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

 

[jpiniero]

drop the Radeon name then. Call it something else. Worked for lots of companies

AIdeon?

 

[ToTTenTranz]

I suspect, that this should work. But I expect the offloading only for the DNN part, not FSR4 as a whole. 50 TFLOPS Block-FP16 compared to whatever FP8 TFLOPS of N44/N48 are looking good enough for an iGPU.

I mean, 50 TFLOPS are much more than what e.g. Strix Point's iGPU can deliver (max. 25 TFLOPS FP16 WMMA). And offload compute load from the already stressed iGPU.

AND: It would be a massive uplift and selling point improvement compared mobile RTX40/50, which all get DLSS4.

It's not about how fast the NPU is. For the Phoenix, Hawk Point and Strix Point NPUs that will target 1080p at most, the NPUs' throughput is probably fine.

Requiring the data to go back and forth between GPU caches, System RAM and NPU caches might induce too much latency to make it worthwhile. At that point it might be faster to run it all in the compute shaders.

 

[Novacius]

drop the Radeon name then. Call it something else. Worked for lots of companies

Adreno? 🤔

 

[Win2012R2]

"Should have would have" is not how you justify running a pricey program in front of dr. Lisa Su.

Do you think she regrets her decision to can the big chip?

 

[Meteor Late]

If Top Dog is the only thing that matters, then Lisa Su is directly responsible for this failure, She was the one to cancel "the only thing that matters". Then this means if RDNA4 fails, it's because of her.

 

[tsamolotoff]

You imply that most people are just stupid.

Most people are just buying 'a computer', and the integrators are not interested in providing the best bang for the buck for their customers, they just want less RMAs and hassle with tech support (which is naturally easier with nVidia offerings as they support the CURRENT_THING thanks to nV hands into AAA dev processes everywhere). Also, as this CURRENT_THING is a software component, nVidia can just adjust the metrics each time AMD is threatening to get close (tesselation, hairworks, frametime pacing, encoding, RTRT, hallucinated frames, now 'neural shaders' and whatnot) and win with 100% guarantee. I personally believe the decision to propel RTX and various forms of TAA abominations into the industry standards was a big loss long-term for AMD

 

[Abwx]

I have some bad news for you, ATi tried all this with RV700s and Evergreens and they lost horribly. I get it, you're a zoomer and missed out on all the oldskool GPU wars.
No one buys Radeon.

Well, since you mentioned this gen :

Demand so greatly outweighed supply that more than two months after launch, many online retailers were still having trouble keeping the 5800 and 5900 series in stock.

Radeon HD 5000 series - Wikipedia

en.wikipedia.org

There were so few who did buy Radeon that retailers were out of stock, sound logical, isnt it...

 

[maddie]

Well, since you mentioned this gen :

Radeon HD 5000 series - Wikipedia

en.wikipedia.org

There were so few who did buy Radeon that retailers were out of stock, sound logical, isnt it...

Yep.

The difference with Zen 1 & 2 and the 4xxx & 5xxx GPUs was that Zen improved again with Zen 3, but Radeon stalled with the 3rd iteration 6xxx cards.

 

[gaav87]

9070XT on 5070Ti level for 600$ would already deliver a quite attractive Perf/$ ratio. AMD is not the Salvation Army, but they need to deliver 1.2x or higher Perf/$ ratio in order to convince people to buy AMD. Anything lower and people will still buy Nvidia cards.
600$ MSRP should also be maintainable for AMD. The 7800XT was released at 500$, with same VRAM capacity and similar TDP. N48 will probably be a notch more expensive to produce than N32 (assuming 330...350mm2 Die size vs. 200 + 4*37.5mm + organic RDL packaging) but I think not by too much.

