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 · Tuesday at 4:06 AM

Aapje said:
An issue is that the full chip is designed to be a balanced package, and it's not necessarily easy to cut it down in a way that makes financial sense and is still balanced.

Ah no GPUs are very much designed with die chops in mind.

soresu · Tuesday at 1:27 PM

adroc_thurston said:
Ah no GPUs are very much designed with die chops in mind.

Chops?

Do you mean physically fusing broken circuits on the board, or disabling them in firmware?

jpiniero · Tuesday at 1:32 PM

soresu said:
Chops?

Do you mean physically fusing broken circuits on the board, or disabling them in firmware?

Presumably talking about the design of the cut down dies.

Aapje · Tuesday at 4:08 PM

adroc_thurston said:
Ah no GPUs are very much designed with die chops in mind.

If they have to choose between harming the quality of the 'full' die and the cut down one, they definitely will choose to harm the cut down one, with the yields that TSMC has.

igor_kavinski · Wednesday at 11:08 AM

An ex-intern replied:

igor_kavinski · Wednesday at 11:11 AM

Funny that the teams hated this chip and how that guy is still describing Cadence tools as hell.

Wait. The ex-CEO of that hellish tool company is now Intel CEO

branch_suggestion · Wednesday at 8:12 PM

igor_kavinski said:
An ex-intern replied:

View attachment 121149

Very big RV770/Cypress vibes.
Maybe upper management can convince Lisa to give more wafer alloc next time so configs can stop being stupidly undersized.

igor_kavinski · Wednesday at 8:17 PM

branch_suggestion said:
Maybe upper management can convince Lisa to give more wafer alloc next time so configs can stop being stupidly undersized.

Unfortunately, won't work. They clearly didn't think they would be competitive with RTX 5000 so kept the VRAM limited to 16GB and probably had an emergency plan to go as low as $399 for RX 9070 but Nvidia handed them a nice victory on a platter. Next time, Jensen won't be so generous. It's almost impossible competing against a company whose only business is GPUs and both other competitors have tiny GPU divisions in comparison.

branch_suggestion · Wednesday at 8:32 PM

igor_kavinski said:
Next time, Jensen won't be so generous.

You build the X3D GPU, big enough that a mono slab cannot win.
NV is running out of arch tricks.

igor_kavinski · Wednesday at 8:38 PM

branch_suggestion said:
You build the X3D GPU, big enough that a mono slab cannot win.

That's the problem. They prefer their X3D GPU goes in a data center. Not to consumer peasant.

$10,000 a lot more enticing than $3500 for same effort and raw material

branch_suggestion · Wednesday at 8:46 PM

igor_kavinski said:
That's the problem. They prefer their X3D GPU goes in a data center. Not to consumer peasant.

$10,000 a lot more enticing than $3500 for same effort and raw material

It is about sending a message.
Everywhere NV competes, AMD cannot cheap out, they have to build the best.
They could cheap out vs Intel, cause Intel.
A simple enough 500 on 500 stack wins gaming, period.
Zen X3D guys could pull it off.

igor_kavinski · Wednesday at 9:36 PM

branch_suggestion said:
Zen X3D guys could pull it off.

We all wish. It ain't happening

Coz AMD is first and foremost a CPU and enterprise company, despite whatever they say.

ToTTenTranz · Thursday at 2:52 AM

igor_kavinski said:
Next time, Jensen won't be so generous.

Jensen will be generous if all he's doing is looking at AI profits.

Their sloppiness on gaming GPUs is so high they reached a point where even DF is complaining about their drivers.

SolidQ · Thursday at 7:06 AM

GRE or GSE?

https://videocardz.com/newz/amd-reportedly-preparing-radeon-rx-9070-gre

marees · Thursday at 8:38 AM

SolidQ said:
GRE or GSE?

https://videocardz.com/newz/amd-reportedly-preparing-radeon-rx-9070-gre

That was from March-9th right. Any updates after that ?

TESKATLIPOKA · Thursday at 1:42 PM

So when will we see N44 based GPUs?
5060Ti supposedly will be be announced on April 15 and reviews should be out a day later.

jpiniero · Thursday at 3:43 PM

TESKATLIPOKA said:
So when will we see N44 based GPUs?
5060Ti supposedly will be be announced on April 15 and reviews should be out a day later.

