Generally, companies that have that, that also work with non-x86/x64 processors usually have formulas that cut down on the cost for things like this. So they may do something like "Multiple core total by .50"
@@PSjustanormalguy Holy cow, been out of the sys & net admin game for a while, had no idea that VMWare moved to core based licences, that is ridiculous. We used to use VMWare & VSphere at my old company, I'll bet they moved away from VMWare like the plague after that.
@@betag24cn What are you talking about? Windows server has been core based for ages now... 8 core minimum per socket, 16 cores per system minimum... And that's a standard approach to a LOT of the higher end software suites that typically run on windows servers... Not that those normally require you to license the whole thing unlike the OS, but rather that if you license as an example MSSQL and you license say 8 cores, while your system has 32 cores... Well MSSQL will only schedule itself on 8 of those cores. So 24 cores will be idle unless you have other services running on those.
Yeah Edge can be quite as silly as AI with how often the word is slung around. I figure it's just getting servers closer to where they're needed, a pushback from the Cloud-everywhere theme of the early part of this century.
Cloud: euphemism for "someone else's computer". Edge Computing: euphemism for "the cloud was a mistake", and putting shit back on your own servers in your own network. (We've done both for decades longer than stupid MBA's have had trendy terms for them.)
Linus got a 192-core EPYC, you have a 192-core Ampere. These videos are reminiscent of a time when car reviews were taking on the McLaren P1, LaFerrari and Porsche 918 in a narrow time frame.
Couldn't find a comment mentioning it (although I admittedly didn't scroll far) but just as a note the front ports look like SFP not QSFP. If they're 25G they'll be SFP28. Love your videos Jeff
"Nobody puts baby in the corner", and nobody would leave a machine like this sitting idle. It's a total workhorse and probably intended to be maxed out its whole life. Love the content! Great video!
If you sent this guy back in time to mid-2000, it would be the world's most powerful supercomputer. It would still rank among the top 500 as late as 2006.
Back when I did development, we talked about vertical MIPS and horizontal MIPS. Big fast machines were more 'efficient' but lots of 'cores' allow running lots of stuff, but think of them as threads. It is great stuff, but if you can take the MIPS and add it all into one it was 'more efficient' and more computer programming to use more 'cores' efficiently. ... Enough from old history (starting in the '80s when I got my hand on them).
Im a junior dev rn. I really like to think of my personal projects and code as "services". It just makes it feel like a mechanical part that you would use in a car or something. I feel like other ppl realized this too, cus microservices are way more popular than some big monolithic server program. like yea the interactions between them has alot of overhead, but the fact you can scale them so easily the more silicon you spam into it, the better. I used to make alot of roblox games with my friends back in HS. Even Roblox is adding more multithreading support to their games, which is really needed.
2:32: “Don’t think of this as a single 192-core server. Think of it more like 48 dedicated 4-core servers in one box.” Waylon Smithers: “Now, I realize caring for Mr. Burns seems like a big job. But actually, it’s 2,800 small jobs.”
I just like how there is a VGA connector (DE-15 FTW!) sticking out of the middle. If you are running SIMD-heavy workloads Epyc is probably still the better value per dollar.
IBM used to have a refrigerator-sized RISC server called the p690 Regatta. When I stood behind one on the data center floor it was like a full-body hairdryer!
This is a bonkers bit of kit, a truly mind-bending level of power. Loved Patrick's little piece, you have some cool friends Jeff. I will never have or need something like this, but it's great to have it explained.
@@N3m3515 Since QSFP non-plus (i.e 4Gbps) doesn't exist, I've seen it quite common to drop the plus when referring to the 40Gbps option for simplicity sake. The whole QSFP naming saga is honestly kinda crap - especially when you look at stuff like QSFP-DD (which is actually QSFP58-DD) which is 8 lanes because it's Quad SFP Double Density.
Always nice to have Patrick to help in any video! The tech UA-cam is always on it. I really dont know what all of us techies would do with out all of ya!
0:19 Thought you were joking around with the sound of the fan at the beginning of the video Brother Jeff. Then I see your hair flow in the video from the fan I realized it was no joke. You tend to not joke around in your videos which is perfectly fine. Your content doesn’t need it.
The funny thing is, I had to supplement the fans in the thumbnail with a leaf blower-because the louvres direct the airflow down. They actually blow with about the same force as my leaf blower from 3' away, but since the air goes down instead of up, I couldn't get the thumbnail picture without the blower!
@@JeffGeerling we appreciate the effort just wish you could have pulled it off. I would have wore a wig or something if I were in your shoes. Granted my hair is currently like that dude from Gamers Nexus!
I love that companies like Supermicro started to adopt OpenBMC and getting rid of AMI’s bs / their own creation. On the other hand, it’s sad that they don’t publish the code / the layers so that I could build my own image (which is basically the philosophy of OpenBMC).
I was positively surprised to see supermicro using openBMC. Integrating RAID monitoring of supermicro servers in zabbix with storCLI was... Painful* to set up, to say the least. *I'm not sure how old the servers are, but I'd say at least 3-4 Years.
American Megatrends International has been falling woefully behind on hardware support for a good 10 years now, it's good to see them being cropped out of the picture by enthusiast alternatives.
Last time, you ended with "until next time, I'm Jeff Geerling." I asked who you would be after that. This time is now the next time referred to last time, and this time, we have a witness testifying that you are now Jeffy G. Mystery solved. The more you know.
Loved your video title. 192 core is indeed a lot in a single processor. I think it's time to compare it with a cluster of raspberry pi of equal core count.
when @@JeffGeerling and Patrick did comparison with the Ampere Altra version of the same telco edge Supermicro server, the result was it equaled 100 RPi4 and was more energy efficient. ua-cam.com/video/UT5UbSJOyog/v-deo.html
Great Video Jeff, thanks a lot! As impressive as these machines are, as ridiculously are they in a home server context. What is bugging me: there are plenty arm machines on entry level and also machines like these from the end of the spectrum but it seems that other than Apple, no other manufacturer seems to be interested in the mid range so far. I hope Windows on arm will fix this as I have the definitely a strong wish for a 8-16 core 32GB Linux box in my home. If it was more straight forward to run Linux on a Ma Mini, that would be awesome.