But it will stand or fall with FSR4. DLSS4 has improved by so much regarding image clarity and stability, that Transformer-DLSS-P is quite usable even at 1440p. 4K Transformer-DLSS-P looks very good and in my opinion better than CNN-DLSS-Q and in some regards even CNN-DLAA (mainly image clarity and ghosting/smearing, not so much regarding noise and flickering).
- If FSR4 achieves DLSS3 quality levels (Super Resolution part), it will already be a great improvement compared to FSR2/3. But the competitor is DLSS4
- Ray reconstruction should be there as well. I am optimistic here, because AMD has already teasered that a few times
- MFG is not that important, but could be the icing on the cake. 5070 Ti / 9070 performance should often be sufficient to use FG, especially at 1440p. If you have FG, MFG should not be a too far step from it (frame pacing and image quality are getting more important with MFG, that is clear)
- RDNA4 differentiation possibilites vs. Blackwell: FG Extrapolation (no latency increase), RSR @ AutoSR, AFMF3 @ ML/AI/Extrapolation

If FSR4 comes close to DLSS4 regarding quality, RDNA4 could be a banger. And if the last point (differentiation features) gets delivered as well, it could even surpass Blackwell regarding gaming relevant feature set. Unseen from AMD.

But: It's optimistic, I know.

That's a lot of words and none of them mean anything.
People won't buy Radeon, simple as.
It's a good product that is worthless without a 3 kilobucked tiled chop chop on top of it.

Sorry but adroc is right. I talked with a lot of ppl. I asked, a guy yesterday (before igor's lab leak of 8% 5080). If he would buy 9070xt if it were only 10% behind 5080 and 2x cheaper he said NO. There are many many ppl like this for them radeon does not exist even 2x cheaper with 10% less perf.

 

[SolidQ]

There are many many ppl like this for them radeon does not exist even 2x cheaper with 10% less perf.

Mine is difference story, most friends only take price/perf ratio, doesn't matter green/red sticker, and mostly of them 90% have Red card

 

[gaav87]

They should bring back naming
ATI 9070XT

Mine is difference story, most friends only take price/perf ratio, doesn't matter green/red sticker, and mostly of them 90% have Red card

Nah i even considered 5080 few days ago before i heard 5080 perf. "fuxxTAA" reddit youtubers like LTT (like look at his yt thumbnail for 5090) too big mind-share. Ppl see dlss4 performance better than native with increased perf and only see what they want to see. Even if its not tottaly true the ghosting is still there but reduced some games look worse w/o ray reconstruction there is still shimmer.
Hope new ai radeon sharpening + fsr4 will, be good.

 

[Tuna-Fish]

Yep.

The difference with Zen 1 & 2 and the 4xxx & 5xxx GPUs was that Zen improved again with Zen 3, but Radeon stalled with the 3rd iteration 6xxx cards.

Also NV got their lineup in gear. 5xxx competed first against the antiquated GTX 2xx series, and then NV sort of botched the launch of the 4xx series. The GTX 480 had such low yields that it was largely not available to purchase, even if you wanted it. The GTX 5xx series was sort of just the 4xx series, but it finally worked, and people ate that up.

 

[SolidQ]

AMD posted this

 

[adroc_thurston]

You imply that most people are just stupid.

Well yeah, RTX3050 outsold RX6600 by like 7 to 1 at least.

Do you think she regrets her decision to can the big chip?

probably lmao. Lisa loves winning and they could've won by a lot here.

There were so few who did buy Radeon that retailers were out of stock, sound logical, isnt it...

you can always use JPR mss data instead of dropping anecdotes.

 

[Thunder 57]

Well yeah, RTX3050 outsold RX6600 by like 7 to 1 at least.

To bad you cant use a mallet to knock some sense into those idiots.

 

[madtronik]

Well yeah, RTX3050 outsold RX6600 by like 7 to 1 at least.

probably lmao. Lisa loves winning and they could've won by a lot here.

you can always use JPR mss data instead of dropping anecdotes.

When GPU prices started coming down to earth after the latest crypto craze I set out to buy a new GPU. For the money I was willing to spend I could get an RTX 3050 or a RX 6700 (also a 1660 Ti, LOL). Guess which I bought.
It is obvious that if NVIDIA is able to sell for the same price a card with 35% lower performance you need a strong cult of retards buying your products.

 

[gaav87]

AMD posted this

Strix halo 64gb will murder nvidia

 

[Abwx]

you can always use JPR mss data instead of dropping anecdotes.

