RDNA3 - Is it AMD Radeon's Bulldozer Moment?
Вставка
- Опубліковано 17 сер 2023
- In the first two parts I covered Performance per Watt and Performance comparisons to RDNA2. Based on the data, is RDNA3 AMD Radeon Division's Bulldozer moment?
RDNA3 Performance Playlist:
• RDNA3 - Performance
#RadeonBulldozer, #RDNA3Bulldozer, #AMDBulldozer - Наука та технологія
I should have clarified this in the video, this assessment is only based from the perspective of the gamer who buys a GPU for his gaming rig. Not from the perspective of an investor or shareholder. The impact of the RDNA3 generation to the financial health of AMD will not be significant (like Bulldozer was to AMD) as it does not sell many GPUs to gamers in the discrete GPU market.
In terms of performance per watt - yes. But remember that top end RDNA3 offering is on par with 4080, where FX-8150 wasn't even close to i5-2500K, while consuming tons of watts.
False the RTX 4080 is really a 4070ti. Remember Nvidia is selling their GPU’s one tier below what they actually are aside from the 4090. So the best that AMD has can only match the mid range of Nvidia. This is why AMD is pulling the plug on the high end. They will just focus on APU’s for low end and mid range discrete GPU’s.
Yea dude amd literally with maintaining their nomenclature strategy from last gen when comparing gpu shaders n clocks couldn’t compete so they named what previously would’ve been their higher end cards a lower sku the next go around and cut out actually offering a true competitor of a higher end card. Nvidia did the opposite naming weaker configurations with higher names (except 4090) and it still ended up balancing out in the market. That’s why AMD made up the XTX name, it appearing as 7950 XT would’ve exposed them so they padded that tier lol. Nvidia could’ve buried AMD but chose for their own reasons a different strategy
@@ChiquitaSpeaks Exactly imagine how hilarious it would have been seeing a 4070ti crush AMD’s 7900xtx.
@@ZackSNetwork yea a 4070ti on par with a 7950 would’ve been sad and it would’ve been cheaper and more efficient
It really isnt like Bulldozer. Bulldozer actually had some regressions versus its previous arch. Rdna3 is poor as an upgrade but it is always faster than rdna2 and more so in RT. It isnt good enough but it isnt a bulldozer moment.
How much boost to RT does it actually have? I see people saying there is an improvement in RT between RDNA2 and 3 but not really seeing much of that in the actual benchmark numbers.
@@olnnn Check Chips and Cheese on his RDNA3 deep dive and RDNA2/3 RT benchmarks.
But technically, it is around 70-80% better in RT (note - the lower Raster improvements neuter this in actual games!) for big RDNA3 and around 20% for small RDNA3.
I can't see how AMD can produce a radically better GPU architecture while maintaining backwards compatibility for Sony and Microsoft's consoles. Nvidia and Intel can completely redesign their architectures with every generation, AMD are fighting with one arm tied behind their back by maintaining hardware compatibility with not just older RDNA but GCN.
AMD is always the underdog, year after year they do not catch up to nvidea. So i would not have any hopes with them. The truth is peak gpu tech is only with Nvidea. Thats why they set such high prices because they know there are no real competitors.
That’s true for those who seek peak …. But I’m willing tho bet most users and oems are willing to put less money down on good performance…that’s AMD’s hand…. And it’s a good hand,I have a feeling it’ll be THE only hand the next few generations giving nVIDIAS meal ticket lies in AI.
I've been an ATI Rage/Radeon user since the mid 90s and never disappointed with the price performance ratio. I would never buy an nVidia card after some of its early misbehavior and its current issues on Linux.
Great job picking up on the lack of progress from clock speed and efficiency. Apparently rdna4 is just going to be improvements on efficiency. Good for laptops. Have to wait till rdna5 for good generational improvements apparently.
That makes sense, just focus on laptops until they can get RDNA5 for the desktop.
It'll be interesting to see RDNA3 performance in UE5 games. The first UE5 game utilizing both nanite and lumen is Immortals of Aveum, coming out in 2 days time!
RDNA3 will look good compared to RDNA2 and Ampere, but will look bad compared to Lovelace.
@@johnc8327Know imagine if Nvidia didn’t purposely make their Lovelace line up aside from the 4090 be compete garbage. I mean by giving us great specs and prices for the 4080 down to the 4060. AMD would be %100 dead.