Yeah, that's the huge gap that I wish wouldn't exist-Qualcomm has the Snapdragon X Elite, but it's still kind of Apple-ish in terms of not being used in contexts where you get upgradeability or hackability, like you would with a typical PC.
STILL not as many cores as I want! I'm a hobbyist programming who's into HPC. It was so exciting to get a Dell Precsion Workstation t5500 with two 6-core Xeons, for 24 hardware threads. And it choked trying to write n-body simulations with over 1000 bodies. So I had to start learning GPU programming. SO difficult. I just want 10000 arm cores and to program it with OpenMP. Sigh. 192 is a nice start, but I need about 500 of them. Still, a server like that with a few GPUs would be one helluva compute node. So many threads for MPI, other threads running CUDA code...This would be fun to program on!
@@SchoolforHackers I'm just getting started but I wish to code complex, biological-like systems. With millions of agents, chemistry, physics like diffusion. Or imaginary physics. Like, design my own particles with their own forces - like but not the same as electromagnetism, the strong and weak nuclear forces, gravity and watch them interact. I got started just writing simulated flocking behavior, and 2D imaginary physics. But those xeons bog down with over 1000 bodies. I just started learning CUDA after discovering OpenACC isn't adequate. And OpenMP slows down when I offload to nvidia! What I wouldn't do to just write all that stuff to run on normal CPU threads!
@ Dayam - you are WAY out there in the fun zone! What you’re doing reminds me of complexity theory, and you’re playing a hugely upscaled Game of Life. The, um, security tester in me imagines a few milliseconds hijack of most of the Amazon cloud here, a few milliseconds there…but of course that would be unethical. Ethics being, perhaps, the fifth force.
@@richardzeitz54 So just possibly something like this. The girls, who are named for gems, “clink” when they run into each other. And the dogs … well. ua-cam.com/video/gFlw55xZoSE/v-deo.htmlsi=pCfann45TXKDvP0k
I'm almost 80. The first computer I used in 1969: Philips P1100 (1C1T; ~6 cubic meters), 64KB (1MHz), ~$750,000? 55 Years from 192 cores. In Apr 2019 a Desktop with Ryzen 3 2200G (4C4T); 16GB; CPU+RAM for $175 and on 11 Dec 2024 Ryzen 5 5600GT (6C12T); 32GB; $179. I combined desktop and server in one PC. Host OS = Minimal install of Ubuntu 24.04 LTS; Hypervisor = VirtualBox 7.1.4; File System = OpenZFS 2.2.2. The Apps run in VMs; Xubuntu 24.04 LTS (Email and Social Media); Ubuntu 16.04 ESM (Finances); Ubuntu Budgie 24.04 LTS (Multimedia); Ubuntu 24.04 LTS (Experiments); Windows XP Home 768MB (Jukebox); Windows 11 Pro (if needed) and ~60 other VMs. Note that the CPU + RAM upgrade is the same price as the original ($175 vs. $179), while the CPU is 3x faster and memory is 2x the size.
@@Zimbob2424 In the sixties and seventies the IBM 360 and 370 had computers, that could be bought with 1 up to 4 CPUs. I think Multithreading comes much later in the 21st century.
From experience of my colleague, I do know one thing, because this is a more monotone sound, it is possible to sleep in a server room, even though it's loud.
I finally subscribed. I see you're on your way to a million. I'm interested in a Pi cluster. Or maybe some other. It has to be cheap though and draw low power. And it needs to run Linux. One of the things about all those cores is that they are still kinda slow. So many of my programs are single threaded. They only burn one core. I do some threaded software that can burn all the available cores. But that is in the minority. Even so, I would love a load of cores. Parallel processing is the only way I can speed up certain software.
Yeah definitely different strokes for different folks. I really wish Apple made the Xserve again, with like 64+ of their performance cores. Would be beastly.
@@JeffGeerling I don't know why they don't go with just performance cores in their plug in computers like the Mac mini. There's no battery to save unless you are smart enough to plug it into a UPS.
@@JeffGeerling I definitely feel like a "blow out" is the better part of Crohn's - the alternative is massively painful - as you and I know. Yes, you are correct - servers blow out a lot of hot air... The wife tells me I am full of hot air too... but that's something totally different! Been lovin' your channel for awhile now. Keep up the good info reviews and maybe help me figure out where I went wront with Proxmox and OPNsense on top in a VM... I cannot seem to get SSL for PM cluster working right. My domain name resolves to my private static IP (I have 5) and hits the WAN side of the PM server which has the OPN (VM) running exclusive on that physical port that then routes out the rest of the traffic back over the private lan side (I do not have managed switches so no vlan tagging ). I can get out from the cluster vm's to the real world IP and have perfect connectivity between nodes behind opn - but I cannot seem to get reverse proxy caddy or nginx or anything to work with OPN. Probably some dumb firewall rule I did not open but dang it is frustrating... been trying for about a year to get my various subdomains from my main domain to route through my OPN and to the correct vm ... Got any tips on where to read more to get it to work right?
I do like SuperMicro gear. I used to run a company on (Intel) SuperMicro boxes, and hearing the "ready for takeoff..." noise when you started up the machine brought back memories...
As someone who has to spend time every now and then at the server room I can relate to that intro. Also I find it very funny every time I see a set in movie or or television series where the set designers thought that server racks can be casually placed around the office like water coolers (and the noise is certainly not the only reason not to do that).
