AMD just proved they're not your friend - Threadripper Pro 5000 Announcement
Вставка
- Опубліковано 29 чер 2024
- Visit www.squarespace.com/LTT and use offer code LTT for 10% off
Create your build at www.buildredux.com/linus
Threadripper upended the HEDT market when it released, but now it seems the more expensive Threadripper PRO is here to take over. Is AMD making a mistake, or is enthusiast HEDT simply obsolete?
Discuss on the forum: linustechtips.com/topic/14170...
► GET MERCH: lttstore.com
► AFFILIATES, SPONSORS & REFERRALS: lmg.gg/sponsors
► PODCAST GEAR: lmg.gg/podcastgear
► SUPPORT US ON FLOATPLANE: www.floatplane.com/
FOLLOW US ELSEWHERE
---------------------------------------------------
Twitter: / linustech
Facebook: / linustech
Instagram: / linustech
TikTok: / linustech
Twitch: / linustech
MUSIC CREDIT
---------------------------------------------------
Intro: Laszlo - Supernova
Video Link: • [Electro] - Laszlo - S...
iTunes Download Link: itunes.apple.com/us/album/sup...
Artist Link: / laszlomusic
Outro: Approaching Nirvana - Sugar High
Video Link: • Sugar High - Approachi...
Listen on Spotify: spoti.fi/UxWkUw
Artist Link: / approachingnirvana
Intro animation by MBarek Abdelwassaa / mbarek_abdel
Monitor And Keyboard by vadimmihalkevich / CC BY 4.0 geni.us/PgGWp
Mechanical RGB Keyboard by BigBrotherECE / CC BY 4.0 geni.us/mj6pHk4
Mouse Gamer free Model By Oscar Creativo / CC BY 4.0 geni.us/Ps3XfE
CHAPTERS
---------------------------------------------------
0:00 Intro
1:08 What AMD announced
1:45 What AMD didn't announce - And why
2:57 Why bother with Threadripper at all then?
3:24 What killed HEDT?
5:12 Why HEDT no longer makes sense
9:05 Conclusion - If you're "PRO", you can still buy it - Наука та технологія
"And I'm Anthony" was great, Linus really is building a great team, I love it
Best moment in the video for sure haha
It was badass, but really, using that as an anecdote to say you don't need the extra cores is terrible tho. There exists lots of usecases that were dismissed without mention
I had to run it back 🤣
He's just Anthony. He's an enthusiast, but he doesn't do HPC or any of the tasks that the users of HPC clusters do. Yeah, you can sort of prototype on 4 cores... but 32 cores gives you a much better idea of how well things are scaling, etc...
@@n0k0m3 Browse the comments, I've seen a ton of them. I've made a ton of them.
My business is pretty much running on a 3970X. I easily utilize all those cores and for some tasks even a 3090 at the same time. 128-256gb of RAM is the sweetspot too. Ryzen just wouldn’t cut it and TR Pro actually has no benefits to justify the higher price for me. It’s the perfect high-end media-creation platform in my opinion.
I have 4 VMs and a bunch of dockers running on Unraid server. 3960X has enough cores but it would be sweet to upgrade to zen 3 for higher IPC.
May I enquire what kinda media you create that is so CPU bound?
Not skeptical, just curious😁
@@MAKESZENZE i guess you could look at their channel!
@@MAKESZENZE We are doing 8K+ timelapse/video production and post-production. A lot of denoising, deflickering, object removal, AI sharpening etc. And a lot of it with huge files (16bIt uncompressed TIFF). But for sure the future goes toward more GPU computing, but so far CPU is still just as important.
Last/current gen threadripper, RAM quantity is set to increase quite a bit with next gen (ddr5). Until the next get Ryzen is announced there is no way to know if it will cover your use case as good as a theoretical next gen TR
Mmm yes the cycle:
- Company does something people wants
- Company takes over top spot
- Company becomes complacent and people aren’t please
- Other company notices complacency
- repeat.
This is why competition in business is good kids!
And it works, unless government lobbying comes in and big dominant corporation can no longer be overthrown.
@@Mbeluba "For your safety"="For the corporation's safety"
@@Mbeluba pshhh whatever, i cant hear you listening to music on my Amazon Basics headphones while eating my AmazonFresh groceries on my Amazon Basics plates with cuttlery while telling my Amazon Alexa to add some Amazon Goodthreads in my size to my shopping basket before I'll watch some Amazon Originals via my Amazon FireTV stick 4k
This chip is for engineers to run fem sims on clusters. Why the F is Linus and the gamers here even talking about this chip?? Beats me...They want a friend apparently LOL
Until AMD DISABLES over clocking on low end cpus, stops multi generational sockets and rehashes old architecture for literally 5 generations…. There’s nothing to worry about. You submissive nerds wouldn’t do anything about it anyways besides making a witty video and comment
AMD is hugely constrained by die area, so they're going after the most profitable business first. This is hardly surprising, even if it's disappointing for some sections of the market. AMD is a corporation out to make money, they like to play at being a friend to the cash strapped tech enthusiast but only when it suits them.
This is not a dig at AMD by the way, just pointing out that corporate interests only coincidentally align with those of end users.
And that makes AMD different than how many companies?
If Intel did this no one would blink an eye, but for some reason AMD can't operate like a company that has growth in every market area they produce products for. A little hypocritical don't you think?
And good advice for you, don't feel like you need to jump right out and buy a product just because it's there. I'm doing a build right now for a family member with a 5600X that I bought for $210 from Micro Center. See, what I know about AMD after buying their products for 2 decades, is they discount them towards the beginning of a new release and sometimes pretty steep, so why jump out and buy a product near launch? The 5800X has been between $300 - $330 at Micro Center for months. In fact probably 50% of the time it's been available, it's been on discount.
"they like to play at being a friend to the cash strapped tech enthusiast but only when it suits them. "
Zen 3 was the best product on the market when AMD released it. What's wrong with AMD charging what they feel it's worth? Intel has done that for decades. When they're the industry leader they don't have a right to make money off their product stacks?????
I applaud them. They're sitting in the best financial position they've been in for years. It's boosted their capacity for R&D, acquiring a company that makes sense for them, ROI for investors, since after all they are a public company and if you ALWAYS refuse profits when you clearly can be making them, your investors start to want to go elsewhere.
Lastly, Intel is no one's friend either. Don't think their prices wouldn't be VERY high right now if they didn't have to overtake AMD who's been kicking Intel's butt for the PC market the last 3 years. I'm thinking if AMD weren't competitive, well, first you wouldn't even have Alder Lake in its form. They would have stretched out releases to make more money on inferior designs. Next, they've be prices about 25% higher. The 9900K released at over $500 (ACTUAL COST for the consumer). The 5800X, really the competitor of that part since they're both 8c/16t released at $450. But here is the difference. I was able to buy at launch the 5800X with a brand new game worth $60 USD. Then, about 6 months later when doing another build I got it for $340. The 9900K didn't drop below $500 USD until AFTER 10th gen had been out for a while.
@@johndoh5182 Cope much?
@@johndoh5182 I think you've misunderstood something. Nobody is saying that AMD is greedier than Intel--we're simply saying that AMD is _as_ greedy. This needed to be said because AMD has been perceived as fairly consumer-friendly. Most people already knew that Intel is a profit machine.
This pretty much described any company. Capatalism means your loyalty is to profit not the customers interest. While there maybe overlap sometimes there aren't.
@@johndoh5182 if Intel did it I be upset. Guess what people aren't fanboys don't assume lol
No big company is your friend. They're always going to hold their own interests above your own. The best thing you can do is do your own independent research.
Which typically results in never buying anything
@@genericscottishchannel1603 yep that's the issue with idealism, it is not realistic (well you can move into the woods and life without electricity xD)
its easy to see nintendo, apple and dell as example.
@@genericscottishchannel1603 Beautiful! You save a lot of money!
The thing is that their own (company) interests is to sell, so they can have profit to continue existing, and they won't sell unless you (market) value the product more than the money they are asking for it, all the companies that do not pay attention to the market (person's needs) go to bankruptcy sooner than later.
If AMD wants to kill off the HEDT market they could at least give more PCIE lanes to their higher end consumer CPU’s and motherboards.
And why on earth would they want to kill that market exactly.
@@fVNzO I’m talking about for consumers they killed the market off.
@@ZackSNetwork well you can still buy these things you just need to do it OEM. And I would wager that if your life/business relied on threadripper you would accept the potential surcharge for the added performance. If you don't get TR3000.
Maybe AMD is not trying to kill off the HEDT market and just planning their next release that will also include that market? Could Linus just be wrong?
Agreed. I bought a 5800X for my home server and am using every PCI lane at the moment. Still have a full size PCI slot I'd like to use.
Linus - “I hate that AMD has stopped making Thread-ripper”
Also Linus - nobody bought these anyway
i do…
I mean tbf those are mutually exclusive statements. You can like something but also recognize its not popular.
There's plenty of small TV shows that I loved but got cancelled because of poor viewership, I hate that but I also understand it and don't blame them for it.
I did, and now that I'm to VMs I need cores and ddr capacity even more.
@@CodeNeonMC TRON: Uprising
I will still curse Disney for laying off that show
@@CodeNeonMC It does diminish the claim of "They are not your friend" though when they've made a perfectly reasonable business decision driven by low popularity.
As a grad student doing phylogenetics / evolutionary biology work:
I frequently run analyses that take weeks to complete and can scale almost linearly to hundreds of cores (e.g. marginal likelihood estimation using stepping stone sampling). Most of these are run on my university's cluster, but a couple of years back I built a personal workstation with Threadripper and it's been a HUGE help, as I can run smaller analyses locally without hogging my lab's cluster allocation. Threadripper was perfect because my priorities are (1) highest core count I can afford, (2) decent memory bandwidth, (3) decent single-core performance (important for some applications). Epyc is OK, but I'd be paying a lot for features I won't use. Higher memory capacity with Pro is very nice, but if the new versions remain OEM-only, that would be a deal-breaker.