Personally i feel like rdna 3 target was just to get chiplets working
Wait for RDOA4 that i fear will be another fail
AMD cancelling Navi 41/42 (according to rumor) already tells us its RDOA4. Maybe RDNA5...
@@ImaMac-PC We could still end up getting some decent mid range and budget offerings from N43 and N43 even if there is no 5080/5090 competitor. Too early to tell yet.
RDNA3 is like ryzen first gen without the affordability. Ryzen first gen was slower, drivers issues, not efficient, and software was not ready for high core count yet. But set the path for the future.
Interesting analysis!
Thanks!
I definitely agree with that video statement.
I don't think Navi 33 was designed as anything more than a slight die shrink to Navi 23 being on a cheap 6nm process , and compared to Navi 31 and 32 it also has a smaller vector register file per SIMD (see ChipsAndCheese). There's only a 18% increase in transistor density between N6 and N7+ TSMC nodes and the power consumption is the same.
I'd like you to rerun this in a couple of months - but with an RX 6800 and whatever SKU the full 60CU Navi 32 chip ends up landing in instead... Or the 6800xt v 7900gre if you can get hold of a copy.
No doubt that the RDNA3 has been underwhelming but I'd liken it to Vega v Pascal (like Sandy Bridge their competitors made a massive step forward . Bulldozer was hopelessly off the pace and possibly worse then an aging Phenom architecture , but at least RDNA3 is in the ballpark with redeeming features like Vega).
What AMD should have done , like they did with Vega is price the GPUs very agressively , the 7900XTX should have been a 7800XT priced at $749 , and the 7900XT should have been an RX 7800 and priced at $649 - I'm sure they can maintain margins while beimg much cheaper than nVidia.
Good analysis
Thanks!
Dude, you are hitting the nail on the head with these videos! They're well put together, factual, and persuasive. These are things I've been saying the whole time through observation. I'm happy you've actually put the ideas to the test and proved it. Hopefully You get a lot of support for your thoughtful analysis. All I've been seeing lately is people getting huge attention for riding AMD's D and dogpiled and attacked for calling them on their Ls.
Thanks!
You're risght, pointing to facts for weeks or months has been a rough ride on the Internet comments lastly...no matter you said 5700XT was amazing (now that almost 2080 super like or 2070 at least) and below 150€, that RX 6000 cards where better tahn obsoleteRTX 30 (by nvidia itself not allowing all new sw tech to be processed by old cards), or that RTX 40 cards were better the cheaper they were, being 4060 better value/money/power and only followed by 4070...but 4090 has ome problems of optimization on games also as 4070 ti is 2x 4060 equivalent and 4090 3x 4060 instead of 3x and 5-6x...FPS are lost all over the gpus and cpus combinations, and stock/overclocked power levels...
Do you think is possible that the new 7800XT includes some hardware fixes/improvement with respect to the first graphics cards launched in 2022?
Moore's Law Is Dead just did a video for this. his leakers are saying that they killed the high end gpu's this gen because of delay's that would push them into rdna 5. the bugs in the mcm are enough that they have to work them out before doing another chiplet design and they think that it'd waste too much money to do a high end next gen. they are trying to do a Polaris this next gen, they are going to give us an affordable low and mid range while they work out the chiplet design that let us all down in the 7900 xtx. since they'd have to push it back into rdna 5, they are just going to improve rdna5 instead. he has die renders of the new chiplet designs and they look totally different than the 7900 xtx.
@@craiglortie8483 thanks for the answer. But I don't mean RDNA4, I mean the 7800XT which is still RDNA3
@@Mastercar77 he said in the video that the leakers aren't counting out rdna3, we will have to wait and see. :)
@@craiglortie8483 I had a typo in my last comment 😅
maybe chiplet technology is the culprit of the bulldozer moment. i think for same reason Nvidia don't want to hop in chiplet anytime soon.
Great video. Keep these unique videos coming.
IMO. RDNA3 achieved little, the dual-issue ALU does not effectively improve gaming. However i heard is useful in productivity applications. Maybe you can make a comparison video between RDNA 2 and RDNA 3 in productivity applications.
Thanks! I'll look into that.
@@ImaMac-PCI like to play around with machine learning and stable diffusion. For that reason I got a 7900 XTX because it had a lot of VRAM. However, your videos are gaming centric and I am not the average consumer of these GPU;s. That said, there is a very strong improvement with stable diffusion where the 6000 series just sucked. So they did improve something, its just not affecting games at all.