The louvre/damper fan is a brilliant move, because the fans themselves cools the CPU, others take over if a fan fails as you mentioned. As a former Solaris sysadmin here, like to add it's best practice decades ago to put "dummy" drives/fan holders and no space as it severely lowers the air flow efficiency.
For some fun, try using your ham radio near those smart plugs. Some of them are very susceptible to rf interference, and some cause a lot of rf interference. I had a smart power strip that started to flick on and off constantly. At first it only did it when I used the radio a few feet away, then it started doing it permanently after having my radio keyed down for a few seconds nearby.
Back in 2010 or so, I had an Itanium Superdome with 128 CPU cores and 2 TiB of RAM, and I've been eagerly waiting for the day when that sort of power can fit on a small desk instead of being as big as two full-sized refrigerators.
True; but I had 10 GbE and installed the U.2 drives a while back, actually-10 GbE is honestly plenty of bandwidth for multiple 4K streams of ProRes RAW or what I more commonly edit, H.265.
@@JeffGeerling Oh just joking around! :) I had some 40GE several years ago, but it was to span switches, not for edge devices. I'd love to play with 25 or 100, but funds tend to restrict that.
Really nice to see Patrick in this video. Love the constructive collab between some of my favorite channels and organizations. Would love to see Steve from Gamers Nexus do a formal interview of a Panel consisting of Jeff, Patrick, and Wendell from L1T!
Yep, in a DC it's hard to break, lots of kit around to display it and trivial to design in a board. Quite often it's the BMC driving the display output.
It's honestly super glad to see so much happening in this space, can't wait to see where we end up or at least when this core race ends... Wonder if when the ceiling will be reached in this current race, 1024 core CPU??? :P
Jeff had worried about going over 20 minutes but I told him it'll be fine, folks will love it! I even suggested Jeff try doing long form 2-3 hour videos like @drachinifel ua-cam.com/video/ApOfbxpL4Dg/v-deo.htmlsi=rPBBRSU9tA70-sQ2
"Except, you can actually buy one of these." Objection. I emailed Ampere's support several times and ASrock Rack's product pages are a nightmare and dont tell you if you are just ordering the board OR the board AND the CPU. I also tried alternative sources like ITScope but only saw ASRock Rack listings with even less description on there. I really, really want their 32core version - it's perfect for a homelab with high perf! RK3588s are great, but from microSD hell to annoying vendor support...I just want to move to something proper and keep it deployed for a while. The things I run from home are what I rely on daily - private, and work. So this would be epic to have - no pun intended. :p I also hear that these CPUs have ridiculous efficiency as well!
They do tend to be tough in terms of availability of all SKUs. It seems like they release batches of the 32/64/96/128 core variants here and there, and when they release, it's easy to get them... until it isn't again. And they're usually bundled with motherboards. Would love to be able to pop into Micro Center and buy an Ampere chip!
@@JeffGeerling I live in germany... Hardware availability here is terrible, even for most consumer things, since we often import from the states or china. Now imagine trying to get a "niche" thing; hence why I attempted to email them, hoping to hear back from them. Do you happen to know of a shop/retailer that sometimes sells them? I might as well bite the bullet and order it from overseas with import and shipment...i just need a "where".
Smart plugs on your servers is an interesting choice. Find a used APC datacenter PDU -- you'll get remote management for all your power outlets and more.
server market is an extreemly fickle market as it takes decades for there to be major movement in terms of adoption due to software. amd has been out competing intel for nearly a decade with their epyc servers which were better priced and more performant, and they still dont have a majority in the x86 server market, its very very dificult for other arm based server manufacturers to go out and compete without having extremly low prices, hyperspecific components and amazing efficiency. its a hyper competitive market, but with everything being said seeing the current server market revolution over the last nearly decade has been a very exciting time. we have to remember the top end intel xeons from 2017 were (im going off of memory i might be wrong) 20 core cpus with 300w tdps and costed anywhere between 20 and 50k$ per chip, to say we have seen a huge reduction in price is an understatement.
I like the aside with Patrick about benchmarking. It goes to show how important it is to pick good benchmarks, and why good hardware reviewers use many benchmarks and workloads.
Fanless or low-speed fan ftw! I've been trying to get as much of my rack gear fanless as possible, even if I can close the door, it's nice to be in there at 40-50 dB and not like 60-80!
Compiling the Linux kernel always used to be good fun, especially once it we started getting multi-core systems and being able to build with "make -j". And then of course "make world" for the BSD fiends. It wouldn't test the networking or the RAM (much) but running these various, very core-dense systems against each other as compiler farms could be an interesting comparison.
Right you are; and I realized I said QSFP during the edit... but forgot to fix it lol. Oopsie! I just told someone yesterday "I know someone's going to point that out in the comments"... and you win :)
I was scratching my head wondering if i had my monitor aspect ratio wrong every-time i heard "QSFP". Alas, it's just a sign you're working too hard, Jeff.
2 місяці тому+1
This is a paradise for those who prefer virtualization! :)
I totally don't need one of these but after hearing it roar when you turned it on I want one really bad. No idea what I would do with it except stare at it and turn it on.
@@ramdynebix pretty much. Even the official distributor here in EU has trouble getting them. They recommended I get a pre-built one from SuperMicro. I don't want a pre-build one. 😅
I have never met a micro USB port I could trust. They all failed me. I have never met a mini b port that I couldn't trust. Might be more durable than C.
Good job! Seems like there is enough space for X86, ARM, RISCV and Quantum computers to co-exist - depending on perpouse. Trust your health is doing well!
It would be about $5-6000 total for that setup (plus the cost of PoE switches!). Not completely out of the question, but I'm waiting for wider availability before I start thinking about that :)
When you showed the fans with the anti-backdraft dampers I was like oh yeah they've been using that in HVAC for a long time and surprised servers haven't been using that already.
240volt power is often used for efficiency it saves a couple of percent when applying it to the system and when deploying it in datacenters that saves $$$, particularly when you look at the power draw of the systems, not to mention that much of the world uses 240 volt.
