Discussion RDNA4 + CDNA3 Architectures Thread

[DisEnchantment]

With the GFX940 patches in full swing since first week of March, it is looking like MI300 is not far in the distant future!
Usually AMD takes around 3Qs to get the support in LLVM and amdgpu. Lately, since RDNA2 the window they push to add support for new devices is much reduced to prevent leaks.
But looking at the flurry of code in LLVM, it is a lot of commits. Maybe because US Govt is starting to prepare the SW environment for El Capitan (Maybe to avoid slow bring up situation like Frontier for example)

See here for the GFX940 specific commits

History for llvm/lib/Target/AMDGPU - llvm/llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies. - History for llvm/lib/Target/AMDGPU - llvm/llvm-project

github.com

Or Phoronix

More AMD "GFX940" Enablement Work Landing In LLVM - Phoronix

www.phoronix.com

There is a lot more if you know whom to follow in LLVM review chains (before getting merged to github), but I am not going to link AMD employees.

I am starting to think MI300 will launch around the same time like Hopper probably only a couple of months later!
Although I believe Hopper had problems not having a host CPU capable of doing PCIe 5 in the very near future therefore it might have gotten pushed back a bit until SPR and Genoa arrives later in 2022.
If PVC slips again I believe MI300 could launch before it

This is nuts, MI100/200/300 cadence is impressive.



Previous thread on CDNA2 and RDNA3 here

Question - Speculation: RDNA3 + CDNA2 Architectures Thread

Man I have been dying to make this one for a while now. First rumours for RDNA3 are here so new thread time! Just going to start off with this one for now: kopite7kimi on Twitter: "@VideoCardz Ah, I mean a simple mcm design with 10240 cores is not enough. Because the lift from RDNA2 to RDNA3...

forums.anandtech.com

 

[Win2012R2]

Gross profit or net income?

Both

Saying 270mm2 die means nothing... because the die itself is barely +40-50 dollars MAX on N4.

For TSMC it might be. If only they got into business of selling GPUs...

 

[ajsdkflsdjfio]

Both

For TSMC it might be. If only they got into business of selling GPUs...

No, for the customer it is. I calculated this different using the wafer prices TSMC charges the customer, not the prices TSMC pays to produce these wafers.

Also what do you mean by both? Intel literally made 43 bil gross profit versus a 79 bil revenue in 2021. This includes not only production costs but marketing and sales costs. Now think about what the gross profit is on the products themselves minus sales/marketing. Greater than 50% if not 60%. 60% of 600 is equal to 360$ so.... Not entirely unfeasible if you use your brain.

 

[Win2012R2]

No, for the customer it is.

No, it isn't.

Intel literally made 43 bil gross profit versus a 79 bil revenue in 2021. This includes not only production costs but marketing and sales costs.

Marketing costs are not included into calculation of gross profit, neither is R&D and a whole load of other very expensive stuff.

 

[ajsdkflsdjfio]

Marketing costs are not included into calculation of gross profit, neither is R&D and a whole load of other very expensive stuff.

Why are you talking about R&D and other associated costs when I specifically said I wasn't factoring them in. Marketing costs are indeed not part of gross profit, but regardless Intel/AMD has around a 50% gross profit depending on the year, so once again it is not unfeasible for a 270mm2 card to have 200-250$ gross profit at 600 dollars. Now in this specific scenario while using 7600 parts it can be as high as 350-400$.

If a rx 7600 is making 30-40 dollars gross profit at a 250$ sale price, the same card but with a 270mm2 die on N4 (+35 dollars) selling at 600 dollars would be netting a 350$ gross profit.

My argument was always that the B580 was not a negative gross profit at 250$. Kepler seems to refute this idea by ridiculing a 350$ gross profit for a 270mm2 die card at 600$, which is ludicrous if you are only talking about die cost alone.