Exactly this.
Same here.
Wouldn't it be better to save the money on buying a threadripper and just buy a higher end GPU I'm that case? You can offload those calculations to GPU cores which are far more than CPU cores either way
Threadripper Pro is already available for retail.
@@KishoreG2396 They probably have a high end GPU aswell, calculations take A LOT of resources.
threadripper CPU are so usefull for animator like me, i'm extremely sad this market end pretty much BUT, i have nice hope, if intel start release some good HEDT CPU, AMD will have no choice to fight back them with the threadripper line.
nahhh they let amd have this market share for about 2 years, why would they suddenly think thats a good idea now?
Exactly. Doing Marvelous Designer and Zbrush can easily utilize the cores and ram usage of a TR. Even Maya animations run smoother on my wife's 5950x vs my 5900x. I'm sure VFX artists wont be happy about this either. Feels like a huge let down for the Freelance/Indie VFX and Games artists all in all.
So good news, Zen 2 TR is still a great product.
@@johndoh5182 why are you commenting some form of this on every chain?
Yes. It exists. But you do realize that the guy who had a 3990x wasn't looking at the 5990x to buy. He was likely looking at the 6990x or 7990x when the performance would no longer be comparable. But when a company throws away a product chain, it generally doesn't come back.
This is an issue for people in the long term. When the 3990x no longer cuts it, what do they upgrade to? Well, now it's going to have to be an entirly new computer on a chip that's significantly more expensive (instead of just upgrading the Mobo and CPU).
AMD has removed the ability for a freelancer to get industry performance at home without paying industry prices.
Why you're defending this? I have no clue. You seem REALLY keen on making sure AMD has the most positive reception out there. As linus said, AMD is not your friend. Stop treating a corporation like it's your friend.
Compiling huge c++ projects as well nothing like waiting ten minutes every time I made a change lol
I'm a rare consumer, I know. But, I utilize a Threadripper for quantum simulations as part of my research. I was really counting on newer Threadrippers to come out so that I can upgrade to a better CPU for cheap (3960X -> 3990X)
Me too…..
Can you make quantum simulations without a quantum computer?
@@_DT_ my man there are like 2 of those rn
HEDT and cheap don't really go together. HEDT solutions from three years ago cost a lot back then and are generally outperformed by more affordable modern top-of-the-range consumer CPUs and motherboard solutions like 12th gen Intel plus DDR5 plus PCIe gen 4. The Threadripper as such was a performance compromise, fitting between the Ryzen Zen 2s and the Epyc server chips. The Threadripper Pro is its obvious successor with features that the original Threadripper didn't have like ECC support, something that is finally FINALLY coming to the consumer desktop with DDR5.
You might consider investing in older Threadripper systems turning up on the second-hand market in the near future and build a cluster of machines for distributed simulation, or see if Intel 12th gen or new Zen 3 Ryzens will do the heavy lifting for you for fewer dollars than your original Threadripper system cost you.
@@robertsneddon731 ASUS and Asrock have generally supported ECC for AMD CPUs.
Currently running an AMD Phenom(tm) II X6 1035T with ECC RAM.
I even have a old Pentium board that supports ECC RAM (chipset feature only, not CPU supported: kind of like DDR5).
The scene @ 5:56 about the double burst length with DDR5 really took me back for a moment to the DDR1 days and the difference of Ganged-vs-Unganged memory. It was a difference between a single 128bit lane versus a pair of 64bit. Meanwhile now with ddr5 it is split again and a similar idea once more.
This video broke my heart. I'm an AI PhD candidate who built my Threadripper 3x system to fuel my research. I have every one of these cores going 24/7 to generate the data I need for my dissertation, and I've been waiting over a year for the next gen to swap in for an upgrade. I'm also a capable enthusiast, and I don't want some company to make my own machine for me. Thank you, Linus, for keeping it real and keeping us informed. This is truly a sad day.
:(
Honestly I have countless issues with Threadripper. My 3990X specifically has gone through two motherboards because of manufacturers rushing them, then killing them off when issues arise.
That's ignoring that the CPU can't be used properly on Windows due to it splitting it as two separate ones.
I'll miss it and I'll likely not upgrade for years, but I'm happy to trade the crazy core counts in the future for a more adopted and therefore troubleshooted platform. 😭
This is a suprise, just finished watching one of your GT7 videos
ELLOOOOOOO
Theo are you running Windows 11? If so, you might want to disable fTPM and go for a dTPM. You can find a compatible dedicated TPM on eBay.
@@RamjeProductions still on 10 on my main machine, no stuttering but my motherboard has been discontinued with SATA SSDs being broken 🥴
@@BlackPanthaa I feel sorry for you Theo. I've settled for a X570 platform (MSI Prestige X570 Creation) I'm still rocking the 3950X and planning on upgrading to a 5950X. I had some major issues at the beginning with setting up bootable NVMe RAID 0 and even trying to POST with an ODD connected to port SATA-III_5 in the system. But MSI has earned back my trust because they've steadily came up with updates every time AMD dropped a new AGESA update. It has been steady for the past 6 months. Let's hope that AM5 is gonna serve us all well and won't have reliability issues from the get go.
As a Data Scientist working with Machine Learning, I have to say i'm extremely dissapointed that there isn't any update for my Threadripper 3960X.
I actually DO use all the cores, both in R and in Python, and I would love to get a 5960X or 6960X with higher IPC and more cores at the same price as when I got my 3960X.
I don't really need the 8 Channels of PRO/EPYC or even DDR5. I just want a bloody update for the TRX40
100% agreed. I'm a 3D artist and animator. I use Redshift for GPU rendering and Arnold for CPU rendering of print jobs at truly massive resolutions that only a CPU renderer can handle. When I'm exporting animations from After Effects a four minute film can take six hours to render at maximum quality. All I wanted was as many cores as possible and a CPU that could address to 256GB of RAM that doesn't also have a dead-slow clock speed like EPYC CPUs do.
I also want to be able to expand my computer with more GPUs than any OEM would ever configure for. Locking Threadripper into that market alone is a huge kick in the gut.
me with my 8 cores from 2013 fx 8350
Yup, I do small-time video editing, 3D rendering, compiling, AI upscaling, and have been looking to get into neural networking for fun in my spare time and notice huge differences in having 64 cores in the time it takes to get things done, and I can even use the GPU to game on the same platform while the task is ongoing. Meanwhile I have a gaming laptop that I use if it's particularly intensive. So yea, I'm using those cores, I love it as a dual gaming machine + semi-workstation. What surprises me is there aren't more niche developers like me doing the same and love those extra cores, but have ZERO desire to build a workstation because it's not their main work anyway. It's a sad niche-only world we live in.
Unlike what Linus said Threadripper Pro is available to end users
It's why Asus sells a a standalone WRX80 motherboard
Mac Pro studio?
Short-lived sockets is something I have experience with. I bought a system that used an LGA 1156 (Intel for those that don't know) motherboard and CPU. Little did I know at the time, it was pretty much DOA with regards to it's future (functionally it was fine, CPU and board still works). It essentially existed for only a very limited launch of CPUs (only a limited number of CPUs made within a span of like 1 year), and with a quite narrow range of performance. Despite the fact that I bought a relatively inexpensive CPU for it (around 150 USD I think. I'd call that mid tier, but for a lot of people -especially these days- might call it low tier) there was not even a CPU in existence that gave a significant upgrade in performance.
Really upset with that experience. The CPU has at least held up relatively well over time it seems. At least considering it's 12 years old (it ran Cyberpunk 2077 playable).
*Threadrippers are high end CPUs, generally just for professionals anyway, even without being officially exclusively in a pro class. It's not as bad nor as surprising to drop that socket "support" quickly.*
i have an 1156 as well, its 9 years old at this point. runs fairly well, but having that socket means that when i finally upgrade, i pretty much have to buy a whole new rig.
Let me get this right, the more you pay for something, the less you're supposed to care that it gets longevity? As a 3970x owner who got baited in to buy the whole 'we designed trx40 for long term platform growth, this time instead of x399', fok off joke
my i5-3330 is starting to show its age despite being somewhat ok, but its bottle necking my rx570. AMD looks like they support their platforms for longer than intel so i plan on going amd later on but man i basically have to buy a new rig worth about 200 or 300 usd worth of components for a small boost in performance...
@@simrantoor85Or at least a new mobo and chip.
@@bluphenyx most likely new ram as well, I have ddr3 right now. Also, because I have the system all opened up I might as well make other upgrades. New case, new fans, definitely new power supply.
When you're coding for a multithreaded app, there's generally no limit to the number of cores you program for. It's more, how many threads do you actually need? A webserver might need lots. But a game? Well you need a render thread, and you might need one for physics, and you might need one for game logic, maybe one for AI. But there's actually a limit to how many things it makes sense to multithread. That limit isn't a technical one, it's just the nature of the problem being solved. That said, there are games that would benefit greatly from additional cores (hearts of Iron, stellaris, total war, yes I'm looking at you).
Well.... there are very much different problems that can be scaled to pretty much any number of cores. You mention games, and you're right about the lack of need for many threads... that is unless you look a level or two deeper. Multithreading doesn't *always* have to occur at the topmost level, many subtasks can be parallelized as well. Take game physics, for instance, where you could use thread pools to parallelize and massively speed up collision detection by putting each pairwise intersection check into it's own task... or where the contact and constraint resolution could potentially use a parallelized matrix algorithm to solve for forces. Both of these things can benefit from a rather large number of cores.
@@haniyasu8236 - yes I don't disagree with you generally, the latter part of my statement acknowledges that this is specific to the use-case. Though we would want to be cautious about over parallelizing some subtasks. It's possible you could get to a place where the performance gains are lost through thread/context overhead, and you've just made your code more complex for little or no gain.
