Discussion RDNA4 + CDNA3 Architectures Thread

DisEnchantment · Mar 23, 2022

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 · Jan 25, 2025

IEC said:
If 5090 idle power draw is any indication, GDDR7 doesn't exactly sip power.

That's just 512b and chungus die things.

soresu said:
The coming capacitorless DRAM designs

the what.
What's coming is 3D DRAM, but that's still caps. They're sideways (sort of).

soresu · Jan 25, 2025

adroc_thurston said:
the what.

Imec Demonstrates Capacitor-less IGZO-Based DRAM Cell With >400s Retention Time | imec

Back-end-of-line compatible architecture paves the way to high-density 3D-DRAM memories

www.imec-int.com

Capacitor-less IGZO-based DRAM cell | imec

Device architecture improvement is complemented with new insights in IGZO transistor reliability.

www.imec-int.com

Heartbreaker · Jan 25, 2025

IEC said:
Launching after nVidia is also the right move historically.

Right move how???

If you are ready, you launch first and have the market to yourself for while.

Launching first would also mean you only face your competitors previous generation product in reviews.

Launching first would also mean, when you competitor launches, his reviews will feature your new product, so you get extra free coverage.

All the advantages belong to launching first.

adroc_thurston · Jan 25, 2025

soresu said:
Imec Demonstrates Capacitor-less IGZO-Based DRAM Cell With >400s Retention Time | imec

Back-end-of-line compatible architecture paves the way to high-density 3D-DRAM memories

www.imec-int.com

Capacitor-less IGZO-based DRAM cell | imec

Device architecture improvement is complemented with new insights in IGZO transistor reliability.

www.imec-int.com

Oh no that's IMEC R&Dware, eons and eons away from mass production.

beginner99 · Jan 26, 2025

Heartbreaker said:
Launching first would also mean, when you competitor launches, his reviews will feature your new product, so you get extra free coverage.

Not if the reviewer gets a list of what he can and can't show or compare and has to use multi fake frames in any comparisons.

I think AMD wants the reviews of their own cards be directly compared to 5000-series. Which means no NV interfering with what can and can't be shown.

The other option is software issues. yes these could be fixed later but all the effing reviews would be full about that and people reading it months later would think it's still broken. Like the 7900 xt pricing failure which is in every 7900 xt review.

adroc_thurston · Jan 26, 2025

beginner99 said:
I think AMD wants the reviews of their own cards be directly compared to 5000-series. Which means no NV interfering with what can and can't be shown.

Ding-ding-ding.
People can't wait™ for Blackwell© when Blackwell® is already here and it sucks.

Thunder 57 · Jan 26, 2025

beginner99 said:
Not if the reviewer gets a list of what he can and can't show or compare and has to use multi fake frames in any comparisons.

I think AMD wants the reviews of their own cards be directly compared to 5000-series. Which means no NV interfering with what can and can't be shown.

The other option is software issues. yes these could be fixed later but all the effing reviews would be full about that and people reading it months later would think it's still broken. Like the 7900 xt pricing failure which is in every 7900 xt review.

Nah. Tom's reviewed the 5090 fairly and gave it 4.5/5 stars despite early drivers and certain things just not working. I'm sure AMD would be treated equally /s. It's their most egrigous Nvidia shilling since the "Just buy it" BS.

coercitiv · Jan 26, 2025

Thunder 57 said:
Nah. Tom's reviewed the 5090 fairly and gave it 4.5/5 stars despite early drivers and certain things just not working. I'm sure AMD would be treated equally /s.

Of course they would:

marees · Jan 26, 2025

beginner99 said:
Not if the reviewer gets a list of what he can and can't show or compare and has to use multi fake frames in any comparisons.

I think AMD wants the reviews of their own cards be directly compared to 5000-series. Which means no NV interfering with what can and can't be shown.

The other option is software issues. yes these could be fixed later but all the effing reviews would be full about that and people reading it months later would think it's still broken. Like the 7900 xt pricing failure which is in every 7900 xt review.

If I were Lisa Su / Jack Hyunh (& I really wanted to sell GPUs), then I would have released RDNA 4, 6 months ago ($800 9070xt & $650 9070). Dropped prices gradually as stock moves & released FSR 4 whenever it was ready in an Incremental fashion

Heartbreaker · Jan 26, 2025

beginner99 said:
Not if the reviewer gets a list of what he can and can't show or compare and has to use multi fake frames in any comparisons.