As noted 7600 rx is the same as 6650xt. From what I have seen I assume that those cards aim at 1080p and they can do the job even at 1440p but at this resolution it is beeter to use cards with more vram. So what is the point for rdna or xess or dlss if you are playing in 1080p? Doesn't the image degrade too much if you do that? Because if that is the case then even if rdna3 worked as intened it would have been a waste for 1080p users.
I switched from nvidia a while back when I purchased a second hand rx480 which I was able to play froza horizon 4 and gears of War 4 at 1440p 60 fps, replaced that with an rx5700, flashed xt bios and primarily use it to play automobalistia 2 at 1440p 90hz. Both AMD cards left me with the "fine wine" feeling. Looking to the future I don't see myself buying another RDNA card due the current power efficiency of the nvidia counterparts. That being said I'm finding it hard to justify the purchase of a current gen card at msrp.
Great comparison, thanks for your discovery.
Hopefully new generations both Nvidia and AMD will be more price/perf competetive than current.
Still not complaining much, got used 3080 10gb for 550usd in november 2022. Wasn't intensivly mining one if mining even. Probably someone changed to 4080 or 4090 and sold his.
Funny enough, used 3080 are still 550usd (in Ukraine) after 1 year has passed. While new are at 649$ the cheapest.
Quite like your videos, glad the algo directed me your way. I've gotta say, I don't get why you're called "I'm a Mac" though? Anyway, another interesting analysis, I would say we're much closer to whenever the equivalent of Zen 1 was, not because RDNA hasn't been successful, but because RDNA4 is similar to Zen 1 in that it's a new architecture, chiplets, and V1 isn't all that it can be, RDNA5 we should all be rooting for, whether it's versus Intel, Nvidia, both or neither. If RDNA5 is like Zen 2, then we're not far away. Lastly, 7900XT vs 4080, if we can get a scenario like the Vega 56 vs 1080 (where when you overclocked the Vega 56 it got to 1070ti/1080 performance at massive power consumption), I'm interested, so good luck with that!
Thanks. The name is how I started... I need to work on that. The 7900xt is always compared to the 4070ti however, it should be more capable, like you mentioned with Vega.
I think the difference is: back then, it was their CPU division that did bad. And that's their main division. Now it's their GPU division and they got a lot of money in the last 4y or so.
Now they do: Ryzen, Threadripper, Epyc, PS5 and Xbox chips, handheld APUs.
They have a lot more streams of income and I think they could support a fall in GPU sales.
What's more, it's hard to compete with Nvidia in high-end. They don't have the mindshare and features.
All in all, I think it might not be a bad idea to focus on sub $500 GPUs, as long as they offer better price/performance.
It's bad for us consumers though. Imagine how much Nvidia's gonna charge for their top cards once AMD gives up
I agree. CPU division is where they make the money. The retails sales from Radeon on the desktop is miniscule. And that's why they don't care how they price the GPU.
@@KurisuKun I agree. Just like with intel in the 4c era.
On the other hand, if options are good at 200-500, it's reasonable. Some people will pay whatever for the best. For the rest of us plebs that won't pay that sort of money, no matter what, the market would enter equilibrium, as long as there are options.
Too many people pay way too much money for this hobby in order to sustain absurd pricing.
At some point, the AI market will be saturated, so they would have to return to gamers.
Hey man i just found your channel recently and let me just say i’m impressed by all of your research and findings on GPU architectures. I’ve always wanted a channel like this. You come off as if you could be a computer engineer or something!
Thanks!
I don't get the comparison of the RTX 4070 to the 3070,Ampere was on a crappier node so the uplifting in has to be the enormous increase to frequency on top of architecture improvement. You didn't stabilize the frequency as close to Ampere to show both generations of GPUs compared to one another.
I've been posting exactlythis take on here for months. RDNA3 = Bulldozer.I remeber in 2011 I had a 1090T that did 4ghz stable, that thing sold above what I paid for it the morning Bulldozer launched lmao.
I upgraded from a 6950xt and it’s a night a day difference to a 7900xtx. It’s funny ppl keep crapping on RDNA3 but it is a great GPU. I am a huge overclocker. In games that take advantage of the new architecture it’s a huge difference. Even in games that don’t it’s still a decent jump and if your talking to performance to $ it’s a no brainer.
The XTX is the 2nd fastest GPU in the world and with some sales, can be almost half the price. I really like the value of the XTX.