There is no doubt a happy medium here. Also worth noting that commercial code may not always be as performant as possible; it simply needs to be good enough to pass muster with consumers, not perfectly optimized.
But I do agree with you that there are definitely areas where games could benefit from better parallelization, it's just not every game (not every game has a physics engine, for example).
@@tacticoolrick5562 oh, for sure. I disagree with nothing here. Unless the actual problem in the code or app has grown to enough complexity to warrant it, more parallelization can definitely hurt more than help.
@@haniyasu8236 - well met!
"Well you need a render thread, and you might need one for physics, and you might need one for game logic, maybe one for AI" This is the old way of thinking about parallelising games. As Haniyasu said, these days everything is broken down into many subtasks (or jobs) with a dependency tree and then these subtasks can be scaled to a larger amount of cores e.g. Unity's built-in job system or ECS. However (as per your last sentence) I definitely agree it depends on the game itself and there definitely is diminishing returns as you scale to 10s of cores.
"They're legally obligated to do whats best for their shareholders."
Not actually true but made true by force of frivolous lawsuits. Shareholder Primacy isn't a legal requirement
Yes. Came here to focus on this. They legally obligated to put the company’s best interest above their own, but this does not forcefully obligate them to preform that duty by maximizing shareholder value. Additionally, even the case for maximizing shareholder value doesn’t mean a company will (nor normally should) put profits today so far ahead of profits tomorrow.
Sounds like it de facto is.
@@masshockey49 thanks for saying this. Linus was looking mighty ignorant with that one. You'd think he'd understand this as a business owner.
Might not be legally required but it’s best to keep the board happy if you want to stick around.
The thing that gives most people the idea that maximizing profit is a legal requirement, is the fact that courts allow lawsuits claiming damage to shareholder equity to go forward.
It is this fear of costly legal action that leads to a business environment of maximizing short-term gains.
That and Linus is not the best when it comes to legal language....see his recent video about ad blocking being piracy.......
Threadripper isn't replaced by Ryzen for *most* of the benefits the platform provides. As someone who's now upgrading from TRX40 to WRX80 for MORE PCIe lanes - and the cost of WRX80 is prohibitive and insane - I SCREAM every time it's said that the 16-core Ryzen magically makes this goes away. Genuinely insane.
I hear a lot of mentions about ebay epyc systems in the discords, ever looked into those? Might prove a viable alternative.
Here's the good news. You don't have to BUY anything. You can do the same thing that people all over the world do, which is use a desktop and figure out how to make it work for whatever it is you do, or, except the cost of your work, expense it, and move along. Home users don't need TR. People who do work need TR depending on what they do, such as graphics studios who contract with the major media companies, different fields of science, etc.... People who make videos for UA-cam don't even need an HEDT. They need a good GPU. I know reviews LOVE to use a lot of video editing and render software to benchmark CPUs, but it's kind of silly since people ACTUALLY use the best GPU they can afford to render.
In other words, it's still your problem, not AMD's.
Maybe instead of complaining and screaming like you said, wait for the next gen of AMD and Intel hardware to see what they offer. You said you're upgrading but then you say it's cost prohibitive. That's a contradiction. If it's cost prohibitive, by definition that means you're NOT doing it, by all definitions of the word "prohibit". So, what you REALLY mean is "I'm complaining about the cost of a system I'm buying"
@@johndoh5182 you good?
@@johndoh5182 sounds like excessive copium
Then get the pro stuff and stop crying about it. Consumers DON'T need Threadripper, and even prosumers are overwhelmed. You're in an enterprise field, you kinda just have to swallow the cost and buy the stuff made for your needs. You get more features you'll use anyway...
8:21 the antec 900 hits me in the feels every time. i had one as my first case in like 2007.
I had the same feeling. Had one around the same time.
I would love to see a video where you test how different modern games utilizes multicore setups. I know BFV uses 100% of my 8700k, and is one of the reasons I consider upgrading to a system with more cores. But I have no idea how other games use more cores 😅
My sister is an architecture student, and the additional cores on the 3960X made all the difference. With render times in the hours, the time savings are worth it. I definitely picked the wrong case, though... The SilentBase 801 is too closed off to cool that beast.
@@SB-pf5rc what software?
@@rhamdhani2600 A lot of CAD software doesn’t do GPU compute for rendering I don’t think REVIT uses GPU and I know Solidworks won’t.
@@rhamdhani2600 sadly, it doesn't work with the specific kind of software her university uses.
@@krikun98 that's why I ask what software, but the answer is nowere near lol
@@rhamdhani2600 If you reread your questions, you'll find that one was addressed to someone else. I recommended GPU rendering too, but she's using Corona Renderer with 3D Max. That particular bit of software is proudly CPU-first. While the latest versions support GPU rendering, it requires specific assets/techniques, and she's wasn't tech savvy enough to make the switch.
The intro also applies to LMG. No corporation is ever your *friend*.
I appreciate that Linus has been very clear about that on the WAN show
corporations are a friend of money not people
I would argue that if you own the company, they are your friend
You do understand though that LMG is *not* a corporation, exactly becau7se Linus doesnt want to get stuck in that rock andhard place of needing to increase profits no matter what right? sure big company not your friend but that integrity of not going all out for profits sure speaks volumes
@@battyflaps5410 corporations are horrible for money, remember 2008?
I’m one of those guys that used to buy HEDT parts for home LAB stuff, as it almost always has more memory slots and PCI-E lanes.
I have really been looking into running multiple gaming servers recently, on top of running multiple games,OBS, editing software and discord. I was looking at using the 32 core Threadripper 3979X to do all of that at once.
That’s very dumb.
I understand AMD on that, but man. Threadripper was perfect without having to go EPYC or Pro series which had no benefit for many of the software we use. The 3990X was excellent as our renderfarm and the 3970X was excellent in ratio of single core and high core performance.
Using it for unreal realtime prodution, Photogrammetry, AI training with multi GPU setup as Ryzen doesn't have the lanes to supports multi GPU setup.
I bought those chips for time saving, intel xeons are the worse as of now for our workload as they are not versatile enough. For 1 HP Z8 G4 workstation (40k USD) we can buy at least 4 threadripper system with way better specs and performance for the buck. Yes, it's isn't certified or warranty like HP (still crap anyway) as that machine only gave us headache and pain.
AMD was the alternative, it had given the choice to buy better hardware for excellent value. Killing that line is obviously a bad choice for consumers.
And depending on the workload, you can even distribute the work across the 4 threadrippers, making it even quicker.
CO
It seems from the comments here, and in my own work in bioinformatics/genomics, that scientific computing and the trends within it, are a huge grey area for intel and AMD. We use all the cores, have more money to spend vs a regular consumer (but not as much as enterprise), and threadripper was a good option for the sweet spot of building a data processing rig for many applications. We use a threadripper for running real-time analysis of genomes while they are sequencing (along with a RTX 3090) for diagnosing neurodegenerative diseases and benchmarking new algorithms/software we write.
My own personal gaming rig has a 3950X in it because I also use it for my research. 16 cores/32 threads goes a LONG WAY when doing computational analysis, and means hours instead of days when processing a few hundred gigabytes/even terabytes of genomics data.
I think u cant complain when u need it for work.. yes its nice when u can buy a rig for 3k.. but if it were to be 30k, ud just have to deal with it.. its either that or u cant do the work.. same with software.. some software is expensice af...
I definitely remember hearing about that, and I definitely remember my own genomics data being roughly 15 gigs raw and unprocessed. Too bad I never had a chance to actually use my rig for processing so I really wouldn’t know the pain of slow computations.
Ayo! Found another bioinformatics in the wild. I built a new workstation for another friend with a 24 core threadripper With GT710 though, because they don't really use nanopore or other gpu intensive stuff
Maybe apples m chip is the answer
To be fair you sound like the target market for TR Pro. Thats proper real workstation use if I've ever seen one. "Enterprise" is often more about servers then powerful workstations.
Actually surprised how good the context you gave to the clickbait was
I think they are just trying to minimize issues. The Threadripper lineup was awesome for consumers, enthusiasts, and prosumers, the market wasn't large enough to flesh out all the issues and still have it segmented with Threadripper pro chips that had full memory lanes
To me anyone that buys a Threadripper is a "pro" lol. There is no normal consumer market for that chip, period. Computer nerds have no idea where they are in the market. I don't know a single person in my life that runs one.
@@riothero313 My brother does... because it's the system I got him. He doesn't utilize it anywhere near its full potential, but it should shrug off anything he throws at it for a long time to come.
@@zvxcvxcz as long as whatever you throw at it can use all the cores. It's considerably slower than a 5950X in almost any reasonable workload.
Even if you use it for something like rendering, if you already need that much performance, price is probably not much of an issue. It was more of a social experiment to see if there was any point of releasing this kind of a product. And it ended up being more of a bragging right for AMD. I know more people that own or rent an Epyc than those with Threadrippers. Ryzen 9s already cover workstations, while epycs cover data centers and servers.
The whole reason I invested in the AMD Threadripper is because of its reputation of generational support and upgrades. As someone who is part of the Lighting and Rendering community, not a lot of us want to build a render farm or 2nd workstation. The skillset for doing that is a full time job and taking on the task of getting a separate machine configured is a huge burden. There are licensing limitations, hassles, and costs that greatly outway the ability to just press render on our local machine. I can still keep rendering while I work by reserving a very few cores just for the UI in Maya and to keep playing UA-cam. The last thing I want is some closed-off Lenovo box that costs a fortune, has hurdles of its own, and no creative sense to its design that appeals to me.
Yes, thank you for posting this. I am mostly GPU rendering by now, but the moment my VRAM is full I am SoL. Rendertimes just go to an absolute crawl. I don't have a Threadripper because I can't afford it, but I have been thinking about it for the future. Now? Well, fuck me I guess.