Prove this happened on any GPU review in the last decade. Every GPU has always been tested against the competitions latest cards and usually the previous generation as well.

I've yet to here the compelling argument for launching second.

IMO, the only reason to not launch first, is if your product needs more work.

Win2012R2 · Jan 26, 2025

marees said:
I would have released RDNA 4, 6 months ago ($800 9070xt & $650 9070)

They had lots of unsold RDNA3 inventory and software most likely was not ready, because it never is

reaperrr3 · Jan 26, 2025

adroc_thurston said:
Ding-ding-ding.
People can't wait™ for Blackwell© when Blackwell® is already here and it sucks.

The other aspect is, availability of Blackwell in general and anything above the 5070Ti in particular seems to be atrocious for the first couple of weeks according to rumors and could possibly remain that way until RDNA4 launches or even beyond.
So in practice, all this "time to market is so important, now everyone will buy a 50x0 before RDNA4 even launches!" is likely bogus, since the Blackwell supply for that just won't be there.

Ranulf · Jan 26, 2025

Heartbreaker said:
Prove this happened on any GPU review in the last decade. Every GPU has always been tested against the competitions latest cards and usually the previous generation as well.

I've yet to here the compelling argument for launching second.

IMO, the only reason to not launch first, is if your product needs more work.

Eh, off the top of my head, maybe Polaris vs Pascal in 2016. Though I doubt releasing early for Polaris would have net them that many sales.

Hail The Brain Slug · Jan 26, 2025

Heartbreaker said:
Prove this happened on any GPU review in the last decade. Every GPU has always been tested against the competitions latest cards and usually the previous generation as well.

I've yet to here the compelling argument for launching second.

IMO, the only reason to not launch first, is if your product needs more work.

Hardware Unboxed got banned from nvidia FE reviewing for not showcasing the specific games and settings nvidia put in their review guide, even though they still covered DLSS and RT, because their review wasn't entirely about the things that Nvidia cards would be significantly faster at (and the technology nvidia keeps trying to gaslight is into thinking we asked for and wanted).

I think it was Linus on the WAN show that pulled out the nvidia review guide and shared the details inside?

Even though nvidia backtracked the entire thing, the message was clear: showcase what we tell you in the review guide with the settings we prescribe.... or else.

If you wonder why all the reviews tend to show the same games with the same configuration (DLSS4, RT ultra, whatever) its because the review guide demands it. Only the brave reviewers who aren't afraid for their livelihood test outside the review guide.

The fact he points out the email they got was written more like a press release, seemingly with the intention it would be publicly shared means it was a greater message to the entire reviewer community.

Heartbreaker · Jan 26, 2025

Hail The Brain Slug said:
Hardware Unboxed got banned from nvidia FE reviewing for not showcasing the specific games and settings nvidia put in their review guide, even though they still covered DLSS and RT, because their review wasn't entirely about the things that Nvidia cards would be significantly faster at (and the technology nvidia keeps trying to gaslight is into thinking we asked for and wanted).

That wasn't the claim.

The claim was they would be blocked from testing against competitor cards. Which I've yet to see happen.

Hail The Brain Slug · Jan 26, 2025

Heartbreaker said:
That wasn't the claim.

The claim was they would be blocked from testing against competitor cards. Which I've yet to see happen.

That's not what you quoted. The issue isnt being told they can't test a specific competitor card.

The issue is for the official Nvidia review, they basically have to follow the review guide which tells them which games and settings they have to use. This means if AMD releases first, the nvidia review will follow nvidia prescribed format and make nvidia look as good as possible.

If nvidia releases first, this is done in a vacuum against last gen cards. Then when AMD releases, nvidia units are retail and now AMD reviews can include nvidias new cards with zero restrictions on what games and settings (and features) they test with. In general it will put AMD in a better light.

Heartbreaker · Jan 26, 2025

Hail The Brain Slug said:
That's not what you quoted. The issue isnt being told they can't test a specific competitor card.

The issue is for the official Nvidia review, they basically have to follow the review guide which tells them which games and settings they have to use. This means if AMD releases first, the nvidia review will follow nvidia prescribed format and make nvidia look as good as possible.

If nvidia releases first, this is done in a vacuum against last gen cards. Then when AMD releases, nvidia units are retail and now AMD reviews can include nvidias new cards with zero restrictions on what games and settings (and features) they test with. In general it will put AMD in a better light.