Totally bombed
Yep they should just move on soon as possible and they made so many
..I don't see a fixed 3.5 RDNA coming unless for pro console
I have an FX cpu on a shelf - back then I didn't know better. It go tthe job done, wasn't the best buy but was cheap. I am glad the division survived and I hope radeon survives. If they can maintain their deals with Sony and Microsoft they will.
I wonder how far off they are from getting two GCD (? proper term) on one board operating efficiently. Then they could survive with RDNA2 performance through die shrinks and having frames render to alternating hardware targets
I have heard from a lot of people who built with the FX and thought it was great because it was cheap and as you said, it got the job done.
I see RDNA3 more like chiplet design test drive, the potencial for chiplet design its great, the first gen was not impressive but the idea is. Let them work the kinks, then when they make a BIG chip like a GPU Threadripper, NVIDIA will be behind.
haha I bought an FX 8350 at launch, needless to say I learned a lesson in buying into hype back then..
I look at it from the other end, they're going to launch the 7700 & 7800 which are going to be pretty much DOA because we've had that level of performance. However, 7900xt is down to 750 at microcenter with free starfield premium which makes it effectively what the 6800xt launched for and it's like a 50% performance uplift over my current 6800. Sure, I could wait 2 years and maybe get that performance at $500 or get it now and enjoy starfield maxed out and pretty much be set for years and years.
I still maintain the 7900xt is the true successor to the 6800xt and they should have launched it at $749. Then, with the $99 game, it is a worthy upgrade to thh 6800xt.
@@ImaMac-PC facts, if the 7900xt is the 7800 non XT for 700... It's an amd world. Something has to have gone wrong late, for such a rigid lineup. Cards have to been made and ready to go with higher performance expectations. Higher efficiency per core and this lineup is fire effectively pushing the 6950xt to the 6800xt slot.
The hard truth is , compute, laptop, hpc , console and AI are taking up resources from discrete GPU over at AMD
Looking at earnings they can make money there and better do so regardless of our feelings.
AMD. Actually was poised to be very successful after the athlon64 and phenom days in other fields such as mobile GPU(imageon) embedded CPU (Alchemy) and heterogenous processing. All of which was ignored to focus on the core business....that failed with bulldozer. The future was fusion but it was the FAR future.
Accept it now, AMD is a server /hpc CPU maker, and GPU compute accelerator/AI card maker first and foremost. Desktop and hedt are tertiary, discrete GPU even further down than console GPU.
@@ImaMac-PC 100%, I've been saying this since these cards first launched. The 7900XT should've been called the 7800XT had AMD not tried to do what Nvidia is doing and the 7900 GRE be the 7800 non XT and the 7900XTX just a 7900XT.
5:20 total system power draw @ wall is the way to go, even including display monitor, thanks FPS need a lot of hardware to be made and seen by someone..
my experience with the 7900x was giving me driver issues. But not the 6600 nor the 6750xt...
If people want a reason of why RNDA3 is a failure it’s not just the sales. Performance per watt it is a failure. Raytracing performance wise it’s a failure. AMD’s top end GPU the RX 7900xtx is competing against the 4080 which is really a 4070ti. So AMD over hyped their new architecture of GPU’s and their very best matches Nvidia’s mid range in rasterization performance. But, then gets completely annihilated in wattage/efficiency and raytracing performance. RNDA3 is a complete failure and embarrassment to what AMD has achieved with Ryzen as a whole and even RNDA2. No surprise they are getting out of the high end GPU market.
The 4090 is 26% faster than the 7900xtx and if the 5090 will be 50% faster what is possible thanks to 3nm AMD knows they have nothing to look for in the higher-end segments.
With 50% faster 5090 the 8900xtx would have to be 89% faster to match it. Good luck with this especially if this chiplets design is broken and 529mm² (304+225) 7900xtx competes only with 70-class 380mm² monolithic AD103 with half the cores (53% to be precise) of the full die. RDNS3 indeed made close to zero progress. The 7600 with the same specs and clocks is 3% faster than the 6650xt, 7900gre with the same specs and faster memory and clocks is 3% faster than 6900xt. RT also didn't improve at all. Intel will catch them sooner than they will Nvidia.
I agree, Intel will likely catch them sooner than most think. Maybe if Intel goes after their console business, AMD will get serious with Radeon.
I totally agree with u but i must say somehow most of them think rdna 3 worth the buy i dont get it
Bulldozer CPUs were still available and selling in late 2018 and early 2019...I never understood that
I hope AMD gets things figured out. My main concern is going for a technically more challenging chiplet design while having less R&D resources makes it hard to close the technical gap to Nvidia (for as long as Nvidia stays monolithic).