At work we have a single Threadripper Rendernode that is also watercooled and overclocked and oh man, that thing just chews through renders like there's no tomorrow. Everybody uses and loves it.
Hey cryptominer you're ruining gaming experience how do you expect people to run OverWatch with very good shaders
that is one of the most legit uses for a high core count cpu and the performance benefits are almost linear. gpu rendering is still not a proper alternative but it's closing the gap. unfortunatelly, amd seems to have made up their mind, so consumer threadripper seems to be at an end. fuck...
I could not have said it better! That part of the video was completely devoid of practical sense.
Plus a CPU that will be permanently locked to that awful lenovo box
Lost all respect for them
LTT always fails to mention music production. When you have a song only mixed and mastered inside of a DAW Every core matters. The ability to play back the song and record in real time matters.
That is true. But I personally find that even for a lot of music production uses, you could frankly get away with just a 3700X for most uses, even if you have a ludicrous samplerate of 192khz set in your DAW & use many 3rd party plugins. So perhaps they don't bring it up because the consumer chips are more than enough for that use.
@@blanknam3d Both of you are right. In some of my bigger mixes, I've used a significant chunk of the 32 threads in my 3550X, but not all of them. The only software I actually use semi-regularly that uses and benefits from all of the threads is Izotope RX7. Having these extra cores has saved an immense amount of time that would've been spent sitting around. I will say though, I mostly bought the 3950x just because I wanted one.
@@Splitface2811 Thing is, when you absolutely need every bit of time at the tail end of an important project, you have that power on tap. There's absolutely nothing positive about not having that flexibility except saving a few Ks. The alternative now is spending tens of Ks more for a Threadripper Pro for basically worthless enterprise support. The people who win are mostly the C suite of workstation vendors and their shareholders.
@@Postman00 my personal opinion is that its a supply chain issue. The entire tech industry has been struggling to aquire resources for their chips. For example 3090ti, 3080ti and 3070ti either don't exist or are no longer supplied so that more main stream chips can be produced and supplied. AMD is mostlileky struggling for resource allocation and management because of this supply chain issue, so I chosing to spend its resources on chips that will be mainly bought rather than less popular enthusiast chips. AMD values it reputation and a friend to the consumer alot as shown with FSR and ryzen chips but in a case where they are struggling to even make enough of the main supply chips it makes more sense to neglect the less popular chips. So I don't think they are trying to neglect both their reputation and their relationship with customers for only profit, I think it's more for self preservation.
Tldr, amd is not selling enthusiast chips as they can't make enough of the normal chips and value self preservation of the business.
From what i heard, AMD is skipping ahead to the new socket for consumer & hedt, which makes the threadripper xm40 abandonment a lot of sense.
3950x user here and can confirm, I was in the market for a ThreadRipper until the reviews of this chip came out and its IPC was so high. Now I use my computer for gaming, video editing, VMs and compiling code and almost never tax all my cores all at once and see no need to upgrade again any time soon.
"I'm not even using more PCI Express lanes than a modern Ryzen desktop can handle... and I'm ANTHONY!"
That is a man who is well aware of his own reputation.
Well earned reputation, if I may add
i mean as someone who does a lot on a mere i3 4170 im sure that most modern systems don't need lots of core. im old i3 still keeps up well with modern games. and yeah i have a ryzen build coming soon. but seriously having those extra cores is nice but noone is gonna need them until like 10-20 years in the future.
@@likeablecloud2454 I have a i7 6700k w/64gb ram and running blue stacks and firefox with 6 tabs which has hiccups. My Threadripper 1950x w/32gb ram is smooth sailing. It all depends on the application specifically. More cores is absolutely beneficial if you need it for software that takes advantage of numerous threads. Or if you multitask heavily more cores again are going to be way better than what your i3 can offer.
He is kinda clueless to be honest. All you have to do to use more lanes than a Ryzen can handle is require 2 GPU's and/or 2 M.2 drives; as Ryzen can only run 2 GPU's in 8x 8x, and you only have 1 M.2 to the CPU, any other M.2's go though the PCH and share a single x4 lane uplink.
@@DragonBornish Hardly.
Honestly the biggest issue with non-pro thread ripper was the memory limitation. Sure, 256GB is a lot but when you're buying something with 64 cores, you typically intend to use a lot more than that.
Right thats only 4gb per core
@@lexecomplexe4083 2gb per thread
@@spedkid4348 yes
And the lack of official ECC support. Threadripper has always been in this weird spot where it's a consumer/enthusiast CPU pretending to be a workstation one, while lacking some key features. Sure, many people can do without those features and still use it for serious work, but it still feels like it was a halo product designed to boost AMD's brand image rather than something that was seriously intended to cater to professionals. It's like they'd rather have you buy an Epyc system instead, but oh wait, then you're limited to much slower single core performance.
Once TR pro came out, it finally addressed all of these issues, bringing together the best of all worlds... but it's stupidly expensive and limited to only a few specific OEM machines. Honestly I don't get what kind of a game AMD are playing here, TR pro could easily become the widely successful platform that non-pro TR was supposed to be, but they're sure taking their sweet time actually doing anything with it.
Sure, you want more than that, but that's also getting near the top of your budget anyway as an individual...
I'm loving how so many of UA-cam tech channels are pointing out that the average user probably won't use the extremely powerful side of the tech made in today's market
I just felt like TR was going more on a professional end where the normal line up provided enough for prosumers.
Higher core count cpus have been an absolute game changer for the barrier to entry in the 3d industry. Being able to render on your local machine in a reasonable time is a huge money saver for freelancers. However the price of the completely top end threadrippers was just too much to justify when benchmarks didnt give proportional results. Essentially the lower down in the stack skus were "good enough" if you were choosing to render on your local machine
A huge chunk of the CGI industry just produces stills for catalogues, concept art, architecture, construction, urban planning etc and so your focus is to get a single render out as quickly as is reasonable. Youre only waiting for a single frame usually, not a photorealistic animation. Alot of the time when working you can run your renders in 'interactive mode' where its essentially rendering on the fly as you make changes. If we didnt have high core count cpus this wouldnt be possible AT ALL and it makes the process exponentially easier removing lots of guesswork from how your final image will look providing instant feedback. But once youre talking seconds to see your image rather than hours or minutes, paying an extra few grand to double your core count and shave off increasingly small amounts of time is just pointless.
Just pulled the trigger on a 5950x to upgrade from my 1950x. This news makes me more confident in my purchase. I wanted a 3rd gen threadripper, but just couldn't justify $2200 for the cpu. Only downside is only having 64g of ram. I wanted to keep it simple with 2 sticks.
High end computer components NEVER give proportional performance to their higher price, you should know this by now. A $1200 gpu isn't going to be 2x the performance of a $600 gpu, it never a scales linearly and pretty much everyone knows this
@@randybobandy9828 I do know this. That's why I never claimed double the price is double the speed. Proportional results based on price doesn't just mean double the speed. It's proportional value that's the important factor and it doesn't deliver on either.
That being said whilst a higher core cpu could result in faster rendering, you could end up with it actually being a limiting factor in some applications such as fluid simulation where threadrippers are often outperformed by mid tier Ryzen. So the value proposition of each sku becomes heavily workload dependent rather than purely just a measure of rendering speed.
@@randybobandy9828 Because you need to factor everything else into the build. If that $600 gpu is a bottleneck in a $1500 rig, then spending an extra $600 for say 50% more performance makes a lot more sense.
@@JordanHembrow95 yeah plus if you're decking out your pc with a high core count CPU specifically for rendering, you don't need a high end GPU because you're not using it for rendering.
Thank you for calling out on AMD.
This is exactly the reason why we must have competition.
The ONLY thing it takes for me to like an LTT video is a cheeky segway to a sponsor. ;)
I expect this was their plan from the beginning. Launching new tech is always a risk. They launched the chips at a deep discount price, and they knew enthusiasts would buy it and popularize it. So the consumer launch was to build a foundation for when they moved to enterprise.
They were only able to release this chip for all, Cloud servers and Desktops, the same chips. Just too many cores for playing games!
@@lucasrem1870 True. But there are more demanding applications than gaming, which are not strictly enterprise.
Well I didn’t expect a multit billion Dollar company to be my friend to begin with ….
Some definitely do for some reason
Yeah, it's weird people seem to forget that.
I had to explain this to a few of my old coworkers when they thought we were all apart of the family.... "family means they can take advantage of you"
@@peeonu25
Apart means separate.
A part means a fraction of something larger.
Even space, essentially nothingness, can be significant.
that goes for all companies honestly not matter if their in the thousands, millions or billions . If they make a good product that deserves the money and don't make decisions to go against the customers that helped them becoming something bigger I thin they will be well off ,but unfortunately few are left that go about it that way and can always change if a ceo gets replaced with someone who is profit driven .No matter how big all companies or people make mistakes ,make bad decisions what's important is how they handle the situation after that so stay vigilant and warn people not to gush over companies like they are some divine deity , I see way too many people in the gaming industry like that unfortunately
The additional pcie lanes is the reason why I downgraded from modern ryzen to older threadripper. I actually need them (gpu, spf+, sas for tape, and video capture card) and threadripper has been the cheapest way to get decent performance. And I do have dedicated server running used xeons for server stuff.
I hope this is just an awkward transition to Zen 4, because otherwise cheap Xeon builds are going to continue to be the prosumer choice for PCIe lanes and RAM capacity, with AMD officially bowing out of HEDT. I think that for a lot of Threadripper buyers, it wasn't the core count as much as the extra PCIe lanes and RAM capacity. If AMD could give Zen 4 a quad-channel memory controller but leave it up to motherboard makers to produce 4 and 8 DIMM slot SKUs, then that would be ideal. Having Zen 4 Ryzen be able to have motherboards with 4-6 PCIe slots with quad channel memory (without ECC) would allow Ryzen to satisfy a lot of the prosumer market and take away a lot of sales from lower end Xeon.
