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

 

[adroc_thurston]

I wonder if we at least get an annotated die shot

They always reserve this for ISSCC.

That seems to be the same ballpark as Genoa.

A wee bit more but yeah.

 

[TESKATLIPOKA]

16-24 Gb 32gbps GDDR7 is coming at the end of 2024.

It doesn't look like RDNA4 will use It If It's released next year, unless some other manufacturer won't be faster.

 

[jpiniero]

16-24 Gb 32gbps GDDR7 is coming at the end of 2024.

It looks like availability is Q1ish 2024. But it is likely too late for RDNA4.

 

[F-no]

Hynix and Samsung have been sampling GDDR7 since august, Micron appears to be behind: https://hothardware.com/news/samsung-sk-hynix-sampling-gddr7-next-gen-gpus

 

[Tigerick]

Ugh. No?

wot

Man, do you still think that RDNA4 would use GDDR7?

 

[igor_kavinski]

Man, do you still think that RDNA4 would use GDDR7?

They might go with GDDR6X, leaving the expensive GDDR7 for nGreedia.

 

[adroc_thurston]

They might go with GDDR6X

not JEDEC thus functionally doesn't exist in AMD's eyes.

leaving the expensive GDDR7 for nGreedia.

joseph macri - AMD | LinkedIn

Experience: AMD · Education: wa · Location: San Francisco · 390 connections on LinkedIn. View joseph macri’s profile on LinkedIn, a professional community of 1 billion members.

www.linkedin.com

say this to his face ok?

Man, do you still think that RDNA4 would use GDDR7?

yeah, AMD tends to opt for fastest-anything JEDEC whenever it is available.

 

[Tigerick]

not JEDEC thus functionally doesn't exist in AMD's eyes.

joseph macri - AMD | LinkedIn

Experience: AMD · Education: wa · Location: San Francisco · 390 connections on LinkedIn. View joseph macri’s profile on LinkedIn, a professional community of 1 billion members.

www.linkedin.com

say this to his face ok?

yeah, AMD tends to opt for fastest-anything JEDEC whenever it is available.

But RDNA4 cards should be launching pretty soon and GDDR7 only available end of next year..

 

[adroc_thurston]

only available end of next year..

from Micron.
there are other DRAM vendors you see

 

[Tigerick]

from Micron.
there are other DRAM vendors you see

Below statement is from Anandtech regarding Samsung GDDR7's launching even though Samsung and Micron are supposedly HVM GDDR7 in 2024:

Given the timing of Samsung’s announcement (as well as Micron’s), the initial market for GDDR7 seems to be AI and networking accelerators, rather than the video cards that GDDR7 gets its name from. With both AMD and NVIDIA barely a quarter of the way through their current architectural release cycles, neither company is likely be in a position to use GDDR7 in 2024 when it’s ready. Instead, it’s going to be the other users of GDDR memory such as networking products and high-performance accelerators that are likely to be first to use the technology.

OTOH, RGT claimed that performance of RDNA4 cards to be between N31 and N32, I will let you guys decide which to believe,,,

 

[igor_kavinski]

OTOH, RGT claimed that performance of RDNA4 cards to be between N31 and N32, I will let you guys decide which to believe,,,

It would be stupid to release such cards, if true.

 

[adroc_thurston]

Below statement is from Anandtech regarding Samsung GDDR7's launching even though Samsung and Micron are supposedly HVM GDDR7 in 2024:

Nothingburger words.

It would be stupid to release such cards, if true.

It's a cheap midrange pair so the perf is right where it should be.

 

[Tigerick]

from Micron.
there are other DRAM vendors you see

Apparently, Micron is launching JEDEC's approval GDDR7 memory next year. We don't know whether Samsung and Hynix's sampling of GDDR7 is it 100% compliant or not? So I am still thinking RDNA4 cards will use GDDR6 instead, we shall see...

Micron Updates Roadmap, Promises 32 Gbit DDR5 and GDDR7 for 2024

During yesterday's HBM3 Gen2 memory products yesterday, Micron also shared an updated roadmap with select media and partners. The most interesting details on that roadmap were updates to DRAM and GDDR memory products, with increases in capacity coming for both types of memory. Micron is aiming...

www.techpowerup.com

 

[Tigerick]

It would be stupid to release such cards, if true.