Like why should we have confidence RDNA5 will be competitive on the high end, if by that time AMD has struggled all RDNA gens but #2.
And RDNA2 was looking good just because Ampere was made on outdated 8nm Samsung so AMD had a huge node advantage.
Once Nvidia switched to TSMC we can see where RDNA really is. Is not just about broken chiplets design where 204mm² full N33 competes only with 159mm² 50-class AD107.
well the way I see it, nvidia will eventually have to go chiplet design as die shrinking isn't really a think anymore, maybe we can get down to 2nm? but either way its going to end soon, so unless nvidia can somehow make the same manufacturing process far more efficient they'll probably have to switch at which point they might experience some of the issues amd is facing adopting this method early.
this is really a stretch. This assumes everyone upgrades gen to gen as well, which is simply not the case. Just comparing to the last immediate last gen only makes no sense. I had the hd 7970 before I even bothered upgrading to RDNA2 6000 series
Are you really a Mac? Do you hackintosh?
You should see my earliest videos on "Big Mac" and "Mac 'da Ripper".
🐘🦣🏕: 10:26 "Premium RDNA3 7900XTX: Architected to exceed 3GHz - industry 1st, 61 Tflops, boost FP32: 2.7x increase, 1.54x PErf/watt for a 3rd generation" why iit FPS are somewhere between 4070/4070Ti/4080? RTX 30 to 40 Tflops FP32 are equal to 61 Tflops FP32 AMD? addition and multiplication intensive code could benefit from AMD architecture according to LAplacian part 4 finite element on GPuOpen, for sure?
I bought a new GPU yesterday. With my budget, the choice was between a discounted 7900XTX for 1,200 EUR or a standard price RTX 4070 Ti, also for 1,200 EUR. The RTX 4080 costs a staggering 1,500 EUR at minimum, even though that includes 25% VAT in my country that's still too much for what you get. I haven't had an AMD GPU in decades since they first acquired Radeon, I always get the top-tier Nvidia cards but now they are simply out of my reach lol. I'm a poor AMD pleb again, kekw
That was the right choice. For the same money, I would have done the same thing. The 24GB on the XTX will stretch out the useful life of that GPU far longer than the 12GB on the 4070ti.
Gr8 analysis again and there is another elephant in the room.
HiGH Power Consumption of AMD GPUs @ video playback !
Last Gen's RX 6800 (and up) AMD GPUs & the new chiplet RX 7000 series have serious power issues @ video playback (youtube, Netflix, video players etc.)
40+ W is stupidly HiGH / meaningless... and while the previous 6700 XT consumes only 16W, the new 7700 XT doubles that power draw with a whooping 33W on "video playback"
AMD should have solved that "power issue @ video playback" on the current Gen but it's gotten worse... Can u make a video about it too ? 🤔
Rdna 3 is jsut rdna2 but relying on 5nm and extra specific additions live av1 and extra Ray core
The 7600 was supposed to cost 299 btw
Yes, NVIDIA had to publish the 4060 data and pricing early so that AMD didn't get destroyed in the reviews.
I'm not worried about AMD or Radeon. The Ryzen 8000 APUs with 40 CUs will destroy the mid range GPU market. Event the 12 CUs was enough for me to replace a desktop with a GTX 2060 with a BeeLink SER6 (Ryzen 7736, Radeon 680m) . I can play any game I want at 45 Watts without all the nVidia Linux issues.
Amd realised they can`t compete with nVidia so they just stopped. Eventually nvidia will stop alltogether making consumer gpus for plebs like you and me (although i have a 4090 i payed 2k euros for my 4090, nowhere near the 30k for a hopper gpu soooo.... dirt cheep basically) so nVidia will leave this sector of the market to AMD and Intel i pressume as selling soemthing for 2k when you could sell it for 30k+ makes no sence what so ever. Love your content mate please keep it up because they way you explain things is EXCELENT!
Thanks. Before AI, I would have never have believed NVIDIA would ever leave the gaming market, but you're right, they will likely follow the money.
That's mad, 30k! I would leave gaming too lol
It's a setback for current plans, but IMO, chiplets are still the way to go, and AMD is the only one making headway into that.
And while the rumors of RDNA4's high-end being cancelled does sting, there's also rumors that Nvidia is also going to skip out on some mid-range dies, so RDNA4 could still work out.
As a 7900XT owner I agree with that, RDNA3 is disappointing.