@@alexv8940 wont happen because that’ll also cannibalize future TR pro chips
The extra pcie lanes is the exact reason that I have been waiting to replace my system from 2011 that struggles with editing HD footage.... Gpu, spf+, video capture, video output... possibly a raid card. Was going to build it last year when the new Threadripper was going to be released but nope, didn't happen. Now I don't know what direction I can go to get those extra pcie lanes when the Pro and Epyc series is out of my price range.
@Chocolaa-P tbh I don't think that it would. A lot of prosumers use lower end/used Xeons for their workstation and server whiteboxes. Going past 16 cores for most people, even people with home servers, isn't needed; it's all about the PCIe lanes and RAM capacity. And if they need more than 16c/32t, then they just do a dual-socket Xeon setup. IPC won't be as good as a single socket AMD solution, but if a workload benefits from more cores, then it is better to add more cores than have higher IPC (with the exception of NUMA unaware workloads of course). AMD could easily capture a lot of market away from used Xeon sales if they added a few more PCIe lanes and quad-channel memory because then a Ryzen whitebox would be cheaper than a Xeon one.
In the last year, about 260,000 Xeon CPUs (both chip only and in-system, not counting lots of more than 1 or systems with more than 1 socket) have sold just on the US eBay site alone, totaling $65,700,000. And a lot of server component sales don't go through eBay but companies that specialize in server part resales. That's a lot of people trying to save money by not buying new workstation/server hardware at enterprise prices. Giving Zen 4 Ryzen those two extra features would eat into those sales.
I have a 3950x I edit video. I regularly use all 16 cores. I have deadlines and a high volume of content that I produce. If I didn't have this cpu I would be in trouble. I wanted to go up to 24 or 32 cores but I didn't have the budget for a threadripper. Ryzen 9 was the best thing I could get. I have no regrets.
I get that but, you are still a rare use case. I batch edit photo's and even my 8c/16t laptop crushes it not to mention the overclocked 8c/16t desktop cpu. Most people don't need that HP.
@@riothero313 Most people doesn't mean no people, and Linus' statement was not made with such qualifiers.
8:13
OSHI-- that's my old case on the left! My ole Antec 900.
Had a AMD FX 4350 in it.
Consider that with the somewhat low sales of non-pro Threadripper CPUs compared to their other products, the ability for Ryzen to do most of the same things, and limited output due to various world crises... Maybe then consider that Threadripper is just being shelved, not discontinued. When the need arises again, it'll probably return.
Kinda like Batman?
@@RudyBleeker It's the CPU we deserve, but not the one we need right now.
I need it now. I would buy a 24 core 5000 series TRX40 tomorrow if the opportunity presented itself. Upgrade my 3000 series. (Presuming 24 core is still the sweet spot for cores vs clocks)
"Even power users are having trouble imagining how to fully utilize 16 cores" there are plenty of independent 3d artists like myself that have functionally unlimited need for compute power and are only limited by price. I don't know how many cores I can meaningfully make use of before it stops making a difference (ignoring cost), but its gotta be in the high triple digits. And on the memory side, Ryzen's stupid memory limits are an even more pressing problem (to the point that some projects are simply not feasible, regardless of compute time needed). Zen3 Threadripper has been a hugely awaited release for me, but the lack of direct to consumer sales is going to be a huge pain in the ass
Right this statement is shortsighted af.
Levono's exclusively agreement ends in 3 months, so you'll be able to buy them after that
part of this video is based on a rumour (AMD hasn't officially stated that these parts won't be purchasable)
You don't fully utilize 16 cores. Sure, you max them out for a few minutes, a few hours, maybe even a day or two at a time, but then they idle down again and sit there doing nothing while you queue-up the next batch of work. When computer companies say "fully utilize", they mean "run at 100% load continuously for _years."_
@@deusexaethera you underestimate my workload. The queue is always full.
@@deusexaethera By utilize we mean the combined compute power of those cores
7:16 Nice. I'm still running my old AMD build from 2009. Same initial bios screen except the last bios update was in 2012. CPU 955 x4 black edition 16 gb ram Asus M4A Motherboard and dual monitor. The only thing I've upgraded over the years are the hard drives to terrabytes and the video card Radeon Rx 480 8gig. I wait to upgrade until I'm happy with what's being offered. Took from 2009 to 2016 to just upgrade the video card. I was happy with the price per performance at the time. Still haven't been sold on any CPUs that have come out in the past 13 years. I can still play all my old favorites great with no issues (other than actual game bugs) from Skyrim, GTA V to the Witcher 3 and even the older games such as Gothic II. Though, I am getting the itch to upgrade fully soon.
The bottom line for most users HEDT use is the trade off between more cores or higher clocks. Some task cannot be programmed to use several core at the same time. This often make 4 or 8 higher clocked core a better option for certain workloads. I would personally say 16 cores with high clocks and the option for 256GB to 512GB of ram memory would be the sweet spot for most HEDT users.
I would have loved for mainstream to get quad channel memory support, now that Threadripper gets 8 channels
Just cluster a fast CPU with a massive multicore CPU. the OS does the choosing where the task runs.
I was amazed the dual Xeons dual chipsets/motherboards didn't allow this approach. Its not rocket science to deduce this would be a desirable option for many.
It seems marketing people do the design choices at AMD and Intel. not engineers.
@@joefish6091 Creative! Makes a lot of sense and would make cooling them easier.
@@joefish6091 Having different chiplets with different clocks sounds like a nightmare. Edit: Oh, dual chips. I can't read. I still think different clock speeds would incur heavy costs.
I love how they put Anthony in to give his opinion, tbh I almost trust him more than Linus on this specific topic
Well, I guess I was one of those "very few people" who built a 3000 series Threadripper in the fall of 2020. My 3960 24 core TR is paired with 128Gb of 3600MHz CL16 DDR4 memory and a 2080 Super.... Of course I missed the 3000 series Nvidia cards by a month or two.... but even so, my Threadripper build kicks as much ass today as it did in the fall of 2020 when I first built it. It's plenty of PC for my rendering needs. If the chip doesn't die prematurely, that machine should last me a long time.
VERY good! Congratulations. You figured out how not to BITCH about AMD producing a product that many professional studios will use.
No one HAS to upgrade to a PRO part, not even professional studios need to upgrade. People can be content using what they have.
I didn't bother watching the video because I find people who complain about what a company makes to be whiny little bitches. There's nothing in AMD's charter that says "we will make a product for EVERY SINGLE use case and do our best to have everyone LOVE our prices, at the expense of our profits!"
They have in fact produced a high quality HEDT with Zen 2, when they had more wafer capacity. Right now AMD has grown so much in every market, they'd have to cut production on something to produce an HEDT version of TR this time around, and I don't blame them for wanting to produce as much EPYC CPUs as they can since they're selling every single one of them, partly before they ever make it out of the Fab because of contracts they made. On top of that they're getting a lot more business than they ever had in laptops, and they can't make the custom APUs for Xbox and PS5 fast enough. I assume they've also made contracts for part of these Pro TR systems, since they aren't even putting them out for discrete sales yet.
And as far as all the whiny bitches go, they can always wait for Intel to put out an HEDT on Alder Lake or once again wait for both Intel and AMD to release a new gen of products to see what comes out then. Maybe AMD will have a lot more capacity from TSMC than they did for Zen 3. At least they can get a lot more die off of each wafer.
same. i picked up a 3000 shortly after. Pretty salty about this.
@@johndoh5182 Idk if you're an AMD exec or what but I do recommend a shower now and then and also to get some fresh air. Might as well touch some grass while you're at it bud
@@johndoh5182 Shill harder.
@@johndoh5182 I agree with your reply and quite frankly, am impressed you put so much time and effort into your reply so that it has a lot of information and is well presented. My only problem with it is that it's a reply, not a comment. Maybe copy/paste/modify it into a comment (especially since Linus even made the remark himself about "Here he goes again, Bitching about AMD" in the video) so that everyone who reads comments can see it, not just everyone who reads the replies to this comment (so you get the most out of the time you just spent).
Edit: I myself, have made a few long comments ranting about something before, I know the feeling of having a fuckton'o'shit to say, and that "Did I really just blow 2 and a half hours writing a fucking UA-cam comment?" Feeling. Haha. Oh and now 20 more minutes to edit, copy and repost so others see perfection without the "Edited" mark next to it, lol.
Ended up with a 5800x in my desktop. Works perfectly for gaming at any resolution, and I set mine up to boost to 5ghz when it has the headroom. I could use a gpu upgrade since I’m running a 2080 which is basically a rebadged 1080 ti but most people don’t need anything even close to threadrippers core counts. I do far more with my rig than a “typical” user and 8 current gen cores is fantastic. But I will be very happy when devs start making better use of those cores, as I often dont see more than 65% total utilization across the entire chip unless I’m encoding or rendering etc, which a lot of encoding can be offloaded to the gpu these days.
I like how Linus is such a realist backed by good info from good sources. Helpful. thank you
I found Threadripper to be nearly perfect for my UnRaid server. I had a 2950X and 128GB of ram, run AdGuard-Home, Minecraft, Emby, Grafana, Home-Assistant-Core, MotionEye and others along with 25 virtual machines. More important to me than the cores was the availability of PCI-Express lanes, where I needed to fit a graphics card, an HBA for drives, a card for four NVMe cards and a 40Gbps network adapter. Threadripper and the X399 chipset was a good solution.
I'm planning on running a threadripper pro on a msi TRX40 board w/256G of ram. Is that a good fit for a server that runs a business site, (200 or less simultaneous users) a blog, a 7D2D server, and a NAS (private cloud storage)?
