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 · May 2, 2024

soresu said:
while Sammy are starting up a 2nd gen nanosheet/GAAFET 3nm process node this year.

where are the shipping parts with gen1.

soresu · May 2, 2024

adroc_thurston said:
RDNA4.5 APUs are also dead!

Wait, wat?!??!

First I even heard of them.

adroc_thurston · May 2, 2024

soresu said:
First I even heard of them.

Kraken 1/2 (the OG ones), had Zen5+ and RDNA4.5, direct strix1/2 replacements.

soresu · May 2, 2024

adroc_thurston said:
where are the shipping parts with gen1.

Quite, I haven't heard of a single one which makes it sound a bit like TSMC N3B, only at higher difficulty due to a new device type.

soresu · May 2, 2024

adroc_thurston said:
Kraken 1/2 (the OG ones), had Zen5+ and RDNA4.5, direct strix1/2 replacements.

Yeah that makes sense, much more appropriate naming vs a value part.

adroc_thurston · May 2, 2024

soresu said:
Quite, I haven't heard of a single one which makes it sound a bit like TSMC N3B, only at higher difficulty due to a new device type.

N3b actually ships in chungus volumes (Apple).
Where is SF3?

soresu · May 2, 2024

adroc_thurston said:
N3b actually ships in chungus volumes (Apple).

Oh, I thought that only Apple had managed to tape something out on it because it was b0rked from being too customised for them.

adroc_thurston · May 2, 2024

soresu said:
from being too customised for them.

No such thing at foundry.

soresu · May 2, 2024

Ooooof, case in point why consumer gfx cards and APUs are not likely to get HBM in the near future.....

SK hynix Reports That 2025 HBM Memory Supply Has Nearly Sold Out

www.anandtech.com

ToTTenTranz · May 3, 2024

soresu said:
Ooooof, case in point why consumer gfx cards and APUs are not likely to get HBM in the near future.....

SK hynix Reports That 2025 HBM Memory Supply Has Nearly Sold Out

www.anandtech.com

SK Hynix aren't the only ones doing HBM, but my guess is Samsung is also booked completely for AI.

Joe NYC · May 3, 2024

marees said:
Intel will do this to AMD with battlemage

Intel is in worse position than AMD to be a charity.

inquiss · May 3, 2024

SolidQ said:
Because they think, if AMD can't compete with NV, they've out(there enought posts about it), also trusting in Intel to compete vs NV

If you read difference gaming forums, there always praise DLSS to heaven. called DLSS "Black Magic"

Yeah, b cause Nvidia made it. If it wasn't that it would be drivers or anti lag, the tech picked is irrelevant, really.

coercitiv · May 3, 2024

Joe NYC said:
Intel is in worse position than AMD to be a charity.

Nonsense, all they need is to update their sticker:

Intel inside. Subsidized with US taxpayer money.

There you go, make gaming great again!

Panino Manino · May 3, 2024

ToTTenTranz said:
Alex / Dictator is a massive propagandist for nvidia and it's honestly hard to believe he's not on the payroll, but to be honest I find Oliver Mackenzie and to a point even Richard and John to be fair-er in their assessment.
In reality, I think it's just Alex who's a really poor fit for the team because the man has reduced himself to a talking advertisement.

He probably has his favorites, but I think this bias is a bit unintentional.
He likes technology, the became a member exactly for this reason. Companies are giving him access to their tech, giving him the opportunity to talk about and explain how they work, so from his point of view he is just doing a "public service".

Problem is that AMD is in the gutter and have no party to invite him.

igor_kavinski · May 3, 2024

Panino Manino said:
Problem is that AMD is in the gutter and have no party to invite him.

Really? I don't smell anything from the two RDNA2 Radeons I have.

ToTTenTranz · May 3, 2024

Panino Manino said:
He probably has his favorites, but I think this bias is a bit unintentional.

No freaking way.

Panino Manino said:
Problem is that AMD is in the gutter and have no party to invite him.

AMD gives Digital Foundry the same access as every other tech outlet with a reasonable following.
The other members of DF aren't flagrant marketeers for Nvidia like Alex.
With suggestions that the RTX 2060 shows similar performance to a PS5 and other preposterous claims throughout the years, the guy has pretty much memed himself.

adroc_thurston · May 3, 2024

Panino Manino said:
He probably has his favorites, but I think this bias is a bit unintentional.

It's very intentional and he dreams of working at NV product marketing.
It's so transparent a toddler can understand that.

ToTTenTranz · May 3, 2024

adroc_thurston said:
It's very intentional and he dreams of working at NV product marketing.

To be honest, the faster this happens the better. Digital Foundry's legitimacy as a pro-consumer team gets lost a lot because of his blatant nvidia marketing, IMO.

adroc_thurston · May 3, 2024

ToTTenTranz said:
To be honest, the faster this happens the better. Digital Foundry's legitimacy as a pro-consumer team gets lost a lot because of his blatant nvidia marketing, IMO.

Why hire him when he does an amazing job for free?

Saylick · May 3, 2024

adroc_thurston said:
Why hire him when he does an amazing job for free?

Basically, this. He is essentially a corporate mouthpiece, intentional or not. A useful idiot, if you will. Doesn't matter if his intentions are innocent; the results speak for themselves. Also, Nvidia probably doesn't want him as an Nvidia employee because then it removes the thin shroud of neutrality since he's supposed to work for an independent media outlet.

adroc_thurston · May 3, 2024

Saylick said:
Basically, this. He is essentially a corporate mouthpiece, intentional or not. A useful idiot, if you will. Doesn't matter if his intentions are innocent; the results speak for themselves.

He's not an idiot, oh no.
It's still a gross miscalculation since he does an amazing job at his current position thus no need to hire him.

Mopetar · May 3, 2024

adroc_thurston said:
Naa they gotta move units.

Best way to move more units of big expensive datacenter GPUs is to make smaller consumer GPUs that don't use as many wafers.

What are NVidia customers going to do in that scenario, switch to Intel?

adroc_thurston · May 3, 2024

Mopetar said:
Best way to move more units of big expensive datacenter GPUs is to make smaller consumer GPUs that don't use as many wafers.

What are NVidia customers going to do in that scenario, switch to Intel?

No it's a very specific tradeoff between moving units and ASPs.
People not gonna buy kilobuck entry GPUs.

ToTTenTranz · May 3, 2024

Saylick said:
Nvidia probably doesn't want him as an Nvidia employee because then it removes the thin shroud of neutrality since he's supposed to work for an independent media outlet.

At this point, that shroud is basically made of a graphene sheet.

SolidQ · May 3, 2024

adroc_thurston said:
Why hire him when he does an amazing job for free?

it's not only Alex. Alot "journalist" when talking about system requirements, they writing like this "You need RTX 2070 for 1080/3080 for 1440", absolute ignoring Radeon cards. For them only NV exist

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Platinum Member

Platinum Member

Platinum Member

Platinum Member

Platinum Member

Platinum Member

Platinum Member

Platinum Member

Platinum Member

Member

Platinum Member

Member

Diamond Member

Senior member

Lifer

Member

Platinum Member

Member

Platinum Member

Diamond Member

Platinum Member

Diamond Member

Platinum Member

Member

Senior member