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

TESKATLIPOKA · Jan 15, 2025

I don't understand why the discussion turned into what chip AMD should have made along with N44 and N48 to compete at higher end against Nvidia, when we don't know the actual performance of N48 or N44.

adroc_thurston · Jan 15, 2025

TESKATLIPOKA said:
when we don't know the actual performance of N48 or N44

we do.
N48 full die is 7900XTX, chop is 7900XT.
n44 oh well who cares about that one.

SolidQ · Jan 15, 2025

adroc_thurston said:
N48 full die is 7900XTX, chop is 7900XT.

no 12gb version N48?

adroc_thurston · Jan 15, 2025

SolidQ said:
no 12gb version N48?

not in the launch window so idk what it is.

SolidQ · Jan 15, 2025

adroc_thurston said:
not in the launch window so idk what it is.

if Full XTX and cut down XT, i'd assuming 12gb is 7800XT perf

gaav87 · Jan 15, 2025

SolidQ said:
no 12gb version N48?

Maybe later like 5600xt

eek2121 · Jan 15, 2025

Win2012R2 said:
Hope somebody gets fired for that losing attitude.

It’s intentional. From what I have heard AMD may try to work on optimizing clocks. That doesn’t mean they won’t have bigger parts, but die size is the focus going forward. Smaller die = less expensive to make.

I do how we get a 96-144CU card next gen, but if they gave us even faster clocks that wouldn’t be bad either.

Keller_TT · Jan 15, 2025

So, any rumors or whispers about when really are we gonna see RDNA4?
Was hoping at least we get an idea today, but bed time now...
Just launch it already, Lisa! Set a date!

Bryo4321 · Jan 15, 2025

Keller_TT said:
So, any rumors or whispers about when really are we gonna see RDNA4?
Was hoping at least we get an idea today, but bed time now...
Just launch it already, Lisa! Set a date!

AMD still adamantly saying “soon”. And apparently some retailers already have stock.

Just weird we have no official details yet. You could speculate in a positive or completely negative direction I feel.

Josh128 · Jan 15, 2025

adroc_thurston said:
we do.
N48 full die is 7900XTX, chop is 7900XT.
n44 oh well who cares about that one.

Erm..sauce please? I got a big doubt that its going to match XTX in anything but RT.

Josh128 · Jan 15, 2025

eek2121 said:
It’s intentional. From what I have heard AMD may try to work on optimizing clocks. That doesn’t mean they won’t have bigger parts, but die size is the focus going forward. Smaller die = less expensive to make.

I do how we get a 96-144CU card next gen, but if they gave us even faster clocks that wouldn’t be bad either.

Its the Intel Raptor Lake approach, I cant really say I dig it very much. 3GHz, 330W midrange is not my cup of tea. 3GHz, 220W I'll take all day.

adroc_thurston · Jan 15, 2025

Josh128 said:
Erm..sauce please?

Pay me.

Josh128 said:
I got a big doubt that its going to match XTX in anything but RT.

Yes I know it HAS to suck right. It's Radeon.

CastleBravo · Jan 15, 2025

Heartbreaker said:
Someone just posted more NVidia 50 series, performance numbers without Fake Frames, and they are brutal:

Page 78 - Discussion - Nvidia Blackwell in Q1-2025

Page 78 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Looks like Nvidia has put AMD in an extremely awkward position where they are going to have to put in more effort than usual to ensure they miss the opportunity.

eek2121 · Jan 15, 2025

Josh128 said:
Its the Intel Raptor Lake approach, I cant really say I dig it very much. 3GHz, 330W midrange is not my cup of tea. 3GHz, 220W I'll take all day.

You’ll note that those specs are rumors, and you don’t even know how it will perform, so it is kind of foolish to make assumptions, don’t you think?

I’d wait for reviews. Based on my own interpretation of everything, tons of folks are bashing AMD like they did with Zen 1, and everyone is talking about how cool NVIDIA is, even the vanilla RTX 5070 has a higher TDP (to say nothing of TGP, which is a different number) than you quoted.

EDIT: As an example, tons of folks are claiming 380mm2 for a die size, they are way off.

Josh128 · Jan 15, 2025

eek2121 said:
You’ll note that those specs are rumors, and you don’t even know how it will perform, so it is kind of foolish to make assumptions, don’t you think?

I’d wait for reviews. Based on my own interpretation of everything, tons of folks are bashing AMD like they did with Zen 1, and everyone is talking about how cool NVIDIA is, even the vanilla RTX 5070 has a higher TDP (to say nothing of TGP, which is a different number) than you quoted.

EDIT: As an example, tons of folks are claiming 380mm2 for a die size, they are way off.