@@JeffGeerling Also, what the heck is up with your plugs? They're terrible! No fusing, the sockets aren't switchable, the sockets wear out... I mean come on! Anyway, that's enough of my 'power outlet' based standup routine. How cool would it be if Apple got back into the server market with a new XServe, powered by an M4 'Enterprise' chip... ah, one can but dream...
This isn't anything new, Arm chips usually have a lot more cores than their X86 counterparts, this is simply because to make up for the simplicity of their CPU cores (since Arm is RISC/reduced instruction set), they make it up in cores, so a single X86 core can probably beat all of these Arm cores combined in double digits. Also, having more cores over faster less cores has its own disadvantages, there are many many algorithms which cannot take advantage of the parallel computing of multiple cores so they will run terribly slow.
The specs of this server will be the minimum hardware requirements for Windows 12
does it have a TPU? :P
Which is why windows 12 will only be in the cloud lol.
Too bad Win12 won't install without an NPU.
Sorry sir you lack 12 Nvidia H100 PCIE
There will be no Windows 12. The next version will be called Windows Cloud.
Software Vendors with core-count based licensing: "breathing heavily"
Yep, stay clear of Oracle and VMWare now!
you made me remember microsoft wanting to ask per core licenses, then they had to go to per socket license
Generally, companies that have that, that also work with non-x86/x64 processors usually have formulas that cut down on the cost for things like this. So they may do something like "Multiple core total by .50"
@@PSjustanormalguy Holy cow, been out of the sys & net admin game for a while, had no idea that VMWare moved to core based licences, that is ridiculous. We used to use VMWare & VSphere at my old company, I'll bet they moved away from VMWare like the plague after that.
@@betag24cn What are you talking about? Windows server has been core based for ages now... 8 core minimum per socket, 16 cores per system minimum... And that's a standard approach to a LOT of the higher end software suites that typically run on windows servers... Not that those normally require you to license the whole thing unlike the OS, but rather that if you license as an example MSSQL and you license say 8 cores, while your system has 32 cores... Well MSSQL will only schedule itself on 8 of those cores. So 24 cores will be idle unless you have other services running on those.
Finally, a sensible explanation of “edge computing.” Thank you.
Yeah Edge can be quite as silly as AI with how often the word is slung around. I figure it's just getting servers closer to where they're needed, a pushback from the Cloud-everywhere theme of the early part of this century.
Cloud: euphemism for "someone else's computer". Edge Computing: euphemism for "the cloud was a mistake", and putting shit back on your own servers in your own network.
(We've done both for decades longer than stupid MBA's have had trendy terms for them.)
Cloud is defined as IaaS provisionable via api instead of UI.
you could do a lot of edging with this computer
who up edging they computer rn
Linus got a 192-core EPYC, you have a 192-core Ampere. These videos are reminiscent of a time when car reviews were taking on the McLaren P1, LaFerrari and Porsche 918 in a narrow time frame.
What a time to be alive! Would be fun to see a smartphone with 64 cores haha
The holy trinity fr
@@ps3guy22 But for it to be.a true holy trinity we need a third player with a comparable chip. I was thinking Intel but maybe Nvidia has an answer...?
@@FlyingCIRCU175Apple?
@@FlyingCIRCU175 it would need to be a 3 architecture like RISC V to be truly a trinity
Couldn't find a comment mentioning it (although I admittedly didn't scroll far) but just as a note the front ports look like SFP not QSFP. If they're 25G they'll be SFP28. Love your videos Jeff
she's taking off in 5. 4. 3. 2. 1.
I should have put something in the box as an Easter Egg for Red Shirt Jeff and JeffyG! That is a fun little server!
Missed opportunity! Could've burnt the Linux src on a disc. ;)
JeffyG😂
The bright red STH shirt wasn't a Red Shirt Jeff reference?
Marketing should use teraflops going forward. Plus “Tflop” sounds like a new music genre.
Tay Zonday is a geerling guy ?? Hope you are well, man.
I'm sure LOOK MUM NO COMPUTER would be all over the Tflop genre!
@@seanunderscorepry he also watches cathode ray dude. Apparently he likes computers modern and retro
Doesn't work, because it flops.
chocolate rain tflop version when
"Nobody puts baby in the corner", and nobody would leave a machine like this sitting idle. It's a total workhorse and probably intended to be maxed out its whole life.
Love the content! Great video!
Ironically, ARM is VERY efficient at idle. x86 is EXTREMELY efficient under load.
If you sent this guy back in time to mid-2000, it would be the world's most powerful supercomputer. It would still rank among the top 500 as late as 2006.
A bit faster than the Celeron 333MHz desktop I was running in 2006
Nice work and collaboration, Jeff, Jeff, Wendel and Patrick! Already subscribed to all of you.
I love turning on servers in my lab & hearing fans get up to 100%. Just sip a hot mug of coffee while they boot & then settle. Badass IT moments.
i love that all my favorite youtubers are apparently buddies who casually compare benchmarks and do e/o favors
The tech side of UA-cam is the best side of UA-cam ;)
Back when I did development, we talked about vertical MIPS and horizontal MIPS. Big fast machines were more 'efficient' but lots of 'cores' allow running lots of stuff, but think of them as threads. It is great stuff, but if you can take the MIPS and add it all into one it was 'more efficient' and more computer programming to use more 'cores' efficiently. ... Enough from old history (starting in the '80s when I got my hand on them).
Im a junior dev rn. I really like to think of my personal projects and code as "services". It just makes it feel like a mechanical part that you would use in a car or something.
I feel like other ppl realized this too, cus microservices are way more popular than some big monolithic server program.
like yea the interactions between them has alot of overhead, but the fact you can scale them so easily the more silicon you spam into it, the better.
I used to make alot of roblox games with my friends back in HS. Even Roblox is adding more multithreading support to their games, which is really needed.