@@scotteric8711 yes
Inb4 next week’s AMD praise video
love the Anthony splice, always love watching .. and always learn something.
Love our 3970x and 5950x machines for Vm, truenas, video crunching, real-time video analytics, etc. We run 256GB ram and a handful of PCI-E lanes.
We would have had to spend double for nothing useful for a TR-Pro CPU.
Hate AMD is killing them off, hope they have a change of heart.
TR is perfect for every serious video editor IMO.
Hmm. It does make sense. My own actions for our project over the last few years more or less mirrors what you said. We have a bunch of build boxes the basically use 100% of the CPU and memory resources handed to them to compile up open source packages. Here is our progression in build boxes. No particular time frame given but lets say over the last 7 years or so:
OP Quad-socket opteron (48 cores)
XE Dual-xeon (16-core/32-thread)
TR 2990WX (32 cores / 64 threads)
TR 3990X (64 cores / 128 threads)
RY 3900X (12 cores / 24 threads)
RY 3900X again (a second one)
RY 5900X (12 cores / 24 threads)
* The quad socket opteron was retired because it became too power inefficient (1000W)
* The dual-xeon is still useful, strangely enough (or perhaps not so strangely)
* The 2990WX was retired so some parts could be moved to the 3990X
* The 3990X is our fastest build box, with the most memory, but we only need one
* The 3900X's and the 5900X are our main bulk build boxes these days for constant chugging
We use the 3990X all the time, but we only need it in situations where either we need to quickly do a full bulk build (1 day turn instead of 2 days), or if we need to compile a large package with full debugging, like chromium, which uses every last byte of the 256GB of ram in that box during the build. Or basically any quick turn... testing patches to chromium while debugging a problem, for example, often requires being able to quickly do a full rebuild rather than an incremental rebuild. The chromium package contains 50,000 C++ source files now and the 3990X can build it in around 35 minutes while anything else takes several hours.
But these days the machines that are chugging away all the time on package building are the 3990X's, the 5900X, and even the dual-xeon for low-priority staging work. In otherwords, it made sense for us to get one 3990X, but it didn't make sense to get more than one. The 3990X *is* the most efficient box we have, in terms of power consumption. But the AM4 systems just got too cost effective. I can build three AM4 systems for the same cost as one 3990X system. It made more sense to throw together a couple of 3900X and 5900X boxes each with 64GB of ram (actually the 5900X has 128G of ram in it), then blow the budget on a second 3990X.
And to be perfectly honest... how does AMD top a 64-core/128-thread threadripper? While you might want to overclock these suckers, anyone using them seriously needs 100% stability and won't. None of the new architectures (Intel or AMD) really have much impact in a power-efficient builds verses the 3990X. The threadripper concept is being squeezed from both sides... it is being squeezed from the retail ryzen side, and it is also being squeezed from the EPYC side.
Threadripper really doesn't make a whole lot of sense any more. What AMD needs to do instead is make more cheap EPYCs available... similar to what Intel has always done with its Xeon line (though hopefully without the same upselling-feature hell that Intel does).
--
Now, I do have to take exception to one thing you said... machines don't become 'unusable' just because all the cores are being utilized. But maybe you are talking about s Windows-specific issue. With Linux or BSD, a workstation can trivially be utilizing all cores for some bulk operation in the background without affecting interactive operations (the GUI, your browser windows, etc...). It doesn't matter how powerful the workstation is... it could be a little mini-pc. The performance of interactive operations in the face of bulk stuff running the background is *strictly* a function of the system scheduler. If it isn't smooth in the face of that load, its a bug with the scheduler, not a problem with the CPU.
-Matt
a lot of times it's not just the cores, ppl need more pci-e lanes and consumer does not cut it...
@@giornikitop5373 At least until the next PCIe rev comes out, then even consumer boards will have far more PCIe bandwidth than anyone needs short of hanging 100 NVMe drives off of the machine. AM5 kinda seals the deal IMHO.
Even now there really are plenty of PCIe lanes for nearly all situations. There are really only three components that need lanes: GPUs, NICs, and NVMe devices. But the actual number of use-cases where a 16-lane PCIe connection to a device is needed are extremely small.
For example, there are virtually no GPU setups that actually need 16 lanes. Not even for gaming. One would have to argue that 5% faster FPS is life-or-death for it to matter. I don't know of any inference or training cases that need the full 16 lanes either.
A 10Gbe NIC is only 1 GByte/sec... *ONE* PCIe v4 lane is enough. So its only when you get into the 50Gbe NICs and higher that extra lanes are really needed. And for NVMe, while benchmarking 30 concurrent NVMe devices all transfering several gigabytes/sec makes for impressive results, there aren't actually very many use cases where that sort of bandwidth is needed. 4 lanes is already overkill and with PCIe-5 or higher it makes more sense to drop down to 2 lanes.
So it isn't really a real limitation for any home environment. Well, barely a PCIe device count limitation, but that's the only one I can think of.
There are two other limitations here that are mostly data-center related. One is, as I just said, the number of discrete PCIe devices (each eating a number of lanes) that the CPU and chipset can support. The PCIe bridges on the silicon have fairly significant limitations in that department. The second are limitations on IOMMU domains. These limitations primarily effect cloud servers that partition-up lots of customer VMs.
That said, many systems in data centers are not going to be constricted by these limitations. And I think most AMD chips can already split their 16-lane PCIe blocks into 4 x 4-lane blocks.
-Matt
are you saying that whenever my windows lags its because its system scheduler sucks ??
@@EpicBunty Basically, yes. The Microsoft windows scheduler does a fairly poor job with multi-tasking, and the GUI itself has tons of stall conditions that we just don't see with Linux or BSD. In some respects, Microsoft depends on multi-core systems just to make things appear more smooth... but even then it does a really poor job if you happen to load all the cores down.
On a linux or BSD system I can do a -j 16 build of the system (even without bothering bumping the nice to +20), or anything else that is cpu-bound, and you can't even tell its running. The GUI is no less snappy than it was before the parallel build was started.
The reason for this is that any kernel can interrupt a cpu-bound user process whenever the heck it wants. It doesn't matter what the cpu-bound process is doing, the kernel can interrupt it in less than 1uS (1/1000000 of a second) and switch to another process (such as an interactive process associated with the GUI, or a semi-interactive process such as a chrome tab). The linux and BSD schedulers do a very good job tracking dynamic process priorities and interactivity, while the windows scheduler does not.
All your system really needs is sufficient ram for the run-time operating size of all the processes... and that's pretty easy to accomplish in modern times when systems can be loaded with tons of memory without any effort. Even when there is insufficient memory, a Linux or BSD system's pager does a much better job than Windows's pager, so paging-related thrashing takes longer to develop.
Also, a lot of people still don't realize that pagefiles (swap space for linux or BSD) should be put onto SSDs these days... if you use them at all on a windows system. Unless the system is thrashing all the time, SSD based paging just doesn't impact SSD life very much and it removes truly idle/unused VM pages from active memory, freeing more memory up for running processes.
But this only matters (on a windows machine) if you are running with 16GB of ram or less. If you are running with more, you generally don't need a pagefile. Linux and BSD systems often configure swap space no matter how much memory the system has because its useful for getting rid of idle VM pages and for TMPFS. e.g. even our 3990X build box, with 256G of ram, we actually have 238GB of swap space also configured because we stage gigabytes and gigabytes worth of data flow through TMPFS, and sometimes the stagings can be huge.
-Matt
I would hold the horses here right now. As we all know, AMD also has AM5 around the corner.
If they pursuit the same goals with it as they did for AM4, this means that you are most likely be (theoretically) able to easily run a 32 core CPU on that socket.
So what I think what will happen: Threadripper will die, but the ultra high end CPUs for consumers might just simply become part of the ryzen series and run on AM5. (I have zero proof and this is all but a theory, do not take this too serious. Apart from that, this may make sense since this would reduce product line complexity from AMD)
reducing product line complexity would help if they fixed their naming schemes for mobile and desktop lol
7:12 virtualisation is key, for people that setup KVM's having those cores is huge.
My build was going to be a 3950x with the same ram he showed, went with the 3900x but honestly i dont think the extra 4 cores was worth the 200$
8 cores is all i really need, but the extra power is nice
"What workload needs more than 8-16 cores"
Well for me that is an easy question to answer... VMs! Not just for production, but pricing and GPUs have lead me down the path of considering one rig for multiple users in the house.
And when you add more competitors for those cores and you have to slice it up a little, suddenly it isn't enough!
Then you move on to RAM, you can either option A pay through the nose for high capacity memory modules, or have more slots and channels and use cheaper smaller modules to get the needed capacity.
This is why I despise 4 memory slotted motherboards, and even more so the amputated HEDT motherboards with 4 slots.
And the PCI-E lanes? Oh don't challenge me there, you can easily have a demand for those in the home with HBAs (Host Bus Adapters for storage), 2-3 GPUs for those VMs.
Suddenly a Threadripper will spank a socket AM4-5 any day of the week, even if it isn't the king of gaming performance, but because the other boards just can't provide the needed resources due to a gimped platform.
VMs, AI, CFD, FEA, SPICE, protein folding, there are a lot of workloads in smaller outfits that need more than 16 cores. I use all 24 cores in my Threadripper and would gladly fill every single PCIe lane with GPUs if they didn't cost an arm and a leg each. I do CFD simulations for rapid prototyping of UAV airframes. It's my hobby but it also pays well every once in a while.
"Oh here we go. LInus is ranting about evil corporations again."
Uh... That's what I like about your videos. Keep doing this please.
more threadrippers or ps5s tho?
@@vincentellsworth7905 You misunderstand, my comment was purely a reply to that quote. I am not interested in buying either of those products, and I don't think this decision from AMD has anything to do with whether or not we will see more PS5s in stock.
@@natechucks you're saying they're "evil" because they would rather make a really popular product instead of a niche one...?