 

[Win2012R2]

Why are you talking about R&D and other associated costs when I specifically said I wasn't factoring them in

Because you said this -

This includes not only production costs but marketing and sales costs

it is not unfeasible for a 600 dollar card to have 300-400 gross profit

Maybe for some other card it isn't but we are talking here specifically current AMD consumer GPUs sized at 270 sq mm on N4, they don't make that much money, we know that for certain from their stock market reports, for AMD GPU business is a dog with fleas.

 

[ajsdkflsdjfio]

Because you said this -

I was wrong about marketing, but I never said anything about R&D and other associated costs.
As for sales costs I was talking about things like shipping, which are in fact included in gross-profit calculations.

Maybe for some other card it isn't but we are talking here specifically current AMD consumer GPUs sized at 270 sq mm on N4, they don't make that much money, we know that for certain from their stock market reports, for AMD GPU business is a dog with fleas.

Yes I agree, a typical Navi 48 card would not be making 300-400$ gross profit, likely much less due to higher costs of the card overall and higher AIB margins on mid-high range cards.

But this is not due to the 270mm2 die itself as its only 35-45 dollars more than the die on the rx 7600 priced at 250$. If all you changed was the die, a 270mm2 die on N4 using rx 7600 parts selling for 600$ would absolutely be making 300-400$ gross profit.

I say all this because Kepler was ridiculing people who said that the b580 was likely not negative gross on a 270mm2 die at 250$. He tries to criticize this idea by implying that using this same logic, a Navi 48 card at 600 dollars would be making 400$ gross profit because they both use 270mm2 dies.

Evil AMD making $400 of profit per card according to some people here

Firstly even with an apples to apples comparison it would be 350$ not 400$. Factor in costs like more ram, higher speed ram, N5 vs N4 costs, lower parametric yield due to denser design, larger memory bus width, a higher end cooler/heatsink config, more expensive PCB for a more power hungry chip, higher AIB margins for higher-end parts and in general more established AIB partners. With all these things factored in you get a gross profit for AMD at closer to 200$ than 400$, which is entirely reasonable. And this is using the same logic that shows that a b580 is likely not negative gross profit wise.

 

[blckgrffn]

Seriously, this is tiring to watch you spar on this. Points are made.

For me, it’s just what we are going to get. Nvidia has made it clear it’s going to lift ASPs and since it’s worked AMD will just follow as best they can.

If it’s a 200W 7900xt equivalent it’ll be a huge upgrade from my 6800 and it’s still happening for me. The RT stuff isn’t that interesting to me but if it’s got 7900XTX+ performance on that front all the better.

Those are my expectations, and if it’s right in the ball park of the 4070 Ti Super performance wise but at a lower power level… I am not seeing the issue. Thats a big uplift from where we are at now.

 

[SolidQ]

FSR4

 

[Kepler_L2]

FSR4

Where is that from?

 

[SolidQ]

Where is that from?

Don't know, ask on gaf

 

[gaav87]

That would be worse than what I imagined even in a realistic scenario. I mean the performance was clear will not be great simply from the die size. that isn't really a surprise. but $599 is just terrible. that's not even mainstream in any way or fashion. And since the die is small and yields high, AMD would what as many on the full chip as possible so the cut version will likley be the same stupid thing as with the 7700 vs 7800.

GPUs don't really age. that pricing would be in competition with 5070 and used 4070 ti supers and especially vs the later not having vra, advantage. given the pro nvidia bias, that won't end well. if this really happens, AMD should just close the GPU division. not even trying is pointless. Either go for the grown or market share but just selling small dies at insane prices, why? what is the strategy here? I don't see it.

Well sorry man but thats what i heard... Also remember 7900xt starting price ? Il remind u 899$... 1 month later lowered 3 months later lowered even more and now its 669-700$

 

[gaav87]

Seriously, this is tiring to watch you spar on this. Points are made.

For me, it’s just what we are going to get. Nvidia has made it clear it’s going to lift ASPs and since it’s worked AMD will just follow as best they can.

If it’s a 200W 7900xt equivalent it’ll be a huge upgrade from my 6800 and it’s still happening for me. The RT stuff isn’t that interesting to me but if it’s got 7900XTX+ performance on that front all the better.