It's not a terrible card but not exciting either it delivers 3060ti like performance in most games at 1080p at 260$ (the XFX swft model)
It’s not all gaming performance. I have an RX7900XT and for compute it’s a total monster. It spends hours every day doing AI noise reduction to raw files and outperforms the RTX4080 we have here.
That is very interesting. Thank you for that feedback. May I ask what program you use for AI Noise Reduction?
@@ImaMac-PC Adobe RAW converter which is accessible through, for example, lightroom.
Its not Bulldozer moment its inovation in the right way but they dont have people nor people for drivers nor money to make it in that small window time frame....
7600 has 2x TFlops vs 6650xt but how to use it.....Even vs RTX4000 series AMD has more TFlops .....
Bulldozer was significantly worse than RDNA3 imo.
Amd IMO, I bet if you normalize clock speeds, the 4060 TI would lose to the 3060 TI.
4060 Ti vs 3060 Ti is less of a like for like comparison though as the 4060 has half the memory bus width and less cores. The non-TI was also cut back from 192 to 128-bit. The change from 30 to 40 series is more like from RDNA1 to RDNA2 in that way, a move to narrower memory bus but larger cache size which makes a like for like per clock comparison a bit more difficult.
it does lose outright, but it's literally less gpu. the7900GRE has roughly identical specs to theRX 6950XT and it's outright slower.
Isn't 7600 using a cut down version of RDNA3, with less ipc?
Several people have commented that it is not exactly the same. And it isn't since the navi 33 die is not chiplet based. However, the changes are not significant to affect rasterization performance in gaming. In any case, I will be testing the chiplet based Navi 31 to find out...
I don't think so. As Bulldozer literally *regressed* in terms of IPC while RDNA3 did not.
That chip was so bad, AMD got sued for false advertising, pushed to the brink of bankruptcy.
I think RDNA3 being a Bulldozer moment is a bit harsh. We still have to see reviews for the 7800XT, which may end up being the best of the lot. Yeah, performance is meager compared to last gen, but it's servicable. To be a Bulldozer moment, Nvidia would have to be making better performance per dollar GPUs-but they're not. AMD is leading the price to performance race in rasterisation workloads. That is not what Bulldozer was.
Bulldozer almost caused AMD to go bankrupt, so while I get your point about a disappointing product, I think it might be a somewhat exaggerated comparison. AMD are doing well in CPU and server/enterprise, we will see if they can do good in "AI"-datacenters. Not going bankrupt because of RDNA3, CPU is where they make money.
RDNA3 is a bit disappointing, and if the rumors about no high-end RDNA4 are true that is also disappointing - and a bit surprising to be honest.
If chiplets are sorted out, massive gains could possibly happen, and they could possibly start taking marketshare and revenue from Nvidia.
IMO, Intel doesn't seem to be able to offer anything higher than mid/low-range in discrete GPUs, even though they said they'd compete at Enthusiast levels. Integrated is good though, AMD will have to compete fiercely there.
Well well, only time will tell :) Nice video again.
Thank you. I only addressed it from a gamer perspective and not a company or investor perspective. RDNA3 is a bump in the road to them financially in the short-term.
Your 7600 is not a "complete" RDNA3: it misses a lot of hardware features.
Without going to the 7900 serie I guess the definitive proof or disproof, of what you are implying, will be next month with the 7700 and 7800 serie compared to thir RDNA2 equivalents.
Not even close to Bulldozer.
7600XT uses the 6nm node which isn't a node advantage over the 6600XT. Also the 7600XT doesn't use the full RDNA3 tech. It should be expected to be roughly the same performance with a few uplifts from the architecture, but suffer from lack of cache comparison, specifically in lower resolutions. That tracks roughly what you should expect. It does that while increasing efficiency by around 10-40% compared to the 6600xt/6650xt depending on application. It achieves that with a 20% increase in transistor count. Overall that is within expectations for the die size and transistor count.
The rest of the RDNA3 uses a chiplet architecture which will give worse perf/watt and therefore performance for any transistor equivalent of a monolithic chip. The benefit of this is that they can go above the transistor equivalent of a monolithic chip while maintaining yields. Thus, the benefit of a chiplet architecture for performance isn't achieved with RDNA3 as the largest chip they've used is in the 7900xt which is only ~300mm^2. Instead they used it to cut the cost of the chip. The 7900xt being under the cost of the 4080 while maintaining a performance advantage to it consistently. The perf/watt is lower than it, but you would expect that again, because having to communicate further distances and run logic in 6nm chips of course gives you worse perf/watt.