Im not bashing anything. 3GHz aint coming on a 4nm GPU without comparatively high power consumption for ANY given die size. Im just stating facts, and the facts are that the only interest I have in >250W GPUs is reading about them.

They are the best rumors we have. If rumors are off limits to discuss, what the hell is everyone doing here?? 380mm2 was estimated by photo comparison, what proof do you have that its way off?

adroc_thurston · Jan 15, 2025

Josh128 said:
3GHz aint coming on a 4nm GPU without comparatively high power consumption for ANY given die size

You sure?

Josh128 said:
380mm2 was estimated by photo comparison, what proof do you have that its way off?

People with better estimation skills narrowed it down to 350mm^2 or less.

gdansk · Jan 15, 2025

Josh128 said:
3GHz aint coming on a 4nm GPU without comparatively high power consumption for ANY given die size.

I know it's comparatively high but some RTX 4060 can be overclocked to 3GHz at ~160W.
And I don't think it was ever designed for that.

del42sa · Jan 16, 2025

https://videocardz.com/pixel/amd-says-radeon-rx-9070-series-deserves-its-own-event-stay-tuned

the comedy has a sequel 🍿 🍿 🍿 🍿🍿🍿

beginner99 · Jan 16, 2025

Heartbreaker said:
Pricing against NVidia is tricky. Especially with the price cut on the 5070. For the following I'm assuming that 9070 has better performance than 5070, but 9070XT underperforms 5070 Ti.

exactly that is why a 9070 XT would be $150 cheaper than the 5070 Ti at $599 and the 9070 $50 less than a 5070. the 5070 will have a lower generational increase than the Ti so in general is weaker and closer to 9000 series and the 9070 will also have 4 gb more RAM as additional selling point over a 5070 while being clearly faster in raster.

Also supply and demand. if the performance rumors are true and a 9070 XT is $549 with 7900xtx performance, could AMD even supply enough to satisfy demand?

adroc_thurston · Jan 16, 2025

beginner99 said:
could AMD even supply enough to satisfy demand?

what demand lmao.
People do NOT buy AMD GPUs. That's the rule. Radeon exists to force geforce pricecuts.

Saylick · Jan 16, 2025

beginner99 said:
Also supply and demand. if the performance rumors are true and a 9070 XT is $549 with 7900xtx performance, could AMD even supply enough to satisfy demand?

Honestly, they should just pull an Nvidia and have an unrealistically low MSRP, say $499, and let the street price be what it is. The market has historically perceived this strategy better than the opposite, which is to set a high MSRP and let prices fall over time.

SolidQ · Jan 16, 2025

adroc_thurston said:
People do NOT buy AMD GPUs. That's the rule. Radeon exists to force geforce pricecuts.

Globally yes, but not everyone buying GF.

Abwx · Jan 16, 2025

adroc_thurston said:
what demand lmao.
People do NOT buy AMD GPUs. That's the rule. Radeon exists to force geforce pricecuts.

We re not talking OEMs or urban legends, in the DIY market they do substancial sales.

https://twitter.com/x/status/1878582444049289689

adroc_thurston · Jan 16, 2025

SolidQ said:
Globally yes, but not everyone buying GF.

Abwx said:
We re not talking OEMs or urban legends, in the DIY market they do substancial sales.

https://twitter.com/x/status/1878582444049289689

Anecdotal stuff compared to JPR data.

Keller_TT · Jan 16, 2025

Saylick said:
Honestly, they should just pull an Nvidia and have an unrealistically low MSRP, say $499, and let the street price be what it is. The market has historically perceived this strategy better than the opposite, which is to set a high MSRP and let prices fall over time.

That's the better way, but not sure if 499 is unrealistic. I certainly don't think they can go beyond 549 for the reference XT, or the whole GPU market needs a lowdown and shakeup. I do hope for Intel to catch up with drivers and refined architecture by 2026-27 launch cycle. Funny if they can have their Ryzen moment through GPUs when AMD fights for relevance against NV in CPUs too.

The market does reward sanity waiting out though. I just got a msg from a retailer buddy that I can have the last 7950X3D box that he has left for ₣450. What a star!
Now, that's a far better deal than the disappearing 9800X3D, which has the halo attached to it now, and somehow this crazy world thinks that few extra FPS for games makes it better.

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Platinum Member

Diamond Member

Golden Member

Diamond Member

Golden Member

Senior member

Diamond Member

Member

Member

Senior member

Senior member

Diamond Member

Member

Diamond Member

Senior member

Diamond Member

Diamond Member

Member

Diamond Member

Diamond Member

Diamond Member

Golden Member

Lifer

Diamond Member

Member