If AMD was smart, they would have dropped N44 for the time being for more N48 supply.

soresu · Thursday at 4:50 PM

ToTTenTranz said:
Jensen will be generous if all he's doing is looking at AI profits.

Their sloppiness on gaming GPUs is so high they reached a point where even DF is complaining about their drivers.

Ohhh how the turns have tabled..... 😆

soresu · Thursday at 5:00 PM

jpiniero said:
If AMD was smart, they would have dropped N44 for the time being for more N48 supply.

Define smart.

If AMD can make more N44's per wafer and sell the total for more (or even just the same as market share increase is its own reward) money than they sell however many N48's they get per water then that is the smart way to go.

At the end of the day 9070 XT's are selling really well, but there is almost certainly plenty of room in the market for a cheaper card, and that will be even more true as the impact of tariffs is felt.

GodisanAtheist · Thursday at 7:27 PM

ToTTenTranz said:
Jensen will be generous if all he's doing is looking at AI profits.

Their sloppiness on gaming GPUs is so high they reached a point where even DF is complaining about their drivers.

-Jensen knows that as soon as the AI bubble pops, he can throw gamers a bone and they'll all come crawling back to gnaw on the scraps.

Gamers aren't exactly the model of discerning customers, Nvidia is more or less banking them for the time being.

soresu · Friday at 2:03 AM

GodisanAtheist said:
Gamers aren't exactly the model of discerning customers

The same non discerning customers that still have a beef about AMD drivers that really hasn't been a thing for years now?

GodisanAtheist · Friday at 10:55 AM

soresu said:
The same non discerning customers that still have a beef about AMD drivers that really hasn't been a thing for years now?

-*Kame Hame's his best Adroc Impression*

They were never going to buy AMD anyway.

adroc_thurston · Friday at 11:59 AM

GodisanAtheist said:
They were never going to buy AMD anyway.

Truth-tsarbomba.

Geranium · Friday at 1:41 PM

Vikv1918 said:
Idk its very hard to think of anything other than some kind of market politics being the reason. I dont think consumer preference has anything to do with the current state of Radeon laptops having 0.05% marketshare. At that marketshare the consumer is irrelevant because they dont even have a choice in the first place. Do we have a choice to buy a $700 7600m laptop? Or a $1000 7800m laptop? Or a $1400 7900m laptop? No we don't, so we can't prove "people dont buy AMD".

Furthermore, its not an "all or nothing" scenario either. Its totally viable from a financial perspective for both AMD and OEMs for Radeon to have 20% laptop marketshare just like their desktop GPUs. Somehow AMD was selling laptops pre-Ryzen and it didn't have a 0.01% marketshare like they do in dGPU.

OEMs simply dont offer such products (and don't tell me its not technically or economically feasible, nvidia themselves sell $750 4060 laptops). And its become even more glaring with the 9000 series which would make the absolute best laptop chips at their price and efficiency.

IMO the gaming laptop situation needs to be researched by the tech or investigative journalists. I have never seen anything like this total monopoly before.

OEMs are definitely shilling for Nvidia. There are more Strix Point laptop with dGPU than Strix Point only laptop; And Strix Point is designed for thin and light-weight laptop.

Let me give you a example:
Specification of Picasso based ASUS laptop : Ryzen 5 3500U + 4GiB Single-Channel RAM + MX230 + 1TB HDD. And you read right no SSD in a laptop launched in 2019, but has a Nvidia GPU. No SSD or dual channel memory, but hey it has a nvidia gpu which is not that faster than the iGPU and consumes more power. Nvidia shilling at best.

Geranium · Friday at 1:46 PM

ToTTenTranz said:
Jensen will be generous if all he's doing is looking at AI profits.

Their sloppiness on gaming GPUs is so high they reached a point where even DF is complaining about their drivers.

Looks kind of fake. Digital Foundry played part for the current GPU situation.

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Diamond Member

Diamond Member

Lifer

Golden Member

Lifer

Lifer

Senior member

Lifer

Senior member

Lifer

Senior member

Lifer

Senior member

Golden Member

Senior member

Platinum Member

Lifer

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Member

Member