Review guides are NOT that restrictive. If they were, they would get called on it, just like they did when they blocked HWUB, and note that had ZERO to do with review guides. They just didn't like that HWUB dumped on Ray Tracing. Pretty much NO reviewer turned on FG when comparing to competing cards.

Even if there was some small chance that NVidia could get away with somehow tilting all the reviews (extremely unlikely) against AMD cards, that is NOT a reason for AMD to give up an advantage and not even try.

Anyone with an attitude that amount's to: "don't even try because their competitor might do shady to counter", should just shut down their company now.

You make full use of any competitive advantage you have (like being able to launch first), and then if your competitor does something shady, you deal with it when it happens.

You don't just roll over ahead of time, because there is some minute chance your advantage will face a counter.

What a ridiculous take.

Still waiting to hear the big advantage of letting NVidia launch first...

Win2012R2 · Jan 26, 2025

Heartbreaker said:
Review guides are NOT that restrictive.

They don't need to - if reviewers don't play ball (in any serious way) then they won't be reviewing hot Nvidia products and getting lots of views.

Hail The Brain Slug · Jan 26, 2025

Heartbreaker said:
What a ridiculous take.

I literally cited sources. I guess you can't accept reality.

Heartbreaker · Jan 26, 2025

Hail The Brain Slug said:
I literally cited sources. I guess you can't accept reality.

Sources of what? NVidia PR getting mad and doing something stupid. That was nothing to do with review guides.

That overstep had NVidia eating their words, backpeddling and much less likely to try something that stupid and shady again.

Again the real ridiculous part, is the take that AMD should just give up now, because NVidia might do something.

That's no way to run any business.

GodisanAtheist · Jan 26, 2025

Only shady thing I've seen reviewers doing with the 5090 is the usual 8/10 solid product review summary, and then expressing a lot of disappointment in the 30% performance/power/cost increase in the actual written text of the review (that most folks aren't going to read through).

Ofc there are the meme reviewers out there but that can't ever be avoided.

Win2012R2 · Jan 26, 2025

GodisanAtheist said:
that most folks aren't going to read through

Most folks won't buy it anyway...

carrotmania · Jan 26, 2025

Heartbreaker said:
Still waiting to hear the big advantage of letting NVidia launch first...

IF RDNA4 is as good as reports (ie possibly better than a.L 7900XTX by now, which means better than a 4080 in some things and therefore close to a 5080) even if they went now, without hard numbers for the 5080, let alone the 5070Ti which is more is price point, there's a high chance all reviews would be like "it's great, but wait to see what NV brings", so people wait, and read a NV review, and then purchase, with a NV review fresh in their mind.

But maybe they are thinking of they can time it right, and price it right, with the right performance and software stack, maybe just enough people this time will read a 9070XT review "last", and order on the spot.

It's not like it's helped them having better performance and earlier reviews previously, so they might be trying something disruptive, even if that annoys YOU.

Heartbreaker · Jan 26, 2025

carrotmania said:
IF RDNA4 is as good as reports (ie possibly better than a.L 7900XTX by now, which means better than a 4080 in some things and therefore close to a 5080) even if they went now, without hard numbers for the 5080, let alone the 5070Ti which is more is price point, there's a high chance all reviews would be like "it's great, but wait to see what NV brings", so people wait, and read a NV review, and then purchase, with a NV review fresh in their mind.

But maybe they are thinking of they can time it right, and price it right, with the right performance and software stack, maybe just enough people this time will read a 9070XT review "last", and order on the spot.

It's not like it's helped them having better performance and earlier reviews previously, so they might be trying something disruptive, even if that annoys YOU.

If it's that much better price/performance better than NVidia it would likely sell out. The later reviews aren't really going to tell anything about the 5070/ti that we don't already know.

A product NOT in the market is guaranteed to sell ZERO. There is ZERO to be gained by artificially waiting, and all the interim sales they could have to lose.

The only valid reason to wait, is if the product/features aren't really in a good enough state to ship.

adroc_thurston · Jan 26, 2025

Heartbreaker said:
The later reviews aren't really going to tell anything about the 5070/ti that we don't already know.

YOU know. you, the tech nerd. GPU launches aren't for nerds.

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Senior member

Diamond Member

Senior member

Member

Platinum Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Senior member

Diamond Member

Diamond Member

Diamond Member

Senior member

Member

Diamond Member

Diamond Member