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

 

[CastleBravo]

It's very sad. We were supposed to have dubious slides from two companies to compare by now. My disappointment is immeasurable and my whole week is ruined.

Give AMD a week or two to cook up some multi frame gen code so their dubious slides can properly compete with Nvidia's dubious slides.

 

[poke01]

Again, flops are whatever. look elsewhere.
They're getting a ton of perf outta 256b 20Gbps GDDR6 so try to imagine how and why.

It’s one game, one game and it’s cod, ugh. Until I see a HUB or Gamer Nexus review I’ll take that “leak” with a grain of salt.

 

[insertcarehere]

Dunno for the final perf but you ll notice that at 4k Computerbase are quite below 99 FPS with all cards and despite FSR/DLSS enabled.

That being said i have no clue about AMD GPUs exact characteristics, according to Computerbase a single SM has two ALUs and can hence do 4 flops/cycle, at 4096 SP
and 2.8GHz this would yield about 45 Tflops, that s the number that matter for raster perf, so either they got dual issue working or they simplified the design by using two separate ALUs per SM.

Edit : That s their presets :

Frankly COD6 benchmarks just seem to be all over the place, this is what a Chinese website is getting with extreme presets at 4k, no FSR/DLSS:

It's better than Arc margins even with a "huge oof". ~380mm² at $400 and $500.
But yes that's 350 and 500 less than Nvidia. Even if GDDR7 was twice as expensive it can't dent that margin much.

c. 350-380mm^2 on N4 would make the ~7900XTX perf claims more believable and explain all the super bulky 3x 8-pin setups. Dissipating 300++w on a 240mm^2 die would have meant heat density that is way out of proportion with what we've seen before.

 

[poke01]

Interesting statement from dev about AMD RT vs NV RT

It’s not AMD RT vs NV RT. It’s AMD RT vs NV, Intel and Apple RT. AMD is the odd one out.

 

[SolidQ]

AMD is the odd one out.

Isn't he saying in long term AMD is better?

 

[ToTTenTranz]

Resolution? Native or upscaled?

Couldn't tell and there was no info about it. I think it was a 4K screen but it was crisp enough not to notice any aliasing.

Does is look to similar C2077 indoors areas or worse?

Asset quality didn't seem as good. The thing is this was clearly just a tech demo to show that path tracing was being done, and not that graphics can look better.

Can you run a COD6 benchmark there?

I can ask if I go there tomorrow, but I doubt it. IGN seems to have done it without authorization.

Yeah, + need check it's real 9070 or XT

It was a 9070 for sure. Why would they have engineering samples of 2 year old RDNA3 GPUs?

 

[poke01]

Isn't he saying in long term AMD is better?

i don’t agree with anything both those guys say. Also MLID saying Nvidia has large dies to bolt on RT is stupid, other than the halo card their dies are PPA leading.

Do I like Nvidias approach?
No
But I don’t like AMDs approach either.

a middle ground is better but it’s one extreme from both companies.

 

[Heartbreaker]

Give AMD a week or two to cook up some multi frame gen code so their dubious slides can properly compete with Nvidia's dubious slides.

If I was presenting RDNA4 for AMD, I'd be tempted to have the first presentation slide to say:

"Performs like RTX 5090 (*AFMF 6X)"...


Then smile, roll my eyes, shake my head, clear my throat, and get on with the real presentation focusing on actual performance.

 

[blckgrffn]

It's very sad. We were supposed to have dubious slides from two companies to compare by now. My disappointment is immeasurable and my whole week is ruined.

Speak for yourself. We are tick-ticking our way out of my XTX return window!

Speaking of… been playing Tiny Tina Wonderlands and while it ran “fine” with the 6800 the XTX with slightly enhanced settings no diffs it easily. But riddle me this, I have chill enabled along with the quiet profile on (dashboard is claiming like 980mv? Have to looking again) and the power usage didn’t change roping it down to 100 fps from a full on 150fps? I don’t really get that - and it’s not turning my office into a sauna despite reading out like 370TBP. I guess I should start a thread to understand it better.

Running about 70c and quiet - if noticeable - as a system. 200mm intake fans seem to be doing their thing.



¯\_(ツ)_/¯

 

[SolidQ]

It was a 9070 for sure

IGN writing it's 9070 non XT, that what i mean

 

[Hitman928]

Speak for yourself. We are tick-ticking our way out of my XTX return window!

Speaking of… been playing Tiny Tina Wonderlands and while it ran “fine” with the 6800 the XTX with slightly enhanced settings no diffs it easily. But riddle me this, I have chill enabled along with the quiet profile on (dashboard is claiming like 980mv? Have to looking again) and the power usage didn’t change roping it down to 100 fps from a full on 150fps? I don’t really get that - and it’s not turning my office into a sauna despite reading out like 370TBP. I guess I should start a thread to understand it better.

Running about 70c and quiet - if noticeable - as a system. 200mm intake fans seem to be doing their thing.

View attachment 114532

¯\_(ツ)_/¯

What are your idle and peak fps settings for chill? Double check the Tiny Tina game profile doesn't have it disabled (if you enabled it globally).

 

[blckgrffn]

What are your idle and peak fps settings for chill? Double check the Tiny Tina game profile doesn't have it disabled (if you enabled it globally).

It’s definitely throttling down to 100 FPS just doesn’t make wattage difference?

95/100 for the settings.

 

[Hitman928]

It’s definitely throttling down to 100 FPS just doesn’t make wattage difference?

95/100 for the settings.

Try setting chill to 100/150 and then check to see if there is a wattage difference between walking/looking around versus no inputs for several seconds.

