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

 

[branch_suggestion]

5070Ti
View attachment 120721

9070XT
View attachment 120722

Its easy, I just call it like it is. The 9070XT Tai Chi is averaging 364W per TPU in gaming, while losing in almost everything to the Galax 5070 Ti, which averages 279W in gaming per TPU. 9070XT is not a bad GPU, but already being pushed out of its perf/power curve. A 400W XTX version which draws 430W is kind of ridiculous, IMO.

You did look at the V-Sync numbers, right?
It is efficient, you just choose whether you want to access the efficiency through various means.

 

[Josh128]

You did look at the V-Sync numbers, right?
It is efficient, you just choose whether you want to access the efficiency through various means.

Of course. I already posted specifically about that a few pages back. What relevance does that have to my previous post, other than re-affirming it?

 

[gdansk]

Of course. I already posted specifically about that a few pages back. What relevance does that have to my previous post, other than re-affirming it?

But it shows lower power usage in vsync. So it's a bit of mixed message.

 

[Josh128]

But it shows lower power usage in vsync. So it's a bit of mixed message.

It shouldnt be. Vsync 60Hz has no bearing on published TBP and clocks while gaming at uncapped frame rates. A non XT, an XT, and an XTX would all draw similar power at Vsync 60Hz, while they wont be anywhere close to each other when you uncap the frame rate.

 

[GTracing]

But it shows lower power usage in vsync. So it's a bit of mixed message.

Vsync reduces the power consumption because it's a lighter load. The fact that the 9070 XT sees a greater reduction than the 5070 ti shows that the 9070 XT is pushed further into diminishing returns on the voltage/frequency curve.

 

[Josh128]

Vsync reduces the power consumption because it's a lighter load. The fact that the 9070 XT sees a greater reduction than the 5070 ti shows that the 9070 XT is pushed further into diminishing returns on the voltage/frequency curve.

Exactly.

 

[gaav87]

5070Ti
View attachment 120721

9070XT
View attachment 120722

Its easy, I just call it like it is. The 9070XT Tai Chi is averaging 364W per TPU in gaming, while losing in almost everything to the Galax 5070 Ti, which averages 279W in gaming per TPU. 9070XT is not a bad GPU, but already being pushed out of its perf/power curve. A 400W XTX version which draws 430W would be kind of ridiculous, IMO.

RT games have increased power draw by alot compared to light or no rt games.
Some games im beating 4080S by few fps with lower power draw at 220-240W
Some games it chugs power like crazy RT heavy games.

 

[marees]

A million 9070XTs would have made AMD like say $700 mln - why can't they get that for launch when Apple sells close to 1 mln iPhones per day and tens of millions on launch?

Because apple's primary product is a iphone

AMD is primarily a data centre company
It is not clear if they want to sell DGPUs to gamers (at the expense of data centre chips)

 

[adroc_thurston]

A million 9070XTs would have made AMD like say $700 mln - why can't they get that for launch when Apple sells close to 1 mln iPhones per day and tens of millions on launch?

you somehow have net negative understanding on how supply chains work.

Because apple's primary product is a iphone

No, Apple primary product is their supply chain.

 

[Win2012R2]

you somehow have net negative understanding on how supply chains work.

Here is what we know:
1) 9 Sep 2024 - Jack Huynh, AMD's senior vice president and general manager of the Computing and Graphics Business Group -""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 [Total Addressable Market] 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."" - https://www.techspot.com/news/104633-amd-confirms-wont-prioritize-competing-nvidia-top-gaming.html
2) "PC discrete GPUs reached 8.1 million units in Q3 2024, down from 9.5 million units in Q2 2024, NVIDIA now controls 90% of the market share" - https://wccftech.com/nvidia-aib-market-share-grow-90-percent-total-gpu-shipments-decline-q3-2024/

Now please explain to me how "40-50 percent of the market" aim reconciles with reported 200k sales with the market being at least 2 mln+ gGPUs per month - AMD could not plan ahead when they decided they cancel top end card 2 years ago? They clearly ordered an order of magnitude less than their claimed market ambitions, so they are either "have net negative understanding on how supply chain works" or something else.

Given delayed launch AMD should have had at least 1 mln cards ready for sale.

 

[Tuna-Fish]

Isn't 60-series most of the market? I think 200k upper midrange GPUs is not little.

 

[Win2012R2]

Isn't 60-series most of the market?

Most certainly, so why are they bothering releasing 70 series if it is not?

I guess having a HALO SKU helps selling others...

 

[adroc_thurston]

Now please explain to me how "40-50 percent of the market" aim reconciles with reported 200k sales with the market being at least 2 mln+ gGPUs per month

that's called having a long-term goal.
AMD is kinda about those.

 

[SolidQ]

I guess having a HALO SKU helps selling others...

for AMD, need few things still fix. PT perfomance/good support software, especially Blender, AMD is ignore this problem, while NVIDIA 3-7 times faster in rendering.

 

[Win2012R2]

that's called having a long-term goal.

