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

 

[stayfrosty]

If they wanted they would've announced the roadmap.

Maybe it's too far out to announce (early 2026?). Going to guess bringing it back to life would take some time.

 

[gdansk]

View attachment 118380

Huh? Are they going to revive big navi?

I read that as he's saying they don't exist.

 

[blckgrffn]

I read that as he's saying they don't exist.

Same.

 

[SolidQ]

If they wanted they would've announced the roadmap.

but they don't have it, secretly playing?

 

[PJVol]

So what is bottlenecking the XT so hard? Memory?

Nothing. +15-20% clocks -> 15%+ performance.

 

[Ghostsonplanets]

RX 9060 coming in Q2 2025 (Apr - Jun)

RDNA 4 certainly is impressive. 9060 might dominate $300 and below bracket.

 

[SolidQ]

9060 might dominate $300 and below bracket.

big question it's still N48 for both or one N44

 

[gdansk]

9060 might dominate $300 and below bracket.

Don't you think it's closer to $400?

 

[ryanjagtap]

RX 9060 coming in Q2 2025 (Apr - Jun)

RDNA 4 certainly is impressive. 9060 might dominate $300 and below bracket.

Navi 44 full die is 32 CU right?

 

[adroc_thurston]

Navi 44 full die is 32 CU right?

Yea, basically half N48.

 

[Meteor Late]

You'll see whatever esoteric encoding/decoding that Nvidia supports and AMD doesn't will be exactly the differentiator.

4:4:4 encode/decode.

 

[marees]

RX 9060 coming in Q2 2025 (Apr - Jun)

RDNA 4 certainly is impressive. 9060 might dominate $300 and below bracket.

5700 xt ($400) = 9070xt ($600)
5700 ($350) = 9070 ($550)
5600xt ($280) = 9060xt ($400+)

5500xt 8gb ($200) = 9050xt 16gb ($330?)
5500xt 4gb ($170) = 9050xt 8gb ($280?)
5500 OEM only = 9040xt ( $230?)

 

[ryanjagtap]

Yea, basically half N48.

If you are sure that Navi 48 is just 2x Navi 44, are you certain that AMD gimped the media engine et al on the 44 die?

 

[Philste]

Nothing. +15-20% clocks -> 15%+ performance.

So we ignore the 56 vs 64CUs (+14%) and that theoretical performance and power draw is up ~35%?

 

[marees]

AMD promises FSR 4 support for 30 games at launch with 75+ games slated to receive support for the tech throughout 2025.

FSR 4 uses a proprietary model that takes the best of CNN and transformer, according to the company.

AMD tells us that FSR 4 utilizes FP8 capabilities of RDNA 4's 2nd gen AI accelerators, which means older Radeon cards won't be reaping the benefits, at least for now. While there could be a possibility of backporting FSR 4 or its subsets to older cards, it's best to overlook that for now.

https://www.notebookcheck.net/AMD-t...limited-to-RDNA-4-cards-for-now.969986.0.html

 

[Panino Manino]

AMD should adopt a PR naming already.
Radeon 7 Series PR70.

 

[gdansk]

AMD should adopt a PR naming already.
Radeon 7 Series PR70.

I actually like their current naming scheme. Too bad it'll disappear next generation.
9070 ~ 5070
9070 XT ~ 5070 Ti
9070 XTX ~ 5070 Ti Super?

 

[coercitiv]

9070 is such a weird product. Hopefully it's not so terribly power locked that a lot of the deficit to the XT can at least be clawed back.

If there is stock of the 9070 XT available whatever they put out for vanilla 9070 will just sit on shelves or get shoved into SI units for people who don't know better.

If they're making fewer 9070s than XTs then it's only normal that the non-XT variant is lower value. So far we have reasons to beleive N48 has more frequency potential than 9070 XT can utilize, so one would expect the binning requirements are not that exotic. They might have more premium than base SKUs. I think AMD really struggled with itself to get these prices out, and the main "beneficiary" was the XT. I don't see a problem with allowing the market to balance the price vs. supply for the base SKU, as long as they came out with a good XT price and supply.

Allowing decent power headroom for the 9070 would be the right thing to do though.

 

[SolidQ]

All Powercolor models use Honeywell PTM7950

https://videocardz.com/press-release/powercolor-launches-radeon-rx-9070-xt-red-devil-hellhound-and-reaper-series

 

[inquiss]

I'll complain that 9070 is about $50 too expensive. Because it is. It's 7900XT/7700XT all over again.

Reportedly the 9070 is about 20-25% slower, but it's only about 10% less expensive, which means the 9070 has worse Perf/$ than the card above, AGAIN.

This is just stupid.

You're not supposed to buy the cut option.

 

[gdansk]

I only found 2 9070 XT cards that will fit in my case so far.

 

[adroc_thurston]

I'll complain that 9070 is about $50 too expensive. Because it is. It's 7900XT/7700XT all over again.

Reportedly the 9070 is about 20-25% slower, but it's only about 10% less expensive, which means the 9070 has worse Perf/$ than the card above, AGAIN.

This is just stupid.

No you buy the full part.
That's how it is built.

 

[ToTTenTranz]

For gaming, it's a no-brainer. Nvidia has to drop the price of the 5070, otherwise reviews are going to tear it apart drop the rating from 5 stars to 4.5 stars.

It's still getting the "Editor's Choice" and "Recommended" badges over at TechPowerUp, no matter what.

 

[itsmydamnation]

Ok then how about simply not launching 9070 non XT? if it doesn't exist, you will also buy the full part.

How does that help you win benchmark wars?

 

[adroc_thurston]

Ok then how about simply not launching 9070 non XT? if it doesn't exist, you will also buy the full part.

gotta do something with bins.