@@vincentellsworth7905 I don't have a serious stance on the issue really. You are reading too much into my comments for no reason.
@@natechucks you took the time to write a comment saying a company is evil...
I am doing FEM Electromagnetic simulations, and the software I use runs in 20 threads, using 80GB of RAM. So I am one of those use cases, fortunately built the rig using 3950X right before COVID struck.
I’m glad I found this. This is the context I’ve been looking for. I’m looking to build for the very first time. I’m a developer that wants to run a bunch of VMs, occasionally gaming will also be done. I am very conflicted what amd cpu I want. I want a bunch of cores but I wasn’t sure how many is “enough”.
I'm looking for the same use case, and also a first time builder. I'm fortunate enough to be in a stable financial situation, but the problem is figuring out the best way to spend it.
I've seen some stuff online recently that AMD will be launching a Threadripper 7000 series "Storm Peak" some time in August or September. Looks like it'll be available for consumers, so I'm planning to hold out for that when (if?) it releases.
Loved and still use my 1950x. It currently serves as my "stream rig" running a custom x264 encoding profile that utilizes the majority of the cores while leaving 4 leftover to handle overlays, minor audio routing (most of it's done by hardware) and overlaid mini games without hiccups, lag spikes, or frame drops. Also runs VM's on occasion. The other utilization of it is in PCI-E lanes running a GPU, u.2 and m.2 NVMe SSDs, several SATA drives (HDD, SSD, and BD-RE) and a 10G SFP+ card to handle transcode output and NDI inputs without saturating. Maybe I could get away with stuffing all of that in a standard desktop these days, but I see no need to replace the rig since it does the job and does it well. Stream never drops frames or gets pixelated when the action gets fast paced vs I've seen nvenc and quad cores running x264 choke and pixelate badly when the action picks up. Never hurts to have an overkill rig to handle anything you can throw at it. Are all cores loaded down 100% of the time? No, of course not. If your rig is running at 100% all the time: you're either running a stress test or benchmarking tool, or your CPU isn't powerful enough to handle the task you're throwing at it. IPC isn't the best, but the 4.05GHz all core overclock does help slightly in this department. (custom loop cooling both CPU and GPU with full coverage blocks to keep thermals tamed)
I was really planning on building a threadripper system with several GPUs for hydrodynamical simulations.
Unfortunately I'm still not earning money.
Utilising "all those cores" is what we in the physics department call a "small simulation" if it takes less than a week to complete.
I knew this would happen ever since they hiked up prices on the release of Ryzen 5000. The moment intel released their 12th gen with better performance and a LOWER price just solidified my opinion
And intel is taking away AVX 512 from 13th gen. The hardware industry wants to make more money off the software industry now.
@Dizzy Gear well arm doesn't make chips but I get what you mean . M1 max is a beast
@Dizzy Gear i think it will all collapse in on itself soon.
All tech
Yeah, SLIGHTLY better performance and the HIGHER price over time because the power consume is almost twice as much.
One more item 1:16 I notice those clock speeds. So it looks like it's true what I read in a tech article a few years ago that the industry is hitting a wall with current technology on how fast the CPUs can get without frying to a crisp. Thought it stated around 5 ghz. Instead what the industry is doing is expanding everything out to eek out that much more speed and performance from the system to compensate for the old CPU thermal/speed threshold.
Most real-world app scenarios (not artificial benchmark code) you reach a point of diminishing returns -- the cores fight one another for L3 cache space, or RAM bandwidth, or both. The added L3 cache on the TR (and Pro) vs consumer Ryzen, probably helps more than sheer number of cores. L3-cache-per-core can be an important number to look at.
Fully agree with Anthony on this one. When I built my machine in early 2020, I opted for a 3900X and most of the time I don't need all those cores. I'm in research so some of my programs can fully utilize all the cores but for whichever calculations I can do locally 12 cores are enough and if they're not anymore, then it's about time to move them on the cluster anyway.
12 cores was the sweet spot for me too. 5900X is also better value than the 5800X, which needs to have higher-binned chiplets.
This is exactly me. Managed to build my system 2 weeks before COVID hit the UK. 12 cores is enough to test the code then it's off to the 64 core workstation with 3 2080ti in it or the cluster if it's truly computationally heavy.
I also went with 3900X for my new build last year, would’ve gone for a 5900X but cost (thanks to graphics card pricing eating the budget) was my issue.
Like Anthony, 3900X is overkill for my everyday needs, it is only when I’m rendering aerial photographs (in open drone map) which uses most of the cores and also requires a bit of memory. Depending on the complexity, on average it takes 20 minutes, on a CPU with half the cores, it would take ~45 minutes.
Someone a couple rooms over from me started playing "Fur Elise" on a piano right before Linus said AMD aren't your friend, and I thought it was some poignant melancholoy background music in the video. I was confused when the song continued during the sponsor spot.
Doing music "producing" and love my vsts (virtual instruments).
Especially those who are modeled like pianoteq and modo bass.
I also like to make my gigantual synths without being limited the way I am now.
In fairness I don't know if limitations are on CPU-level since one of my playthings is a Eurorack simulator Voltage Modular where every one piece is pretty much always running as a process (built in Java).
I also like to create challenges like having these hardware-simulated synths as instruments and running several different ones at the same time.
For fun.
It is a bit like memory.
Never thought I would ned 128GB but it is so good when you use a really good orchestra VSt.
However. I guess it is because of the silicon shortage there is no new normal threadriipper.
I can know for sure but it seems like it to me.
Seems in recent years hardware has come a long way whereas the software optimisation is now the bottleneck and needs to catch up, as a result most of us end up buying machines that are overkill for our needs, including me 😆
No no, software optimisation took the opposite turn. No one is interested in the kind of coding i do. Everyone wants single core performance and browsers. Coders dont even know the basics of software dev today
If you ever happen to watch Level1techs, you'll realize the hardware recently has improved so much, software in general is playing catch up. Depending on application, there are certain pieces of software that can't utilize the hardware properly. We all saw it once CPU's went over 4 cores - 8 threads and those of you who are stuck on Windows, the thread scheduler caused performance regressions as an example.
@@System0Error0Message As a low level developer, I feel really bad that nobody optimize their software and take a lot of resources to make a simple hello world, is like they are too lazy and/or they do it on purpose so people must buy more expensive and newer PCs.
@@romevang I don't see as a system dev a multi thread program that was using 4 thread can't "follow" the hardwars double core. It's a completely stupid stalement, the default for thread pool has always been "nb of thread == nb of core". As a dev I really don't understand what the fuck linus say here but yeah linus IS NOT a dev. Not to mention that, OS are not monolit application anymore since 1980, we CAN have more than one program running. Anyway, I feel linus don't know what he talk about on this subject.
hardware hasn't gone that far recently its been marketed a lot more, there is a difference
I have been patiently waiting since the very day Zen 3 Ryzen was announced (2+ some years now) for Zen 3 Threadripper to come out. It's not, but Threadripper Pro is. Sure fine, now I need to wait longer for the cpus to hit the DIY market and pay even more for them. For the added cores? No, I'm gonna try to get the 12 core model. But it's the PCIe lanes, damn it! I want a system with the Zen 3 IPC improvements and the added lanes to boot. I've been mentally building this workstation since 2019 and it's become painful to think about between this and all the damn shortages. It will probably be out of date by the time I get all the parts together, but so be it. Unless an alternative emerges that has that many PCIe lanes...
Would multiplexed lines work for your case? From PCIe 3 to 4 to 5 is a doubling in bandwith, so I could see them "split" into an earlier standard. That would offer more, but slower lanes.
Or an additional controller on the motherboard that adds more lanes.
Sure, latency would take a hit, but perhaps that could be a workable solution.
@@HappyBeezerStudios Yeah what I really want is at least 2 full x16 4.0 lanes minimum. The Threadripper platform exceeds this for sure but I’ve not found any other platform that can provide that.
@@HappyBeezerStudios Except nobody us building PCIe splitters. Yeah, I would gladly take an x1 gen5 to x4 gen3, but I believe there is no hardware to do that, not outside Intel and AMD labs at least.
Really, all that happened was we returned to type. We had notebooks, desktops, workstations and servers. HEDT was a midway station where you could use desktop priced components (mostly memory) on workstation class systems.
Of course, back in the day the desktop, workstation and server line all had the same socket with some minor changes (often a PIII would work in a Xeon board but a Xeon would not work in a PIII board), it was memory support and expansion that differentiated the segments.
To be honest, my 4c/8t Core i7 3770k is still wayyyyyy enough to do all the actual things on my computer. Even the hardest tasks never fully use it. I guess workstations and other stuff might need more, but even tough games don't really use more than 4 cores, most of the time.
I'm still rockin the 3930k, glad I future proofed, can still run any game with a good video card even with bottlenecks.
I still have fx 6300
I have two machines:
First one with the TR 1950X from 2018 for 3D CGI Rendering.
Second one with the R7 5800X from 2020~2022 for Gaming & Streaming also 3D work sometimes.
Both machines are doing fine I'm very thankful for what I have 😌
I know of a physics/astronomy group of professors that have Threadrippers in their cluster to run calculations. There's a market in heavy data crunching. Also, having seen how the processors are installed on the motherboards, I wish they'd implement that type of system on all AMD sockets - certainly better than looking for that little triangle to ensure it is lined up correctly.
Oooh, the Antec 900, my first self-build pc case :D (well the insides where more core 2 duo and ddr2 tho)
Anthony definitely needs his own shirt that says, "And I'm Anthony!"
"If I didn't dabble with running multiple virtual machines ..." That's basically the singular use profile.
"And I'm Anthony!" best line in a recent video.
Thanks for doing this video. The rationale is valid, and I find I'm agreeing with you. Sometimes more isn't really better.
Me: Literally acquires 3975WX two weeks ago.