2:32: “Don’t think of this as a single 192-core server. Think of it more like 48 dedicated 4-core servers in one box.”
Waylon Smithers: “Now, I realize caring for Mr. Burns seems like a big job. But actually, it’s 2,800 small jobs.”
I just like how there is a VGA connector (DE-15 FTW!) sticking out of the middle.
If you are running SIMD-heavy workloads Epyc is probably still the better value per dollar.
yeah but can it play Crysis?
IBM used to have a refrigerator-sized RISC server called the p690 Regatta. When I stood behind one on the data center floor it was like a full-body hairdryer!
Theme parks should relocate mini datacenters. Just make the hot aisle a tunnel where the kids who come off water rides walk through!
@@JeffGeerling only to keep in mind that kid should be placed at the outlet side ;d
been waiting for this one
192 core mini PC when? :)
I wish Apple would adopt the Ampere philosophy for their Mac Pro.
@@JeffGeerlingjust make a Mac Mini stack and you’re all set 😂
@@JeffGeerling Perhaps Apple feels that they don’t have a big need for a lot of resource hungry services up in the cloud… oh wait.
This is a bonkers bit of kit, a truly mind-bending level of power. Loved Patrick's little piece, you have some cool friends Jeff. I will never have or need something like this, but it's great to have it explained.
0:15 Pretty expensive hairdryer...
I feel like I'm at an airport when I'm turning on a server
The 25Gbps interfaces are SFP28, not QSFP. Just for reference :)
Right, QSFP would be 40gbps, and QSFP28 would be 100gbps, as they are Quad-SFP and Quad-SFP28 respectively.
@@rkrenicki 40G would be QSFP+
@@N3m3515 Since QSFP non-plus (i.e 4Gbps) doesn't exist, I've seen it quite common to drop the plus when referring to the 40Gbps option for simplicity sake.
The whole QSFP naming saga is honestly kinda crap - especially when you look at stuff like QSFP-DD (which is actually QSFP58-DD) which is 8 lanes because it's Quad SFP Double Density.
@@rhysperry111 true. time for a new connector 😀
@@N3m3515 Like OSFP? 😄
Always nice to have Patrick to help in any video! The tech UA-cam is always on it. I really dont know what all of us techies would do with out all of ya!
0:10 Captain, Is there something wrong with the jet?
NO THE JET IS PS4ING
12:31 Craft Computing video inside ServeTheHome video inside Jeff Gierling video. We’re reaching uncomputable levels of inception 🤯
So, I just found out what size of box of RAM I want for xmas, thanks Jeff! :D
Lots of Chrome tabs?
0:19 Thought you were joking around with the sound of the fan at the beginning of the video Brother Jeff. Then I see your hair flow in the video from the fan I realized it was no joke.
You tend to not joke around in your videos which is perfectly fine. Your content doesn’t need it.
The funny thing is, I had to supplement the fans in the thumbnail with a leaf blower-because the louvres direct the airflow down. They actually blow with about the same force as my leaf blower from 3' away, but since the air goes down instead of up, I couldn't get the thumbnail picture without the blower!
@@JeffGeerlingDedication lol
@@JeffGeerling we appreciate the effort just wish you could have pulled it off. I would have wore a wig or something if I were in your shoes. Granted my hair is currently like that dude from Gamers Nexus!
I read Jeff's blog about this during my lunch at work today. Came home and proceeded to watch the video!
I love that companies like Supermicro started to adopt OpenBMC and getting rid of AMI’s bs / their own creation. On the other hand, it’s sad that they don’t publish the code / the layers so that I could build my own image (which is basically the philosophy of OpenBMC).
I was positively surprised to see supermicro using openBMC. Integrating RAID monitoring of supermicro servers in zabbix with storCLI was... Painful* to set up, to say the least.
*I'm not sure how old the servers are, but I'd say at least 3-4 Years.
Seems like a huge improvement so far.
Would be cool if in 5 years we could run self-build Coreboot and OpenBMC on such a thing. 🙂
American Megatrends International has been falling woefully behind on hardware support for a good 10 years now, it's good to see them being cropped out of the picture by enthusiast alternatives.
@@KiraSlith Don’t forget the thousands of dollars in fees that a company has to pay to AMI for every motherboard series (been there, no bye)
This thing is just stupid…
…stupid cool!
I like how the price is listed as “if you have to ask you can’t afford it”.
watched the whole thing, great video. thank you! what a beast.
I absolutley love your show Jeff.. Keep up the exellent work that you are doing. :) Never mind my misspelling to the left. Cheers bro.
Last time, you ended with "until next time, I'm Jeff Geerling." I asked who you would be after that. This time is now the next time referred to last time, and this time, we have a witness testifying that you are now Jeffy G. Mystery solved. The more you know.
I honestly thought the fan spinning up at the beginning was some sort of synth. I was not prepared!
Loved your video title. 192 core is indeed a lot in a single processor. I think it's time to compare it with a cluster of raspberry pi of equal core count.
Once CM5 is more available... would be a fun video to do
when @@JeffGeerling and Patrick did comparison with the Ampere Altra version of the same telco edge Supermicro server, the result was it equaled 100 RPi4 and was more energy efficient. ua-cam.com/video/UT5UbSJOyog/v-deo.html
Great Video Jeff, thanks a lot! As impressive as these machines are, as ridiculously are they in a home server context. What is bugging me: there are plenty arm machines on entry level and also machines like these from the end of the spectrum but it seems that other than Apple, no other manufacturer seems to be interested in the mid range so far. I hope Windows on arm will fix this as I have the definitely a strong wish for a 8-16 core 32GB Linux box in my home. If it was more straight forward to run Linux on a Ma Mini, that would be awesome.
Yeah, that's the huge gap that I wish wouldn't exist-Qualcomm has the Snapdragon X Elite, but it's still kind of Apple-ish in terms of not being used in contexts where you get upgradeability or hackability, like you would with a typical PC.