Bulldozer on the other hand, had no redeeming qualities. It lost on all fronts, badly. It lost to chips smaller than it badly. It lost to the jaguar architecture.
And even though Bulldozer had no redeeming qualities, it was cheaper, and it was the longest running CPU that I know of...its difficult for me to understand how that thing was still on shelves and available for purchase in Jan 2019. How did that ever happen?
@@ImaMac-PC incredibly poor execution from Global Foundry and AMD had a contract with them where they were required to buy so many chips at x capacity. Global Foundry wasn't able to make anything better than that until they licensed someone's lower node. Forget who off the top of my head.
End result, tons of fx chips that had to move or AMD lost even more money. Really sad situation.
something went badly wrong with RDNA3 prior to release. The 7900xt and xtx are decent increases over the 6950xt if not huge increases, but are simply overpriced due to Nvidia's setting the example for greediness...
Sadly we have to wait for RDNA5 to get something meaningful again... at least your 7900XT will be the second fastest AMD card for years to come.
And when the 7900 XT is on sale, not a bad value!
I would say it's even worse than Bulldozer. Bulldozer was super super cheap at least. I bought for 100$ FX 8350 and I had moderate CPU until 2018 when I switched to R7 2700. Right now GPUs are extremely over priced. For example RX 7600 price (269$) is a joke and barely no upgrade comparing to RX 6600 or RX 6600XT. Same goes with other GPUs. Not saying that my current GPU (RX 6700XT) can still compete with RTX 4060Ti/RTX 4060 and probably will compete against RX 7700 if AMD decide to release the new GPU.
It’s a stretch to compare the RDNA3 to Bulldozer seriously Bulldozer almost killed AMD, RDNA3 is actually selling pretty well, and it offers better raster than directly price competing Nvidia products, it’s best to forget names and compare performance at given price
Do you have a source for the data that allows you to conclude that RDNA3 is selling pretty well? I ask because I don't see that in watching the daily sales at Amazon/Newegg/Microcenter.
9:28 Sure! There's a decent chance Radeon could over take Nvidia this decade. You know, if Nvidia is pretty quick transitioning their entire company over to AI.
Yes, then they would just have to be better than Intel!
Not at all! Go AMD!
I hate to say it but i dont see Radeon ever making a true comeback. Our best hope is Intel or Nvidia just deciding to basically move out of the consumer GPU space.
They tried to bullshit us by selling relabelled models, thinking nobody would notice.
I was going to buy the 7800xt but that was the "7900xt". The piece of shit they want me to pay 550 euro for now... No chance!
I'm either holding for intel , or switching to Nvidia.
Now imagine, the 4080 is in fact a 70 class card. So practicly amd is only able to compete with mid range. Their 7900xtx best case is only a 7800 non xt with the current performance. Not only that but doing it meanwhile eating a ton more power. Last time they was lucky nvidia stucked on samsung 8nm. Now they are on the same node. Aaaaaand yeah nvidia is far more advanced regarding the uarch. And I am an amd supporter. Sad. I cannot even imagine what would they have done if nvidia releasing the 4080 as the 4070ti for 600 usd. I guess they would have cancelled the whole rdna3 lineup. It even would have been better just to port rdna2 to 5nm.
That port of RDNA2 to 5nm would have been interesting...
What people should stop is comparing names of GPUs, but simple price to performance. For example what will 500 euros or dolars get you. In my case, im looking for 1080p card, so lets say in Ireland Rx 6650 about 260e, Rtx 3060 300e or RTX 4060 350euros... amd for me is clear winner there, That said Rx 7600 cheapest in ireland is atm is 300e. So for me 6650 is winner. On use market that im looking, there is rx 6800xt for 430 euros asking price in Cork, while RTX 3070 is about same while being weaker card.. So AMD is still good value compared to Nvidia. About RDNA3 being faileir well same can be said for RTX 4060 or 4060TI.. they are even worse
While RDNA3 has been a disappointment because it under-delivered both in performance and efficiency, I won't go so far as to equate it to Bulldozer. RDNA3 GPUs perform fine, its their price that is not attractive for what they offer. Imagine 7900XTX at USD799, 7900XT at USD599 and the upcoming 7800XT 16GB at USD449, and they each will bring quite a lot of excitement to the currently stale GPU market. Edit: 7600 should be USD229.