Those are my expectations, and if it’s right in the ball park of the 4070 Ti Super performance wise but at a lower power level… I am not seeing the issue. Thats a big uplift from where we are at now.

I know ppl expected 4080 for 499$ but lets be realistic here... ~50Tflops of single precision...

 

[Win2012R2]

~50Tflops of single precision...

Dual issue, so in reality 25-ish...

 

[DaaQ]

This isn't the future I was sold in my childhood and I want a refund.

We're supposed to have flying cars and holodecks God damn it, not $600 270mm2 dies!

I know right, I remember in my young adult days, getting a high end Nvidia GPU made by VisionTek.

 

[Win2012R2]

8xxx ref?

Do they really need 3 fans for supposedly lower power card, it's not supposed to be like 600W monster.

 

[gdansk]

Do they really need 3 fans for supposedly lower power card, it's not supposed to be like 600W monster.

More fans = more quiet. Everyone wants a 3 fan graphics card these days.

 

[Win2012R2]

More fans = more quiet. Everyone wants a 3 fan graphics card these days.

Yeah, I like quiet too, prefer FE edition single fans, seem to work very well even for very hot scenarios, 3 fans too many things to go wrong and whine.

 

[blckgrffn]

Yeah, I like quiet too, prefer FE edition single fans, seem to work very well even for very hot scenarios, 3 fans too many things to go wrong and whine.

🤷‍♂️

That has not been my experience… and in the one occasion a fan when there was three and one bearing went out I unplugged that fan and… it was fine. I am good with 3 fan, I am hoping the $599 nets me some ARGB bling too!

 

[SolidQ]

Do they really need 3 fans

i've noticed also on RX7600 temps is much lower on 3fans, than 2

 

[branch_suggestion]

没有9800XT - 电脑讨论(新) - Chiphell - 分享与交流用户体验

没有9800XT,是9开头,但第二位是0,然后第三位是7,以上~,电脑讨论(新),讨论区-生活与技术的讨论 ,Chiphell - 分享与交流用户体验

www.chiphell.com

 

[gdansk]

RX 9070? 4000 ahead of Nvidia now.
Take that Jensen.

I don't think it's a bad naming change. It avoids number overlap with their CPUs.

 

[Kepler_L2]

没有9800XT - 电脑讨论(新) - Chiphell - 分享与交流用户体验

没有9800XT,是9开头,但第二位是0,然后第三位是7,以上~,电脑讨论(新),讨论区-生活与技术的讨论 ,Chiphell - 分享与交流用户体验

www.chiphell.com

70-class branding gives me some hopium for price.

 

[branch_suggestion]

RX 9070? 4000 ahead of Nvidia now.
Take that Jensen.

I don't think it's a bad naming change. It avoids number overlap with their CPUs.

True, but they could've avoided blatantly one upping NV, then again they do the same with motherboards.
Also unless they go to 10k+ branding, it looks like one and done branding.

70-class branding gives me some hopium for price.

Indeed, that is hopefully good news that there is a proper generational leap relative to the name.
7800XT has roughly the same raster perf as the 6800XT for $150 less.
That is dumb naming, either you do a big jump in performance for the same or more money or you name it a class lower for the same perf and less money.
Good thing here is with 7 class naming you can compare to the 7700XT, a big leap in performance at a modestly higher price for the same tier is easy marketing.
Of course RDNA3 naming was intended with 20% more perf but they could've changed names last minute.

 

[branch_suggestion]

I just hope that by the time Zen6 and UDNA are out there is a cohesive branding scheme across the whole portfolio.
Consumers need branding that sets an impression for what is to come and is memorable.

 

[poke01]

I wouldn't touch RDNA4 if your on RDNA3 or Ada. UDNA is where the juicy stuff is and considering PS6 will be based of that it will age much better I think.

Think RDNA1 vs RDNA2. RDNA2 aged much better.


BUT I'm looking forward to the reviews