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

 

[Joe NYC]

I know it's bad, but is it THIS BAD? Really? Is Radeon deader than Xbox?

Nvidia dominates gaming GPU market with 95 percent share as sales of AMD Radeon graphics plummet to a historical low of 5 percent

As the market of graphics cards rises year-over-year.

www.tomshardware.com

View attachment 139500

It doesn't correspond to AMD graphics revenue.

I wonder what kind of "research" Jon Paddie Research is conducting.

 

[adroc_thurston]

It doesn't correspond to AMD graphics revenue.

I wonder what kind of "research" Jon Paddie Research is conducting.

Yeah it does lmao.

 

[Joe NYC]

Yeah it does lmao.

I guess if you classify NVidia dGPU sales as "Gaming dGPU", then yeah, but even NVidia started to come clean and is calling it "Gaming and AI".

So. the low AMD percentage may have to do more with NVidia AI bubble than with AMD selling less.

 

[adroc_thurston]

I guess if you classify NVidia dGPU sales as "Gaming dGPU", then yeah, but even NVidia started to come clean and is calling it "Gaming and AI".

No.

So. the low AMD percentage may have to do more with NVidia AI bubble than with AMD selling less.

AMD does sell less lmao.
People ain't gonna buy Radeon is a comparable geforce is only $100 away.

 

[Joe NYC]

No.

It's right in the NVidia earnings report: "Gaming and AI PC".

AMD does sell less lmao.
People ain't gonna buy Radeon is a comparable geforce is only $100 away.

AMD gaming revenue is up from very low numbers in 2024:
922
648
462
563

2025:
647
1122
1298
843

JP Research shows 2025 much lower than 2024.
The only way to explain (other than the most likely explanation of Jon Paddie being on crack) is NVidia AI revenue from gaming dGPU sales counting in "Gaming and AI PC"

 

[adroc_thurston]

It's right in the NVidia earnings report: "Gaming and AI PC".

Yes, gaming and laptops with Copilot+ certs.
They're called AI PCs you know. marketingstuffs.

JP Research shows 2025 much lower than 2024.
The only way to explain (other than the most likely explanation of Jon Paddie being on crack) is NVidia AI revenue from gaming dGPU sales counting in "Gaming and AI PC"

No, it's just that AMD units are way down while ASPs and margins are way up.
None of the RDNA3 parts were selling above $300.
I don't even think N33 was selling anything either. oh well.

 

[jpiniero]

It means exactly that.

No. Memory ASP hikes will cause channel ASP hikes. AMD margin stays constant and pretty fat.

AMD still sells the AIBs the chip and memory as a bundle though, no? It might be showing up as more revenue but AMD is just passing the extra cost through to the AIBs.

At 5% they obviously aren't shipping much.

 

[adroc_thurston]

AMD still sells the AIBs the chip and memory as a bundle though, no?

Chip yes, memory not anymore.

It might be showing up as more revenue but AMD is just passing the extra cost through to the AIBs.

Means their margin is just as fat as it was on Navi48 launch.

At 5% they obviously aren't shipping much.

They have zero incentive to ship more.
5% is the natural mss for Radeon at acceptable margins.

 

[Panino Manino]

Incredible "copium".
It's like burning the brand to the last of its ashes and becoming insignificant was the grand objective all along.
Praise the Su, she did it.
Finally Radeon occupies the place it deserves.

 

[coercitiv]

5% is the natural mss for Radeon at acceptable margins.

Radeon 5%, perfectly balanced.

I’m looking at scale, and AMD is in a different place right now. We have this debate quite a bit at AMD, right? So the question I ask is, the PlayStation 5, do you think that’s hurting us? It’s $499. So, I ask, is it fun to go King of the Hill? Again, I’m looking for scale. Because when we get scale, then I bring developers with us.

So, my number one priority right now is to build scale, to get us to 40 to 50 percent of the market faster. Do I want to go after 10% of the TAM or 80%? I’m an 80% kind of guy because I don’t want AMD to be the company that only people who can afford Porsches and Ferraris can buy. We want to build gaming systems for millions of users.

One day, we may. But my priority right now is to build scale for AMD. Because without scale right now, I can’t get the developers. If I tell developers, ‘I’m just going for 10 percent of the market share,’ they just say, ‘Jack, I wish you well, but we have to go with Nvidia.’ So, I have to show them a plan that says, ‘Hey, we can get to 40% market share with this strategy.’ Then they say, I’m with you now, Jack. Now I’ll optimize on AMD.’