I can see people nod along with the video. "Of cores, of cores..."
Heh
OMG that sound reminds me of my server admin days. Such nostalgia! Bringing me back to the 2000s!
Finally. A cpu worthy of compiling everything from source on Gentoo.
STILL not as many cores as I want! I'm a hobbyist programming who's into HPC. It was so exciting to get a Dell Precsion Workstation t5500 with two 6-core Xeons, for 24 hardware threads. And it choked trying to write n-body simulations with over 1000 bodies. So I had to start learning GPU programming. SO difficult. I just want 10000 arm cores and to program it with OpenMP. Sigh. 192 is a nice start, but I need about 500 of them.
Still, a server like that with a few GPUs would be one helluva compute node. So many threads for MPI, other threads running CUDA code...This would be fun to program on!
Now you have me deeply interested in what the hell you are doing!
@@SchoolforHackers I'm just getting started but I wish to code complex, biological-like systems. With millions of agents, chemistry, physics like diffusion. Or imaginary physics. Like, design my own particles with their own forces - like but not the same as electromagnetism, the strong and weak nuclear forces, gravity and watch them interact. I got started just writing simulated flocking behavior, and 2D imaginary physics. But those xeons bog down with over 1000 bodies. I just started learning CUDA after discovering OpenACC isn't adequate. And OpenMP slows down when I offload to nvidia! What I wouldn't do to just write all that stuff to run on normal CPU threads!
@ Dayam - you are WAY out there in the fun zone! What you’re doing reminds me of complexity theory, and you’re playing a hugely upscaled Game of Life. The, um, security tester in me imagines a few milliseconds hijack of most of the Amazon cloud here, a few milliseconds there…but of course that would be unethical. Ethics being, perhaps, the fifth force.
Supercomputer type projects, I love it
@@richardzeitz54 So just possibly something like this. The girls, who are named for gems, “clink” when they run into each other. And the dogs … well.
ua-cam.com/video/gFlw55xZoSE/v-deo.htmlsi=pCfann45TXKDvP0k
I'm almost 80. The first computer I used in 1969: Philips P1100 (1C1T; ~6 cubic meters), 64KB (1MHz), ~$750,000? 55 Years from 192 cores.
In Apr 2019 a Desktop with Ryzen 3 2200G (4C4T); 16GB; CPU+RAM for $175 and on 11 Dec 2024 Ryzen 5 5600GT (6C12T); 32GB; $179.
I combined desktop and server in one PC. Host OS = Minimal install of Ubuntu 24.04 LTS; Hypervisor = VirtualBox 7.1.4; File System = OpenZFS 2.2.2.
The Apps run in VMs; Xubuntu 24.04 LTS (Email and Social Media); Ubuntu 16.04 ESM (Finances); Ubuntu Budgie 24.04 LTS (Multimedia); Ubuntu 24.04 LTS (Experiments); Windows XP Home 768MB (Jukebox); Windows 11 Pro (if needed) and ~60 other VMs.
Note that the CPU + RAM upgrade is the same price as the original ($175 vs. $179), while the CPU is 3x faster and memory is 2x the size.
I didn't know the idea of a core and thread went back that far. interesting
@@Zimbob2424 In the sixties and seventies the IBM 360 and 370 had computers, that could be bought with 1 up to 4 CPUs. I think Multithreading comes much later in the 21st century.
Wow, that thumbnail was not an exaggeration!
POV: The server guy having the same ear damage rock stars who play regular concerts next to large speakers do
Hearing protection: underrated!
Affirmative. :D
"Your hearing damage is not occupation related"
From experience of my colleague, I do know one thing, because this is a more monotone sound, it is possible to sleep in a server room, even though it's loud.
@@autohmae suuuuure, a colleague
I finally subscribed. I see you're on your way to a million.
I'm interested in a Pi cluster. Or maybe some other. It has to be cheap though and draw low power. And it needs to run Linux.
One of the things about all those cores is that they are still kinda slow. So many of my programs are single threaded. They only burn one core. I do some threaded software that can burn all the available cores. But that is in the minority. Even so, I would love a load of cores. Parallel processing is the only way I can speed up certain software.
Yeah definitely different strokes for different folks. I really wish Apple made the Xserve again, with like 64+ of their performance cores. Would be beastly.
@@JeffGeerling I don't know why they don't go with just performance cores in their plug in computers like the Mac mini. There's no battery to save unless you are smart enough to plug it into a UPS.
ahh that sound of massive servers starting up...
its like my stomach every morning ... thanks Crohn's...
At least the servers only blow out air!
@@JeffGeerling I definitely feel like a "blow out" is the better part of Crohn's - the alternative is massively painful - as you and I know. Yes, you are correct - servers blow out a lot of hot air... The wife tells me I am full of hot air too... but that's something totally different! Been lovin' your channel for awhile now. Keep up the good info reviews and maybe help me figure out where I went wront with Proxmox and OPNsense on top in a VM... I cannot seem to get SSL for PM cluster working right. My domain name resolves to my private static IP (I have 5) and hits the WAN side of the PM server which has the OPN (VM) running exclusive on that physical port that then routes out the rest of the traffic back over the private lan side (I do not have managed switches so no vlan tagging ). I can get out from the cluster vm's to the real world IP and have perfect connectivity between nodes behind opn - but I cannot seem to get reverse proxy caddy or nginx or anything to work with OPN. Probably some dumb firewall rule I did not open but dang it is frustrating... been trying for about a year to get my various subdomains from my main domain to route through my OPN and to the correct vm ... Got any tips on where to read more to get it to work right?
4:36 “what’s in the back” header - I see what you did there 😂
You found the inspiration for the "THX" sound! lol
Great video!
I was sipping coffee when you showed btop. Almost made a mess 🤣
I do like SuperMicro gear. I used to run a company on (Intel) SuperMicro boxes, and hearing the "ready for takeoff..." noise when you started up the machine brought back memories...
