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

 

[blackangus]

Can't use MST?

That wont swing 4k @ 144 hz right for all 3 right?

 

[soresu]

That wont swing 4k @ 144 hz right for all 3 right?

If the displays supported DP 2.0 and DSC it should, though I doubt many displays support it yet.

 

[blackangus]

If the displays supported DP 2.0 and DSC it should, though I doubt many displays support it yet.

Well mine are DP 2.0 but I thought the display engine on 9070 XT was 13.5Gbps not 20.
Im not sure if the display can manage DSC tho, I would think it can - Gigabyte MU32's

 

[SolidQ]

will games like Kingdom come 2 improve with RDNA4?

That form AMD site. Native vs UPS + FG

 

[soresu]

Well mine are DP 2.0 but I thought the display engine on 9070 XT was 13.5Gbps not 20.

As I said if they support DSC too it should be possible.

UHBR13.5 supports up to 4K187p at 10bpc color, and DSC is supposed to give a 3:1 compression ratio.

 

[marees]

So what are the chances N44 matches the 7700XT?

9070 has the same CU count as the 7700XT and looks to be 45% faster or there abouts in 1440p.

9060XT with 32CUs and a 45% uplift over the 7600XT would be in the same ballpark as 7700XT if not slightly ahead. Bandwidth will probably limit it a bit though but certainly looks as though it can be in the region of 7700XT.

To me this means AMD really do just need to make a 9060XT 16GB and they can probably sell it for $330 as a direct replacement for the 7600XT.

I suspect N44 is sub 200mm so with that in mind I would suspect yields are amazing and I wouldn't bother with a cut part. I wouldn't even bother with an 8GB variant if performance is 7700XT tier because at 1440p it will absolutely need more than 8GB and the performance would be there for that.

Compared to what I suspect will be a fake $450 MSRP for the 5060 Ti 16GB the 9060XT would hold up very well.

Good.

computerbase.de has the 7700XT at only ~35% faster than the 7600XT in FullHD, and in most cases that's gonna be the resolution the 9060XT will play games at (at least the internal resolution before upscaling).
35-45% faster than the 7600XT should be absolutely doable, judging by what we've seen so far.
The 5060Ti might actually struggle to beat the 9060XT in native raster.

If that's the case then AMD should launch a 16GB clamshell Navi 44 alongside the 8GB one.
The 7700XT is a decent 1440p card and and 16GB Navi 44 could be a good product. (maybe name it 9060XTX or just name it 9060XT and put the 8GB model as 9060.... then the cut down version if any, can be 9050 series)

N44 is tiny so I don't see the point in an 8GB or a cut version at all if it is 7700XT or higher performance, it will be received badly unless it is sub $250 and then the reduction in price Vs a 16GB variant will be greater than the ram cost.

They can keep pumping out N33 and drop the price of the 7600 if they want a low end product to replace the 6600 for the $200 crowd and that won't use N4 wafers either.

While I agree with all this, in 2025, an xx60 class card shouldn't be 8gb. Hence AMD should name them as xx50 cards (just like we had 5500xt 4gb & 5500xt 8gb, we can have 9050xt 8gb & 9050xt 16gb)

 

[GTracing]

Hmm seems Charlie got upset https://www.semiaccurate.com/2025/02/28/amds-radeon-9070-isnt-very-good/

What's the fastest turnaround time for Charlie making a fool of himself? I have a feeling this will be in the top 5. His main argument is AMD used upscaling to cheat in all of their benchmarks. All his other claims about performance and efficiency stem from that.

If he put one ounce of thought into it, he'd realize that it's not possible for AMD to have used upscaling in every benchmark. Several of the games don't support FSR4.

 

[Timorous]

While I agree with all this, in 2025, an xx60 class card shouldn't be 8gb. Hence AMD should name them as xx50 cards (just like we had 5500xt 4gb & 5500xt 8gb, we can have 9050xt 8gb & 9050xt 16gb)

I don't think naming matters too much. The point really is that N44 will have very good yields and pairing it with 8GB of Vram caps the price at $250 if they don't want to be absolutely torn to shreds in reviews. With such good yields a cut down die seems kind of pointless so to maximise their profit it would be sensible to pair it with 16GB of Vram as then they can sell it for around $330 which is a larger uplift than the cost of the Vram.

As for a cut variant, seems a bit pointless at launch, maybe way down the road as they build up enough stock of faulty dies and then they could do a 9050 XT with 24CUs and 12GB or VRAM on a 96bit bus for $250 or something. That would probably be better than an 8GB card with a 128 bit bus in 2025 and they get to use dies with faulty CUs or faulty memory controllers so there will be a bit more supply.

 

[Novacius]

Good.

computerbase.de has the 7700XT at only ~35% faster than the 7600XT in FullHD, and in most cases that's gonna be the resolution the 9060XT will play games at (at least the internal resolution before upscaling).
35-45% faster than the 7600XT should be absolutely doable, judging by what we've seen so far.
The 5060Ti might actually struggle to beat the 9060XT in native raster.