Lol, it s you who are dropping hasardous sentences in the hope that it could convince the crowd, and to better look serious throw JPR name , it can always fool whoever is not aware of this tactic, actually youself are not even aware of his numbers, you are just gambling that you could be right.

The data for mobile discrete graphics reveals a 36.5% market share gain, and 87.27% more units sold quarter-over-quarter through the first six months of 2009. AMD now holds the "Number 1" position with an overall 53% mobile discrete market share.

AMD Takes Notebook Discrete Graphics Market Share Lead

AMD today announced that according to the latest industry data, notebook manufacturers have chosen ATI Mobility Radeon Premium graphics processors to drive more than half of today's discrete graphics notebooks. The data for mobile discrete graphics reveals a 36.5% market share gain, and 87.27%...

www.techpowerup.com

 

[adroc_thurston]

Lol, it s you who are dropping hasardous sentences in the hope that it could convince the crowd, and to better look serious throw JPR name , it can always fool whoever is not aware of this tactic, actually youself are not even aware of his numbers, you are just gambling that you could be right.

Those are words.
You can always post total and AIC stats for JPR, it's not that hard.
The highest Evergreen got was 46 or 47% total dgfx share, which is really dire given they stomped the entire NV product stack into the ground.

 

[Abwx]

Those are words.
You can always post total and AIC stats for JPR, it's not that hard.
The highest Evergreen got was 46 or 47% total dgfx share, which is really dire given they stomped the entire NV product stack into the ground.

Lol, you are inverting the reality, you throwed non substancied words and i answeredwith hard facts, live with it rather than taking the denial route with yet another flow of unsubstancied claims, indeed you are making tons of exagerated statements but on this one you were caught parroting the usual urban legends we heard here and there.

And btw see here how much better the 3050 sell compared to he 6600, at least in Germany and likely in the rest of Europe, reality is that there s nothing better than urban legends to prop up a firm.

https://twitter.com/x/status/1883180643091161472

 

[adroc_thurston]

Lol, you are inverting the reality, you throwed non substancied words and i answered
with hard facts, live with it rather than taking the denial route with yet another flow of unsubstancied claims, indeed you are making tons of exagerated statements but on this one you were caught parroting the usual urban legends we heard here and there.

You've posted mobility share, which is neither total dGFX share nor AIC share.

 

[StinkyPinky]

AIdeon?

GeRadeon

 

[Abwx]

You've posted mobility share, which is neither total dGFX share nor AIC share.

Doesnt matter, currently Nvidia has the lead mainly because of mobile, so for Nvidia it count but not for AMD, and while i m here here a hint.

Say i sell 5 GPUs to OEMs for 500$/unity, that makes me 2500$.

On the the other hand i sell 5 FE dGPU cards for 1000$ each using those same GPUs, that makes 5000$ and a total revenue of 7500$ for 10 GPUs and i state that this is my GPU sales.

My competitor sell very few cards and mainly 250$ GPUs to OEMs, and he sell say 8 of those GPUs for 2000$ total sales.

Now all half brained will do the math 2000/(2000 + 7500) = 0.21 and conclude that the competitor hold only 21% of the DT GPU marketshare while the reality is 8/18 = 44.4%.

That s the foolish logic of people who compare AMD s 0.462bn GPU sales to Nvidia s 3bn as they completely obfuscate that most of the latter s revenue come from whole cards and not only individual GPUs, and indeed we can read here and there that Nvidia has allegedly as much as 90% DT marketshare, when you read this number somewhere you know that you are facing a simple mind.

 

[adroc_thurston]

Doesnt matter, currently Nvidia has the lead mainly because of mobile, so for Nvidia it count but not for AMD, and whle i m here here a hint.

The mindshare is forged in AIC sales.
It's the thermostat.

 

[Abwx]

The mindshare is forged in AIC sales.
It's the thermostat.

Mindshare if formed by constant propaganda, viral marketing and lies by the truckload, otherwise a firm that sold dozen millions faulty chips that busted in as much laptops wouldnt be leader these days.