AMD’s Jack Huynh

 

[adroc_thurston]

Finally Radeon occupies the place it deserves.

Long overdue, yes.
Shipping loss leaders to maintain mss was always pure madness.

Radeon 5%, perfectly balanced.

They can go down to 2%, actually.
Just need another round of ASP hikes.

 

[gdansk]

It's not about making a card that can chew through gnarly D3D anymore. There's no room for anyone but Nvidia in post-truth graphics.

Anything AMD or Intel release for non-integrated, non-SR IOV markets is absolutely charity going forward. If they release them, they better be priced knowing they'll sell nothing.

 

[Saylick]

They can go down to 2%, actually.
Just need another round of ASP hikes.

I think 0% is possible too if ASPs go up to infinity. *taps temple*

 

[adroc_thurston]

I think 0% is possible too if ASPs go up to infinity. *taps temple*

They'll probably get to 0-1% late in the RDNA5 cycle.

 

[luro]

They'll probably get to 0-1% late in the RDNA5 cycle.

Better quit at this point.

I honestly don't understand this strategy. 30% margin off 100 units @$700 wouldn't be better than 50% margin off 40 units @$900 ?

 

[gdansk]

I honestly don't understand this strategy. 30% margin off 100 units @$700 wouldn't be better than 50% margin off 40 units @$900 ?

The strategy is actually optimal. Only people who generally prefer Radeon (rationally this is perhaps only Linux users) will buy Radeon. There is no growing the number of units shipped beyond this subset. Reducing the ASP simply reduces the amount of money they get from this group. Everyone else uses the mere existence of the Radeon card to buy a slightly cheaper Nvidia card.

 

[luro]

The strategy is actually optimal. Only people who generally prefer Radeon (rationally this is perhaps only Linux users) will buy Radeon. There is no growing the number of units shipped beyond this subset. Reducing the ASP simply reduces the amount of money they get from this group. Everyone else simply uses the mere existence of the Radeon card to buy a slightly cheaper Nvidia card.

Fair

 

[adroc_thurston]

Better quit at this point.

They did!
RDNA5 already carries zero parts aimed at the AIC DT market.

I honestly don't understand this strategy. 30% margin off 100 units @$700 wouldn't be better than 50% margin off 40 units @$900 ?

NV chops prices and AMD sells 40 units for $700 instead.
They can't win, the market was closed off for Radeon after the HD4870 suicide.

 

[coercitiv]

Anything AMD or Intel release for non-integrated, non-SR IOV markets is absolutely charity going forward.

Good luck to AMD and Intel when Nvidia starts pushing with their own iGPUs on the ARM side. (this is assuming that the Intel - Nvidia iGPU deal was just for show, otherwise good night and good luck AMD)

I never took QC seriously as a short term contender, but post-truth green graphics can definitely chew through the x86 moat on the non-business consumer side.

But hey, AMD has data-center AI in the bag, I'm sure they'll be fine.

 

[adroc_thurston]

Good luck to AMD and Intel when Nvidia starts pushing with their own iGPUs on the ARM side.

WoA sucks and Tegras suck even more.
Serpent Lake will be viable product, though.

I never took QC seriously as a short term contender, but post-truth green graphics can definitely chew through the x86 moat on the non-business consumer side.

NV is utterly incompetent at building SoCs.

 

[marees]

Good luck to AMD and Intel when Nvidia starts pushing with their own iGPUs on the ARM side. (this is assuming that the Intel - Nvidia iGPU deal was just for show, otherwise good night and good luck AMD)

I never took QC seriously as a short term contender, but post-truth green graphics can definitely chew through the x86 moat on the non-business consumer side.

But hey, AMD has data-center AI in the bag, I'm sure they'll be fine.

real competition will come if nv can fab a gpu chiplet in Intel fab

until then the economics are same for everyone

 

[adroc_thurston]

real competition will come if nv can fab a gpu chiplet in Intel fab

That just means fatter margin for NV.

 

[jpiniero]

real competition will come if nv can fab a gpu chiplet in Intel fab

I still think there's a chance nVidia goes back to Samsung for low end products.

 

[adroc_thurston]

I still think there's a chance nVidia goes back to Samsung for low end products.

To pocket the margin, yes.
They're competing with themselves.

 

[madtronik]

They'll probably get to 0-1% late in the RDNA5 cycle.

No problem for me, for a long time I have hoped that I could get rid of my bulky and power hungry GPU and use a big fat APU instead.