I heard they are a bit finicky with niggling issues compared to Dell/HP. Do you think so ?
Intro needed danger zone for that take off
As someone who has to spend time every now and then at the server room I can relate to that intro. Also I find it very funny every time I see a set in movie or or television series where the set designers thought that server racks can be casually placed around the office like water coolers (and the noise is certainly not the only reason not to do that).
Haha yeah.
I think Jeff is the only guy that can get me excited about a server
Don't watch Serve The Home then haha
@@JeffGeerling perhaps I will! thanks for all your great content jeff, love your sense of humor too
The louvre/damper fan is a brilliant move, because the fans themselves cools the CPU, others take over if a fan fails as you mentioned. As a former Solaris sysadmin here, like to add it's best practice decades ago to put "dummy" drives/fan holders and no space as it severely lowers the air flow efficiency.
A cap/cover is better as louvres reduce airflow quite a bit.
For some fun, try using your ham radio near those smart plugs. Some of them are very susceptible to rf interference, and some cause a lot of rf interference. I had a smart power strip that started to flick on and off constantly. At first it only did it when I used the radio a few feet away, then it started doing it permanently after having my radio keyed down for a few seconds nearby.
Back in 2010 or so, I had an Itanium Superdome with 128 CPU cores and 2 TiB of RAM, and I've been eagerly waiting for the day when that sort of power can fit on a small desk instead of being as big as two full-sized refrigerators.
20:45 You haven't run 25GE in your NAS for almost a year. You just installed the card! lol ;)
True; but I had 10 GbE and installed the U.2 drives a while back, actually-10 GbE is honestly plenty of bandwidth for multiple 4K streams of ProRes RAW or what I more commonly edit, H.265.
@@JeffGeerling Oh just joking around! :)
I had some 40GE several years ago, but it was to span switches, not for edge devices. I'd love to play with 25 or 100, but funds tend to restrict that.
❤
Really nice to see Patrick in this video. Love the constructive collab between some of my favorite channels and organizations. Would love to see Steve from Gamers Nexus do a formal interview of a Panel consisting of Jeff, Patrick, and Wendell from L1T!
We just need Wendell to apparate like he does in those GN videos.
That is one loud as hell server. I love it!
1:42 And we're still using VGA.
Yep, in a DC it's hard to break, lots of kit around to display it and trivial to design in a board. Quite often it's the BMC driving the display output.
good. cuz its good. good.
@@AffectionateLocomotivegood.
Most datacenter crash carts still use VGA. Since it's only needed to display information, there's no reason to use anything fancier.
@@0x1EGENSo, no Crysis?
It's honestly super glad to see so much happening in this space, can't wait to see where we end up or at least when this core race ends...
Wonder if when the ceiling will be reached in this current race, 1024 core CPU??? :P
Love the swirl shirt. Didn't even know they had merch like that
Heh, I need to buy a new one, this one's fading fast!
hahaha, that boot-up in the beginning sounded like a starship turning on its engines and put the biggest smirk on my face ^-^
Would this be good for a minecraft server...?
It is certainly a good fan!
Oh man, I have never beaten my ADHD by watching a full 25 minutes video until yours. Thanks, Jeff.
Jeff had worried about going over 20 minutes but I told him it'll be fine, folks will love it! I even suggested Jeff try doing long form 2-3 hour videos like @drachinifel ua-cam.com/video/ApOfbxpL4Dg/v-deo.htmlsi=rPBBRSU9tA70-sQ2
"Except, you can actually buy one of these."
Objection. I emailed Ampere's support several times and ASrock Rack's product pages are a nightmare and dont tell you if you are just ordering the board OR the board AND the CPU. I also tried alternative sources like ITScope but only saw ASRock Rack listings with even less description on there.
I really, really want their 32core version - it's perfect for a homelab with high perf! RK3588s are great, but from microSD hell to annoying vendor support...I just want to move to something proper and keep it deployed for a while. The things I run from home are what I rely on daily - private, and work. So this would be epic to have - no pun intended. :p I also hear that these CPUs have ridiculous efficiency as well!
They do tend to be tough in terms of availability of all SKUs. It seems like they release batches of the 32/64/96/128 core variants here and there, and when they release, it's easy to get them... until it isn't again. And they're usually bundled with motherboards. Would love to be able to pop into Micro Center and buy an Ampere chip!
I guess they just don't have the volume yet, to get a good slice of TSMC production time ?
@@JeffGeerling I live in germany... Hardware availability here is terrible, even for most consumer things, since we often import from the states or china.
Now imagine trying to get a "niche" thing; hence why I attempted to email them, hoping to hear back from them.
Do you happen to know of a shop/retailer that sometimes sells them? I might as well bite the bullet and order it from overseas with import and shipment...i just need a "where".
@@IngwiePhoenix_nb System76 is just getting rolling, you can always email them direct-otherwise, NewEgg is the most consistent retailer to have stock.
Smart plugs on your servers is an interesting choice. Find a used APC datacenter PDU -- you'll get remote management for all your power outlets and more.
server market is an extreemly fickle market as it takes decades for there to be major movement in terms of adoption due to software.
amd has been out competing intel for nearly a decade with their epyc servers which were better priced and more performant, and they still dont have a majority in the x86 server market, its very very dificult for other arm based server manufacturers to go out and compete without having extremly low prices, hyperspecific components and amazing efficiency.
its a hyper competitive market, but with everything being said seeing the current server market revolution over the last nearly decade has been a very exciting time.
we have to remember the top end intel xeons from 2017 were (im going off of memory i might be wrong) 20 core cpus with 300w tdps and costed anywhere between 20 and 50k$ per chip, to say we have seen a huge reduction in price is an understatement.
I like the aside with Patrick about benchmarking. It goes to show how important it is to pick good benchmarks, and why good hardware reviewers use many benchmarks and workloads.
