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

del42sa · Jun 19, 2025

https://videocardz.com/newz/asrock-...-12v-2x6-connector-and-game-clock-at-2350-mhz

who wants 32GB RAM ?

jpiniero · Jun 19, 2025

beginner99 said:
Prices have been coming down and supply for 9070 (xt) cards is now readily available it seems here in Europe. also 5070 tis seems readily available in fact there is "only" a roughly $90 difference between the cheapest 9070 xt and 5070 ti.

Not really in the US. As for the 9070 XT, anything near the Real MSRP ($699) goes by fast but there's stuff in the $750 range. 5070 Ti is more like $899 to start.

eek2121 · Jun 20, 2025

jpiniero said:
Not really in the US. As for the 9070 XT, anything near the Real MSRP ($699) goes by fast but there's stuff in the $750 range. 5070 Ti is more like $899 to start.

Only for eCommerce. I can easily get a 9070XT for $699 if I so desired. The closest MicroCenter has them (3 hours away) and my local best buy likely has a few.

EDIT: Best Buy has them for $659.

Abwx · Monday at 1:02 PM

Test at Computerbase, RDNA 3 vs RDNA 4, 20% better perf/Hz in raster and 31% in RT
for the latter.

Wie viel mehr Performance bringt die neue GPU-Architektur?

Die aktuellen Grafikkarten von AMD und Nvidia haben eine neue Architektur. Ob die Shader mehr Performance pro Takt liefern, zeigt der Test.

www.computerbase.de

maddie · Monday at 2:06 PM

Abwx said:
Test at Computerbase, RDNA 3 vs RDNA 4, 20% better perf/Hz in raster and 31% in RT
for the latter.

Wie viel mehr Performance bringt die neue GPU-Architektur?

Die aktuellen Grafikkarten von AMD und Nvidia haben eine neue Architektur. Ob die Shader mehr Performance pro Takt liefern, zeigt der Test.

www.computerbase.de

Anywhere talks about PPT? Performance /transistor.

Per CU is all well and good, but if the size inflates faster than the performance, then it's a losing proposition, as graphics workloads are massively parallel, unlike CPU workloads, and can use more lower performing cores.

Room for some investigative articles.

adroc_thurston · Monday at 2:13 PM

maddie said:
Anywhere talks about PPT? Performance /transistor.

irrelevant metric.

maddie said:
Per CU is all well and good, but if the size inflates faster than the performance

All that matters is perf/power/area.
But shader core-level perf scaling is also very very nice for workloads that don't neatly fit into super chungus shader arrays.

maddie said:
and can use more lower performing cores.

not really.

maddie · Monday at 3:23 PM

adroc_thurston said:
irrelevant metric.

So you say.

adroc_thurston said:
But shader core-level perf scaling is also very very nice for workloads that don't neatly fit into super chungus shader arrays.

Not really a gaming problem, which is the RDNA part of this thread..

adroc_thurston said:
not really.

Right. Totally accurate when seeing the increase in core count over the years.

basix · Monday at 3:32 PM

adroc_thurston said:
not really.

It is not wrong, if compared to CPUs But a CU will never approach a CPU core. Just a very different use case and focus (throughput vs. latency // massively parallel vs. primarily serial)

Having strong per CU/SM performance helps regarding scaling, for sure. We see it on 4090 and 5090, which get poorly utilized at 1080p (also besides CPU limits). Also compared to their smaller siblings like 4080/5080, even at 4K their performance does not scale too well considering the SM count. The only thing which still scales very well on this big GPUs is pathtracing or even higher resolutions than 4K.

DAPUNISHER · Monday at 3:35 PM

Iceberg uploaded 7600XT vs 9060XT. Good to see he is doing more in game testing again. Serious generational improvements. When he underclocks RDNA 4 it still wins while using approximately 125W.

adroc_thurston · Monday at 3:46 PM

maddie said:
So you say.

It's the truth.

maddie said:
Not really a gaming problem, which is the RDNA part of this thread..

Very much a gaming problem unless you think gaming is exclusively pixel shader waves at 4k.

maddie said:
Totally accurate when seeing the increase in core count over the years.

You're seeing much beefier shader cores than ever before.

techjunkie123 · Monday at 4:02 PM

adroc_thurston said:
You're seeing much beefier shader cores than ever before.

Would you say they're the biggest, most beautiful shader cores ever designed?
😍

itsmydamnation · Monday at 5:16 PM

maddie said:
So you say.

Not really a gaming problem, which is the RDNA part of this thread..

Right. Totally accurate when seeing the increase in core count over the years.

Amd had problems back in the gcn days maintaining shader utilisation as they increased shader count. The complexity and size of the shader code can also be a factor. But also gcn needed 4 times the concurrency at a given shader count then Maxwell/parcel/etc and it hurt gcn performance scaling.

Abwx · Monday at 5:19 PM

maddie said:
Anywhere talks about PPT? Performance /transistor.

Per CU is all well and good, but if the size inflates faster than the performance, then it's a losing proposition, as graphics workloads are massively parallel, unlike CPU workloads, and can use more lower performing cores.

Room for some investigative articles.

In raster at 1080p with 53% better perf the 9060XT consume the same 160-165W as the 7600 8GB, so a 8GB 9060@70W should largely match a stock 7600.

AMD Radeon RX 9060 XT im Test: Benchmarks in WQHD sowie Full HD und die Taktraten

AMD Radeon RX 9060 XT im Test: Benchmarks in WQHD sowie Full HD und die Taktraten / Performancerating mit und ohne RT in 2.560 × 1.440

www.computerbase.de

marees · Thursday at 4:10 AM

This review came out 2 weeks back, for those who missed it

AMD Radeon RX 9060 XT 16GB review: plenty of performance with 16GB

Be wary of the 8GB models, which are a completely different ballgame.

Reviews
By The Staff of Tom's Hardware published June 10, 2025

Along with AMD’s RDNA 4 GPUs, we’ve tested all the initial onslaught of Nvidia Blackwell GPUs, plus Intel’s two (for now) Arc Battlemage GPUs. That doesn’t mean there won’t be mid-cycle refresh GPUs like last generation's RTX 40-series Super cards, but those generally slot into the existing structure without shaking things up too much. If those cards do happen — and they probably will — they’re likely to arrive in early-to-mid 2026.

Many have noticed that our standalone RTX 5060 review didn’t happen, and there are good reasons for that. Jarred has left the building, and while he’s helped us put together this one final hurrah, timing and a lack of direct sampling from Nvidia delayed the testing of that card. So, today’s review marks a two-for-one special of sorts — three-for-one if you want to also count the lackluster RTX 5060 Ti 8GB, which we also had to purchase for testing.

AMD Radeon RX 9060 XT 16GB review: plenty of performance with 16GB

Be wary of the 8GB models, which are a completely different ballgame.

www.tomshardware.com

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Member

Lifer

Diamond Member

Lifer

Attachments

Diamond Member

Diamond Member

Diamond Member

Member

Super Moderator CPU Forum Mod and Elite Member

Diamond Member

Member

Diamond Member

Lifer

Golden Member

AMD Radeon RX 9060 XT 16GB review: plenty of performance with 16GB​

Be wary of the 8GB models, which are a completely different ballgame.​

AMD Radeon RX 9060 XT 16GB review: plenty of performance with 16GB

Be wary of the 8GB models, which are a completely different ballgame.