Lisa su : its because jensen is my uncle
RNDA3 in Navi 33 is diffierent from RDNA3 in Navi 32 and 31. Just so you are aware.
Something unexpected must have happend during développement because the performance jump is too weak. RDNA 2 on 5 would have been probably a better product.
I would love to have seen RDNA2 on 5nm... AND with a larger 384-bit memory bus!
For me it seems like AMD 7000 have a huge flaw in their architecture. Something that they can't fix properly and need to program complex workaround in drivers. This maybe explains late releases of optimization drivers, and (for example) Forza issues.
I don't know if it's a flaw or if it's just different. Will keep on testing...
Is Nvidia leveraging it's AI technology to race ahead of AMD? And what happens if intel leverages their quantum computers to advance their GPUs?
This is simply Lisa Su cancelling the big gpu product to save her cousins company Nvidia, we all know they had a bigger gpu and cancelled it
Sure look like it
rdna3 is a waste of sand. they've doubled the compute power but only got a few percent uplift. disaster.
The entire point of RDBA3 from an architectural standpoint was to gain similar or higher performance while moving to a chiplet architecture. It is a foundational architecture before making the next leap in performance. They couldn't do both at the same time given that this has never been done before. Sure, if RDNA3 was monolithic then this conversation would be non-existent.
That's what I used to think. However, Navi 33 is monolithic and RDNA3 which removes the "it's all the chiplets fault" argument. It's the architecture.
sad thing is amd dont need to win.. they make cpus and apus
all the need to do is "keep it going".. you have seen the insane prices, the only company that can at least do any dent to team green is intel.. they have the cap x to do it amd dont.
You may be right about Intel.
When the nvidia stocks are ready to move and liquidated they will pump AMD then and only then release the nvidia killer gpu
First
RDeadOnArrival 3's launch is actually smoother than Bulldozer or RDNA 1. At least XTX's Raster performance is solid and driver is usable. For Bulldozer I can say no gamer should ever purchase it simply because of its weak single thread performance. And even tho RDNA 1 has competitive performance against mid-range Turing, it certainly took at least half a year for AMD Copeon Terrible Group to iron out many breaking issues for its driver.
On the other hand tho, RDNA 2 exceeded expectation and AMD probably got cocky since then and RDeadOnArrival 3 marks AMD returning to its true form. Nvidia still trashes AMD in terms of power efficiency and feature sets. All I can say for this generation of AMD Copeon GPUs is: RIP AMD🎉
Amd is dead again in gpu.
You can indeed compare it with the Bulldozer or GCN.
GCN only did good one gen on the HD7970.
R9 290 was ok, because Nvidia already was performing the same.
R9 390 Amd was dead.
Looking at the past, RDNA or Navi did the same as the GCN.
Only performed for 2 generations and then it's dead.
Graphic Core Next, what a joke.
Now RDNA does exactly the same.
Maybe jump to Intel next time.
But Intel is also very expensive.
In a few years, there are no cards below 300 anymore.
Just as no cards below 200 now.
gtx1050ti was originally a 150 dollar card.
Now that's entry level. To bad those times are over.
I wish ATI coming back.
Was very bad for us consumers that Amd bought ATI.
Amd wanted to much and also want to enter gpu market.
Now they've killed it.
AMD is finished in desktop discrete graphics. they will bow out of the market by 2025. AMD killed off ATI Radeon.
They had momentum with RDNA1 and going into RDNA2...but that's gone.
now the lesser RDNA3 cards are pure crap, and should only be sold at reduced prices. 7800xt will be a joke, 10% less power and 20% less cost for the same performance as a 6800xt. Simply replacing the prior generation without moving up in peformance at all. That is also following Nvidia's example much too closely.
I thought RDNA 3 was supposed to support frame generation like NVidia but recent leaks say FSR 3.0 will only work on RDNA 4. So I guess AMD is still one or two generations between NVidia. AMD's AI is already a gen behind and apparently NVidia has massive performance gains in this area recently announced. So I guess AMD is not innovating they are just following NVidia so will always be behind.
Radeon is a joke on top of a joke. That's what I think.
Comparing GPUs to CPUs not fair, difference is GPU drivers are very hard, for CPUs entire tech industry works on optimizing it i.e open source gcc compiler used even for proprietary software, there is no such thing for GPUs, radeon has always lagged in software dept. due to smaller teams, but RDNA 3 is still not even close to bulldozer..
AMD is a console company. They dont bring anything valuable to PC gaming.