How so? I thought you are the one who believe RDNA4 card (at least one of two) will perform better than N32 cards?

 

[igor_kavinski]

How so? I thought you are the one who believe RDNA4 card (at least one of two) will perform better than N32 cards?

I didn't know they would be cheaper, as adroc says. $400 card approaching N31 would be nice.

 

[Tigerick]

not JEDEC thus functionally doesn't exist in AMD's eyes.

joseph macri - AMD | LinkedIn

Experience: AMD · Education: wa · Location: San Francisco · 390 connections on LinkedIn. View joseph macri’s profile on LinkedIn, a professional community of 1 billion members.

www.linkedin.com

say this to his face ok?

yeah, AMD tends to opt for fastest-anything JEDEC whenever it is available.

If JEDEC is not formally approved the specs of GDDR7, then Micon should be the first vendor to launch GDDR7 memory by end of the 2024.... hoho, something to think about

 

[Tigerick]

I didn't know they would be cheaper, as adroc says. $400 card approaching N31 would be nice.

That would essentially kill N32 cards, do you believe AMD would do such a bold move?

 

[SolidQ]

It would be stupid to release such cards, if true.

Between 7800XT and 7900XT is like 27-28% difference in 1440p, based on TPU.
If 8700XT, faster 15-20% than 7800XT, and faster in path tracing like 3+ times, than 7900XT, and for 500$, it would be fine

 

[adroc_thurston]

JEDEC's approval GDDR7 memory

Non-JEDEC GDDR7 doesn't exist.
It's a JEDEC standard.

If JEDEC is not formally approved the specs of GDDR7, then Micon should be the first vendor to launch GDDR7 memory by end of the 2024.... hoho, something to think about

what

That would essentially kill N32 cards, do you believe AMD would do such a bold move?

new product, replaces old product.
truly bold move

 

[Tuna-Fish]

It would be stupid to release such cards, if true.

If it's a monolithic design that's cheaper to make than N32?

If the rumors about high end being axed and only N43 and N44 launching are true, personally I would be positively surprised if the N43 beat the full N32.

That would essentially kill N32 cards, do you believe AMD would do such a bold move?

Yes? N32 killed N21. AMD will just cease manufacture of N32 early enough that when they launch it, the stock has ran out.

 

[Tigerick]

Non-JEDEC GDDR7 doesn't exist.
It's a JEDEC standard.

what

new product, replaces old product.
truly bold move

Standards & Documents Search | JEDEC

www.jedec.org

Please update me on JEDEC standard in case I missed it

 

[Tigerick]

If it's a monolithic design that's cheaper to make than N32?

If the rumors about high end being axed and only N43 and N44 launching are true, personally I would be positively surprised if the N43 beat the full N32.



Yes? N32 killed N21. AMD will just cease manufacture of N32 early enough that when they launch it, the stock has ran out.

N21 was launched in Q4 2020 and 7800XT was launched in Q3 2023, exactly three years later. Thanks for confirming my theory

 

[SolidQ]

Big question is vram 12 or 16
If it's gonna be around 4070ti in PT, that good, but if AMD gonna be failed 3rd time with RT, than they lost even more market share.

 

[HurleyBird]

If the rumors about high end being axed and only N43 and N44 launching are true, personally I would be positively surprised if the N43 beat the full N32.

It would be right in line with AMD's history of self destructive GPU business strategies. G80 wasn't the beginning of the end for AMD's aspirations of graphics leadership, RV770 was. AMD needed a R600/G80 moment, and could have had it just one generation after G80, but despite a far superior architecture decided to only take a midrange die to market. AMD's GPU side has always had issues delivering the right product at the right time. You can't really predict performance based on what segments they are releasing to until AMD's history of incompetent GPU business decisions is sufficiently in the rear view mirror.

That said, the narrative going around is that high end RDNA4 was shelved because AMD bit off more than they could chew with advanced packaging, and to the extent that is true the decision to only release low end parts may have nothing to do at all with performance projections, and everything to do with not having monolithic backup plans in place for those segments.

 

[adroc_thurston]

AMD bit off more than they could chew with advanced packaging

Packaging is the easy part.