 

[SolidQ]

There new info about CoD bench.
Some guy testing own Bench with VRS on/off on XTX
with on - 108 fps
with off - 80 fps
if it's was 9070 non XT at 99fps still very good result, sitting between 7900XT and XTX

 

[marees]

an editor from IGN took the opportunity at the CES to run a Radeon RX 9070 (XT) that was already running at the AMD booth through the integrated benchmark of "Call of Duty: Black Ops 6" installed on the system there. The result was a performance of 99 fps at 4K resolution and the "Extreme" image quality setting and without using FSR. Unfortunately, it was not specified exactly whether a Radeon RX 9070 non-XT or a Radeon RX 9070 XT was in the system, and this may not even have been apparent from the CES system.

	CoD:BO6 @ 4K "Extreme"	(Source)
Radeon RX 9070 (XT)	99 fps	IGN
Radeon RX 7900 XT	102 fps	Notebookcheck
Measurements with the integrated benchmark of Call of Duty: Black Ops 6

The benchmark result is appealing either way, a comparison value from Notebookcheck gives the Radeon RX 7900 XT only 102 fps under the same settings, ergo the Radeon RX 9070 (XT) would be almost on the level of the previous generation - at a presumably much better price range. A more precise assessment can of course not be made as long as it is not clear which 9070 model was tested here. However, it is better not to compare this preliminary benchmark with nVidia results, as all CoD titles are considered to be more AMD-leaning, i.e. it is difficult to make a solid estimate of the performance picture compared to nVidia. In any case, this preliminary benchmark fits much better with a recent performance rumor that sees the Radeon RX 9070 XT almost achieving the performance of the GeForce RTX 4080.

The latter also corresponds to an estimated chip size of the Navi 48 chip (from a CES photo) of ~390mm² . This is noticeably larger than Navi 32 (348mm² including MCDs) and even slightly larger than nVidia's AD103 chip in the GeForce RTX 4080 /Super (379mm²) , which means that performance in the direction of these nVidia graphics cards would be very much to be expected. Another earlier performance leak is therefore becoming increasingly unlikely, as the Radeon RX 9070 XT had to be ranked just above the Radeon RX 7900 GRE. In keeping with this, AMD gave a clear indication in a round table discussion with journalists at CES that the performance rumors about RDNA4 that have been floating around so far are said to be quite inaccurate - possibly related to that earlier leak with the (alleged) 3DMark values for the Radeon RX 9070 XT:

The performance data that's out there on RDNA4 is completely inaccurate.
Source: AMD's David McAfee in a journalist discussion with, among others, More Than Moore on January 6, 2025

The other thing I will tell you is that nobody has the final driver. So how can the data be accurate? That's why I can confidently say that the data is not accurate.
Source: AMD's Frank Azor in a journalist roundtable with, among others, More Than Moore on January 6, 2025

https://m-3dcenter-org.translate.go...mark?_x_tr_sl=auto&_x_tr_tl=en&_x_tr_hl=en-GB

 

[marees]

There new info about CoD bench.
Some guy testing own Bench with VRS on/off on XTX
with on - 108 fps
with off - 80 fps
if it's was 9070 non XT at 99fps still very good result, sitting between 7900XT and XTX

My expectation is 7900xt raster (that would beat the 5070)

The 5070 likely has 4070 ti super level of RT. If they match it then they sell $550 or above

Else $500 or below

 

[SolidQ]

I can ask if I go there tomorrow, but I doubt it.

If you gonna get chance try test with VRS off

 

[Heartbreaker]

There new info about CoD bench.
Some guy testing own Bench with VRS on/off on XTX
with on - 108 fps
with off - 80 fps
if it's was 9070 non XT at 99fps still very good result, sitting between 7900XT and XTX

That might explain the discrepancies we have been seeing in reports on 7900XTX.

 

[del42sa]

If I was presenting RDNA4 for AMD, I'd be tempted to have the first presentation slide to say:

"Performs like RTX 5090 (*AFMF 6X)"...


Then smile, roll my eyes, shake my head, clear my throat, and get on with the real presentation focusing on actual performance.

yep, with Raja´s secret features enabled and unicorn drivers

 

[soresu]

https://m-3dcenter-org.translate.go...mark?_x_tr_sl=auto&_x_tr_tl=en&_x_tr_hl=en-GB

Why translate English to English?

 

[soresu]

and Apple RT

Arguably not for the vast majority of current HW RT use, even when you include professionals in DCC.

Even if Apple RT is better it doesn't make a difference due to games availability on their own platforms vs those covered by AMD, nV and Intel, which collectively cover all the current game consoles, Windows and Linux/SteamOS.

 

[gdansk]

Why translate English to English?

Check again. It was a German article summarizing it. Usually people post a comment with it but...

 

[soresu]

Check again. It was a German article summarizing it. Usually people post a comment with it but...

Where Was RDNA4 at AMD’s Keynote?

Timing is Everything

morethanmoore.substack.com

 

[soresu]

Seems like the AMD responses imply that FSR4 is still open for other GPUs to use going by this:

Tim from Hardware Unboxed asked about an assumption based on a footnote that might have suggested that FSR4 was going to be 9070 exclusive (or minimum) and asked for clarity. David stated that there’s nothing about FSR4 that makes it exclusive to a specific model - the RDNA4 updates include better MLOps that makes technology like FSR4 a lot better than it would have been.

 

[gdansk]

Where Was RDNA4 at AMD’s Keynote?

Timing is Everything

morethanmoore.substack.com

The whole post is translated from German except the quotations. 🤔