Took only two generations to turn my 6800 into a 1080p card, that hurts

What's the fastest turnaround time for Charlie making a fool of himself? I have a feeling this will be in the top 5. His main argument is AMD used upscaling to cheat in all of their benchmarks. All his other claims about performance and efficiency stem from that.

If he put one ounce of thought into it, he'd realize that it's not possible for AMD to have used upscaling in every benchmark. Several of the games don't support FSR4.

He makes so many wrong claims I don't even know where to start. He says the 9070 XT is slower than the 7900 XT (wrong), the latter has 50% more memory (also wrong) and it pulls 4W more (you guessed it, wrong), all in one sentence. Was he drunk when he wrote that?

 

[SolidQ]

His main argument is AMD used upscaling to cheat in all of their benchmarks

That strange, he can go on official AMD site and look for Native results

 

[GTracing]

Took only two generations to turn my 6800 into a 1080p card, that hurts

He makes so many wrong claims I don't even know where to start. He says the 9070 XT is slower than the 7900 XT (wrong), the latter has 50% more memory (also wrong) and it pulls 4W more (you guessed it, wrong), all in one sentence. Was he drunk when he wrote that?

The 7900 XT has a 300W TDP and the 9070 XT has a 304W TDP. Real world power draw will have to be seen.

He definitely got the performance and memory comparisons wrong though.

 

[blackangus]

As I said if they support DSC too it should be possible.

UHBR13.5 supports up to 4K187p at 10bpc color, and DSC is supposed to give a 3:1 compression ratio.

Thanks! I had it in my head that that would fall short.

 

[Novacius]

The 7900 XT has a 300W TDP and the 9070 XT has a 304W TDP. Real world power draw will have to be seen.

He definitely got the performance and memory comparisons wrong though.

The 7900 XT has a 315W TDP

 

[SolidQ]

for Spiderman RT results is very high.
Fabio was mention he's suprise result in Spider Man 2 even vs 5080

 

[GTracing]

The 7900 XT has a 315W TDP

Huh, it was announced at 300W and several sites say the TDP is 300W. The official specs on AMDs website say the TBP is 315W.

 

[marees]

Took only two generations to turn my 6800 into a 1080p card, that hurts

N44 will not beat the 6800 is my guesstimate

 

[Novacius]

Huh, it was announced at 300W and several sites say the TDP is 300W. The official specs on AMDs website say the TBP is 315W.

Yeah, it was raised after the announcement

N44 will not beat the 6800 is my guesstimate

I think it's possible, at least in 1080p. The 6800 is around 35% faster than the 7600 XT, and N44 has the same amount of CUs but 40% faster CUs thanks to RDNA4. For higher resolutions the 6800 has a big cache and bandwidth advantage though.

 

[Heartbreaker]

anticompetitive practices...paying devs to block competitors products

Generally once games are out, they stick with the systems they have. NVidia is paying them to upgrade NVidia systems, but you really can't expect NVidia to pay to upgrade FSR versions.

I'm sure AMD can fund an FSR upgrade if they want...

 

[Timorous]

N44 will not beat the 6800 is my guesstimate

6800 and 7700XT are very similar at 1080p and 1440p and the 7700XT has less cache and less bandwidth.

I think at 1080p N44 can be faster than the 6800, at 1440p it may be similar, if it has 16GB of Vram.

It is not a ton faster than the 4060Ti which has 288GB/s of bandwidth so give N44 19Gbps/20Gbps ram for 304GB/s / 320GB/s of bandwidth and it should be enough to be in that range at 1440p.

 

[SolidQ]

if it has 16GB of Vram.

That is big question, will they just stay with like 7600XT or 8/16GB

 

[Timorous]

That is big question, will they just stay with like 7600XT or 8/16GB

As I have argued already I don't see the point in an 8GB version if it can land around 7700XT performance, it would make it too compromised to command the kind of price that a 16GB version could.

If they did make a cut / alternative version I think a 12GB 96bit SKU would be a far better fit for 1080p gaming than an 8GB 128bit SKU.

 

[uzzi38]

FSR4 has fallback mode for non-90 series, so technically FSR4 works with other cards.

I take it y'all missed this slide then, if you're surprised by stuff like this?

(Look on the right hand side).

 

[Kepler_L2]

I take it y'all missed this slide then, if you're surprised by stuff like this?

(Look on the right hand side).

So it does work on RDNA3? Why did they say it was exclusive to RDNA4?

 

[uzzi38]

So it does work on RDNA3? Why did they say it was exclusive to RDNA4?

RDNA3 is likely going to get access a different model (RDNA3 likely doesn't have the brunt for full FSR4) that's launching separately later on, and the screenshot is done by mistake. But what it does show, is it looks like AMD has it working to some extent behind the scenes already.

 

[SolidQ]

is it looks like AMD has it working to some extent behind the scenes already.

I was posted screenshot from Fabio, in this topic