Ok: 200k units that were allegedly sold after at least 2 months of production, most likely including units made before January - that's like what: <4% market share of quarterly dGPU sales, and that's on delayed launch? If somebody got negative understanding of how supply chains work that's AMD - and they got people paid to do that well.

especially Blender

Blender is used maybe by 0.1% users of GPUs, and most of them would prefer Nvidia - it's completely irrelevant for a gaming GPU.

9070 XT is a very good card, they just should have made at least a million for launch - that's about 10% supposed market share of quarterly dGPUs.

 

[adroc_thurston]

that's about 10% supposed market share of quarterly dGPUs.

Not AIC share dawg.
dGFX MSS also includes mobile offerings.

 

[Win2012R2]

Not AIC share dawg. dGFX MSS also includes mobile offerings.

These quarterly dGPU sales refer to desktop AiB -

https://www.reddit.com/r/hardware/comments/1d9jy6b/desktop_dgpu_chip_market_shares_q12024_amd_12

So we have 6-9 mln cards per quarter.

 

[branch_suggestion]

These quarterly dGPU sales refer to desktop AiB -

https://www.reddit.com/r/hardware/comments/1d9jy6b/desktop_dgpu_chip_market_shares_q12024_amd_12

So we have 6-9 mln cards per quarter.

And that includes prebuilts and SIs which NV has well over 90% share in.
DIY is AMD's strongest market, but numbers specifically for DIY are pretty much restricted to individual retailers.

 

[Win2012R2]

And that includes prebuilts and SIs which NV has well over 90% share in.

Well that's (big) part of TAM where AMD has officially stated they want to reach 40-50% market share with more affordable product (hence no halo SKU).

Does anyone see much activity in prebuilds with 9070 XT flying off the shelves? Does not look that way.

 

[branch_suggestion]

Well that's (big) part of TAM where AMD has officially stated they want to reach 40-50% market share with more affordable product (hence no halo SKU).

Does anyone see much activity in prebuilds with 9070 XT flying off the shelves? Does not look that way.

The power of the green sticker and entrenched mindshare with less informed buyers makes that a very slow burn relying on consistent execution.
Intel had the same thing going forever in the same markets, it has always been AMD's kryptonite.
Ryzen, especially X3D parts are starting to build a moat of mindshare, now Radeon just needs the gall to make an X3D GPU.

 

[Win2012R2]

To win market share AMD must make enough cards to satisfy demand - especially for the launch period, but that's clearly has not happened, if 200k units sold is true (seems plausible given known stock figures at a bunch of retailers) then they clearly failed to make anywhere near enough to grab even tiny market share, despite their claims that they want to get to 40-50%, they won't if they got no stuff to sell.

 

[PJVol]

I couldn't properly test it as I've failed to replace pptable, it just locks up if I replace it with upp, or plainly refuses to get replaced via plain copy

They (amdgpu devs) broke it. Is there a smu reset failure in syslog? My Neon distro based on 6.11 so no luck for me ) Supposedly fixed in mainline kernels, and again, the issue is gfxoff (who would have thought? IIRC this has been pain in the ass for devs for about 5 years? )

 

[beginner99]

To win market share AMD must make enough cards to satisfy demand - especially for the launch period, but that's clearly has not happened, if 200k units sold is true (seems plausible given known stock figures at a bunch of retailers) then they clearly failed to make anywhere near enough to grab even tiny market share, despite their claims that they want to get to 40-50%, they won't if they got no stuff to sell.

Yeah volume was certainly too low. I'm debating if I should get a 9070 now which here seem to be sometimes in stock but not at great price or simply wait 2 months and get a 9070 xt probably for that same price I can get a 9070 now and be able to choose which model not just the one that happens to be in stock.

 

[Win2012R2]

if I should get a 9070 now

Nah, wait for proper 9070 XT! It's a no brainer - worth waiting for.

 

[Jan Olšan]

Here is what we know:
1) 9 Sep 2024 - Jack Huynh, AMD's senior vice president and general manager of the Computing and Graphics Business Group -""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 [Total Addressable Market] 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."" - https://www.techspot.com/news/104633-amd-confirms-wont-prioritize-competing-nvidia-top-gaming.html
2) "PC discrete GPUs reached 8.1 million units in Q3 2024, down from 9.5 million units in Q2 2024, NVIDIA now controls 90% of the market share" - https://wccftech.com/nvidia-aib-market-share-grow-90-percent-total-gpu-shipments-decline-q3-2024/

Now please explain to me how "40-50 percent of the market" aim reconciles with reported 200k sales with the market being at least 2 mln+ gGPUs per month - AMD could not plan ahead when they decided they cancel top end card 2 years ago? They clearly ordered an order of magnitude less than their claimed market ambitions, so they are either "have net negative understanding on how supply chain works" or something else.

Given delayed launch AMD should have had at least 1 mln cards ready for sale.

The 200 000 number was some sort of reporting mistake, AMD said they didn't give that figure. Not sure if BenchLife got it off record (or from another source), but they deleted the article that said that. I would basically forget the number ever made news, probably is way off.