AMD: Mwahahahahaha
Sadge
You have done this yourself
We just got our two Lenovo P620 with 3975X in them today.
Sadge
I feel like everytime linus is talking about cpu cores he is forgetting that you don't need to utilize every core on single task to take adventage of multicore cpus.
Honestly, I feel like every time they talk about what developers and how they use their machines do they really don't get it...
@@ZachStein if you develop C++ (or just run Gentoo), compiling is very multithreaded!
Man retail edge was a great program. Did a few modules and got my 4770k on the cheap.
AMD's making me feel like Ryzen was an once in a decade miracle and they're doing their best to let Intel catch up and blaze past them once again.
I mean 1st gen Ryzen was a huge improvement over FX but it was still pretty trash compared to Intel (as Anthony said in this very video, "slow as balls"). It wasn't until 3rd and 4th gen that they really caught up to Intel, and that wasn't too long ago. Doesn't feel like they've been sitting.
Intel has already blazed past them. They've already surpassed amd's chiplet design
@LexeComplexe agreed Intel truly has a superior architecture at least from an IPC perspective
@@lexecomplexe4083 lmao no. AMD is fucking Intel in the ass in the much larger and lucrative datacenter market. That's why Intel is seeing -20% year over year growth from the last quarter.
@@brothatwasepic Zen3 is literally one year older than Alderlake and still holding up quite strong. I don't think it's a fair comparison to say Intel is ahead of AMD.
I'm actually really mad about this, as I use my 2990wx and love it, and was planning to upgrade. HEDT is niche, but among software developers it's really nice to compile across many many cores so quick.
Omg same. I was literally waiting for the new threadripper to be announced so I could figure out when I could upgrade my threadripper build. This is so disappointing.
Guess I'll be riding my 2920x for a bit longer. Wasn't worth the cost going to 3rd gen and having to get a new mb. Certainly not going to pro
Totally agree with this honest review. I have a startup in machine learning and 5950X + 3090 are doing good. I will upgrade to Threadripper Pro in the future for sure, but only because the pre/post processing of huge amounts of data is only feasible locally with more threads you can possibly get. Actual training of the model is the 3090 (with SLI if possible) job.
For you 16 cores and 3090 may be enough, and that's great. If your workload runs beautifully on that system, more power to you for not getting caught up in the latest tech trends.
But for people doing video editing or CPU-based rendering, as actually mentioned in the video, threadripper still serves a significant role there. Just for CPU-performance, threadripper 3000 CPUs from two generations ago, before the monstrous IPC uplifts of Ryzen 5000 and 7000, still hold up quite well compared to the ryzen 9 CPUs offered nowadays. Compare that to the Ryzen 9 3900/3950X whose performance, while still impressive in its own right, has been completely overshadowed by newer processors at the same prices. That's part of what you are paying for with threadripper, alongside the expanded memory capacity and bandwidth that still isn't available on something like AM5.
Another aspect to HEDT platforms that people including myself value is the massive PCIe lane count. Some people may not make use of that, and that's fine, they can utilize only the multi-threaded performance. But for non-SLI/CF systems that instead require faster or redundant networking, more m.2 slots than those that ship on their motherboard, a sata expansion card, or multiple GPUs for hardware-passthrough virtual machines, the 28 lanes available on consumer desktops is simply inadequate. I have not found an AM5 motherboard that ships with 2 x8/x8 and 1 x4 PCIe slots for a remotely similar type of setup. And even if such a >$500 motherboard exists, the limited PCIe bandwidth and core counts would pose a challenge for heavier users.
@@jamesbuckwas6575 I'm really impressed and delighted by your competence on the matter. I can say that I've learned something. Thank you for taking the time to answer my comment in such a complete way.
The reason one might pick up a threadripper first gen in 2022 would be for cheap virtualization rig that can still be efficient, it's got ECC (for ZFS) and 64 lanes (pcie3) you can do a lot with those lanes like attach network cards, have a gpu for rendering, HBC for storage and a lot of enterprise grade PCIe cards are PCIe 2 and 3, not likely to upgrade soon and if they do that means a whole lot of used parts on the market like SolarFlare SFN7122F which is good for people running Truenas Scale
I definitely do not feel comfortable with the "16 cores should be enough" statement. I am a developer that still compiles C and C++, but the cores are also used by a browser to load multiple tabs and multiple images and other objects at the same time, by the operating system, by all the background services and by the applications behind the icons near the taskbar clock (system tray).
As we now know, the 1st generation Zen CPUs, including Threadripper, had bad-ish memory latency. At the time people where enthusiastic about the new generation of CPUs from the underdog AMD, that was again fighting neck in neck with Intel, and nobody was talking about memory latency. But now that latency is the only reason Threadripper 1000 feels slow and sluggish today.
I'm a game programmer, I think his comment was "good enough for now". Honestly I expected true multithreading to be mainstream years ago. I imagine AMD also saw that that was the future, and that's why they went that direction, but it's taking companies and programmers so long to make it a priority for high-demand applications to be multithreaded.
We'll get there though, and I've seen a lot of advancement in the past year/two, with more engines and libraries starting to support multithreading a lot better, which has been great to see finally, and is making me feel less dysphoric with the current climate, to how I know it should be.
Agred. I'm really sick and tired of corporations telling us what we need
We know better than them what we need
This move is just insulting
Also goes along the same lines as Linus unironically saying 16gb of memory is enough... moronic
He just got his Intel paycheck
@@lexecomplexe4083 he said for an average user...
I'm really bummed AMD didn't come out with their non pro Threadripper first 😢
Crybaby ! Why would you even need NON pro Threadripper, while Zen4 is right around the corner !
@@gertjanvandermeij4265 I dunno man, maybe he needs a 64-core threadripper now and not a 16 core 7950X in like 5+ months
Zen 4 is way better than Zen 3, but it's not like Zen 3 is obsolete at all, esp a 64-core variant
Eyyyyy, a bonus Anthony appearance!
love the content, but the camera is fuzzy? is that intentional?
That's true, people who needs a 32 or 64 core CPU probably will also need those 8 channels ECC memories, 128 PCI-E lances, U.2 ports and those sever/workstation functions. normal TR just not make much sense against normal Ryzen
AMD: *Released the Threadripper*
Also AMD a few years later: *Prepare for trouble. And make it double.*
Wait I thought it was Apple's turn for this line
been happy with my 3600x so far, 6 cores and 12 threads seem good, heck my work computer is stll a quad i7 and does fine
I got the 5600X with 32GM RAM, and I can run Visual Studio, Adobe Programs, 2 VM instances, and more at the same time. Responds quickly without much stress on the machine. I'd probably have to be doing something really crazy to need much more hardware right now. And yea, I'm sure there's people out there that need that.
I love to see a analysis of the ideal/optimal build in terms of threads & cores cpu for gaming, because as you said some of this cpu are overkill.
For Games? 6 or 8 cores max.
More cores helps when running Zbrush as far as I can tell, especially when your meshes are hitting millions of active points.
It still could be come back when supply improves
Also AMD desktop lineup has more cores than HEDT for a decade, so it is a "relative" loss
With Apple finally having really impressive hardware on their Macs after lagging behind for so long, and Intel finally starting to make a comeback of their own as well, I feel like this is probably the worst time for AMD to rest on their laurels.
But then again what do I know about business
Rest on their Laurels by spending nearly $50b for Xilinx 🤔 Somehow I don't think so.
Lisa Su must be cooking up something to justify that move!
AMD's TSMC capacity is already maxed out, supplying AMD's customers like MS and Sony. Adding even more chiplets to the mix wouldn't help them prioritise this capacity for a limited number of HEDT Consumers!
By your logic, Intel's 15 billion dollar acquisition of Movidius and the subsequent one of Mobileye in 2016 is not sitting on its laurels.
The fastest M1 mac barely beat threadripper 3970x from 2019 soo yeah
@@drbali
Wasn't speaking for Intel Investments 🤦♂️
But since you brought it up 🤔
Logic and Fact's say Movidius only cost Intel $400m! 🤷♂️
"The price paid by Intel has not been disclosed although it is believed the deal was worth close to $400 million (€355 million). This is significantly higher than the €250 million valuation placed on the company last year following a fundraising round."
🤔 Are you seriously comparing a company the size of Intel, investmenting in a relatively speaking, startup company for $400m, Revenue: $31 Million and 180 employees, to AMD buying Xilinx for nearly $50b, a company employing 4.9k and revenue of $3b? 🤦♂️
Just FYI, Intel Meteor Lake is rumoured to integrate VPU's into the package, coming 2023
As for Mobileye:-
Intel’s Mobileye division has filed with the U.S. Securities and Exchange Commission for an initial public offering, and Reuters is reporting the company could be valued as high as $50 billion.
Mobileye has been a particular bright spot for Intel. The business, acquired in 2017 for about $15 billion, has consistently grown faster than its parent.
Mobileye’s confidential SEC filing means it won’t have to show current financials right away. But the last update from Intel CEO Pat Gelsinger sounded promising. He said in December that Mobileye had posted record 2021 gains that showed a 40 percent revenue increase. Mobileye’s full-year revenue in 2020 was $967 million. The Wall Street Journal reported the company generated revenue of $1.4 billion in 2021.
Intel said it intends to hold on to its majority stake in Mobileye to gain a foothold in the emerging autonomous vehicle market. It surely doesn’t hurt Intel’s case that semiconductors are expected to represent 20 percent of a premium vehicle’s total material cost by 2030, according to joint research by Roland Berger, McKinsey and Intel.
@@WiiNV Xilinx will probably be a good acquisition for them, but I'm not sure that it would do anything for them in the consumer/desktop space. I expect there will be plenty of ML accelerator chips for supercomputers and for executing trained models in automotive applications, etc... but not much change from the Xilinx side for the desktop market.
Amazing video keep up the lovely work