I love that debian shirt by the way!
Also I noticed it wasn't running some outdated Ubuntu, but an up to date LTS machine.
(as it's based on Debian, it is sort of the right shirt)
Multi-creator UA-cam videos to explain multi-socket machines. We are in the golden age of homelab UA-cam boys
Those "moar cores" memes from the early 2010's are about to make a comeback.
Here from short circuit, and now i'm subbed, great video
Hey Jeff, this sounds fancy and all, but can it run Doom?
That's actually a great creation, a good cpu and a good heater, 2 in 1, with the price of 1
Since I don't have additional cooling in the server closet, if I close the doors and run a benchmark, it gets nice and toasty in there, pretty fast!
"it's whisper quiet!" Dr Jeff Riviera.
Jeff .. i can honestly say thanks for introducing me to low powered computing ,, my ears and power bill thank you 🥰
Fanless or low-speed fan ftw!
I've been trying to get as much of my rack gear fanless as possible, even if I can close the door, it's nice to be in there at 40-50 dB and not like 60-80!
2:10 And intel aims to freaking survive XD
Compiling the Linux kernel always used to be good fun, especially once it we started getting multi-core systems and being able to build with "make -j". And then of course "make world" for the BSD fiends. It wouldn't test the networking or the RAM (much) but running these various, very core-dense systems against each other as compiler farms could be an interesting comparison.
3:45 Those look awfully similar to SFP28 ports. I thought QSFP was wider.
Right you are; and I realized I said QSFP during the edit... but forgot to fix it lol. Oopsie!
I just told someone yesterday "I know someone's going to point that out in the comments"... and you win :)
@@JeffGeerling Yay! What do I win?
At least it's the guy who works for a NAS company. lol.
@@MarcoGPUtuber Ooh QSFP 100 Gbps all-flash NAS when? :D
Excited to test out one of the AMD revisions though...
@@JeffGeerling Ooh i wish, but I am trying to convince them to change a bit of direction in our rackmounts. I have an idea.
I was scratching my head wondering if i had my monitor aspect ratio wrong every-time i heard "QSFP". Alas, it's just a sign you're working too hard, Jeff.
This is a paradise for those who prefer virtualization! :)
2:52 I cosplay as Red Shirt Jeff and unleash havoc instead.
I totally don't need one of these but after hearing it roar when you turned it on I want one really bad. No idea what I would do with it except stare at it and turn it on.
Patrick in the middle of a Jeff video???? :D
I am forever grateful at the person who decided Star Trek would invent the term “gigaquads” for performance. It’s aged very well.
I've been looking for an Ampere Altra here in EU, and it turns out they're nearly impossible to get. #sadface
Sadly it's tough to get System76 over there, and I'm not sure about ADLINK but that might be an option.
I wouldn’t be surprised it’s a matter of if you have to ask where to get them you can’t get them?
@@ramdynebix pretty much. Even the official distributor here in EU has trouble getting them. They recommended I get a pre-built one from SuperMicro. I don't want a pre-build one. 😅
0:04 i actually thought you were a about to meme about the old cinema THX intro, then i realised thats actually what the server sounds like 😅😂
3:13 Micro usb 💀
I just got a Pi Pico 2 W, it can be best friends with that port :D
I have never met a micro USB port I could trust. They all failed me. I have never met a mini b port that I couldn't trust. Might be more durable than C.
@@katanyafleet Yeah, minus the size and speed, mini b is great, and I'm glad for the EU usb-c law
why not just use something else? theres so many ports they could've used why micro-b?
@@ThreeTreee Yeah, that was kind of the point of my comment
21:15 yes, please! 👀
Didn’t know fans had loofas.
Good job! Seems like there is enough space for X86, ARM, RISCV and Quantum computers to co-exist - depending on perpouse.
Trust your health is doing well!
192 ampere vs rasberry pi blades with 48 x4 =192 cores. Make it happen.
It would be about $5-6000 total for that setup (plus the cost of PoE switches!). Not completely out of the question, but I'm waiting for wider availability before I start thinking about that :)
@@JeffGeerlingTough decision, are you going to please the crowd, or save some money for Christmas. 🤷
@@JeffGeerling you might be able to get a bulk price, especially you
When you showed the fans with the anti-backdraft dampers I was like oh yeah they've been using that in HVAC for a long time and surprised servers haven't been using that already.
I'm guessing they've been used in some server lines a while, just this is the first time I've noticed them in person!
0:03 THX sound test - Server Edition
Underrated comment!
240volt power is often used for efficiency it saves a couple of percent when applying it to the system and when deploying it in datacenters that saves $$$, particularly when you look at the power draw of the systems, not to mention that much of the world uses 240 volt.
*Laughs in UK standard 240V outlets*
If only we would all be so fortunate!
@@JeffGeerling Also, what the heck is up with your plugs? They're terrible! No fusing, the sockets aren't switchable, the sockets wear out... I mean come on!
Anyway, that's enough of my 'power outlet' based standup routine.
How cool would it be if Apple got back into the server market with a new XServe, powered by an M4 'Enterprise' chip... ah, one can but dream...
I would get that for the white noise to get better sleep
This isn't anything new, Arm chips usually have a lot more cores than their X86 counterparts, this is simply because to make up for the simplicity of their CPU cores (since Arm is RISC/reduced instruction set), they make it up in cores, so a single X86 core can probably beat all of these Arm cores combined in double digits.
Also, having more cores over faster less cores has its own disadvantages, there are many many algorithms which cannot take advantage of the parallel computing of multiple cores so they will run terribly slow.
Right! What matters is power per task. Lightweight web servers? ARM is great. Monster CAD rendering workstation? x64 please.
the sound of servers booting up, music to my ears!
0:48 yk what else is massive?
Your heart :)
My electricity bill?
MY MOM!!!
love the VGA on the rear (err, front)
Thanks for being you, Jeff.