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

 

[MrPickins]

It is.
All the exotica remains contained to HPC parts like H100 or MI300.

Who is even talking about HPC?

I'm talking about current gen midrange consumer GPUs, which by all accounts have had their branding bumped up to the next tier without a corresponding increase in performance.

Yea whatever, you're still gonna buy it.

Considering I'm still using an RX 580 while I wait for an upgrade sweet-spot from either company, I'd say you're wrong.

But if this is how you want to discuss things, I'm not going to bother to respond to you. Have a nice day.

 

[MrPickins]

You saying that, doesn't make it so. Both AMD and NVidia had smaller gains across the board this generation, and they still both have corresponding x060 and x600 competing with each other.

It's not just me saying it. It's the major consensus in the 4060 reviews thread, and the majority of the reviews/commentary I've seen on the 4060 and 7600 releases. Hell, they barely beat last gen's 6 series cards (in VRAM heavy games, the 4060 8GB can even be worse than the 3060 12GB...).

These are 5 series cards being passed off as 6 series. You may be fine with the marketing shenanigans from both vendors, but I'm not.

Regardless, you're not even addressing my original point. I'd be happy with a Polaris-type release (very competitive in the midrange, great bang for the buck, and enough VRAM to make it somewhat futureproof). The 7600 is clearly not that, so bringing it up is a moot point.

 

[PJVol]

Best to just ignore those fake noodz guyz.

So, basically no one knows anything about RDNA4, other than some mysterious navi4c has been canceled?

Yeah, unfortunately those who silently knows and aren't used to talk a lot remain silent, unlike so-called YT "leakers", whom they like to catch in lies or mispredictions, however, after the product launch.

 

[Tuna-Fish]

RX7700XT?

There is just no way, no how they are going to launch a new set of cards whose names are so hard to distinguish from the old ones. Plenty of people in this thread are very misguided about what drives naming. It's not architecture, mm², process node, common sense, or any of that. It's that AIBs and especially laptop OEMs want to release a new lineup of gear every year, with new marketing names on it that are clearly visually "better" than what was sold last year. What they absolutely do not want, is to have sold boxes "7700" printed on the front this year and have to sell ones that have "7700 XT" next year. "8700" is just so much better, see, it has a bigger number!

If AMD keeps some RDNA3 cards in the lineup, the solution isn't naming the new cards to fit the old lineup, it's to rebadge old cards to fit the new lineup. That way, there's no trouble fitting anything anywhere, you can just decide what models you keep and give them appropriate numbers. No connection or logic to the old lineup is required, see for example Pitcairn, which was sold as the: { HD 7870 XT, HD 8870, R9 270X, R9 370X }, and that's just the top cuts of it in each stack.

And yes, they absolutely can throw out "7800 XT" from their active lineup after not even a year, either in reality or just in name, and it wouldn't even be their shortest-lived model.

To be clear, I have no specific information about RDNA4, but I can honestly tell you that marketing execs would recoil from your proposal like a vampire recoils from sunlight.

 

[TESKATLIPOKA]

Guys, here is the final table of speculation of upcoming RX7600XT; once you read through the specs you should have better understanding of why AMD wants to hide the codenames:-

	RX 7600	RX 7600XT	RX 7600XT OC	RTX 4060
Current Price	$269			$299
Launching Price	~ $199 ?	$249 - $269 ?	$299 - $329 ?	?
GPU	RDNA3 N33	RDNA4 N44	RDNA4 N44	AD107
Process Node	N6	N4P	N4P	4N
Die Size	204 mm2	~ 160 mm2	~ 160 mm2	159 mm2
CU / SM	32	32 ?	32 ?	24
Base Clock	1720 MHz	?	?	1830 MHz
Boost Clock	2655 MHz	Lower	?	2460 MHz
FP32	21.75 TF	Lower	?	15.11 TF
Memory Size	8GB GDDR6	16GB GDDR6	16GB GDDR6	8GB GDDR6
Memory Clock	18 Gbps	18 Gbps ?	?	17 Gbps
Memory BW	288 GB/s	288 GB/s	?	272 GB/s
TDP	165 W	~ 130 W	~ 165 W	115 W
TPU Perf	Base	~ +5%	~ +10%	+ 0%
RT Perf	Base	~ +38%	~ +43%	+ 119%

• Basically, RDNA4 is a die shrink of RDNA3 (more likely N32), but due to N4P process and AMD's intention to lower the TDP to save PCB/VRM/MOSFET costs, the clock speed of the GPU would be lower than N33. You can compare to RTX4060's TDP with similar process. That's why AMD has to upgrade the graphics engine to make it more powerful so that 7600XT will perform better than 7600. In CPU term, we refer that as IPC, in GPU term....well that one have to ask GPU specialist
• That's why we should expect two versions of 7600XT with 16GB at launch, standard version with lower TDP (maybe single fan like RTX4060) and OC version which should use same PCB as current 7600.
• Thanks to @adroc_thurston early leaking about how cheap GDDR6 would be, I should expect N44 to come with 16GB not 8GB. My bad , with lower BOM and cheap GDDR6, AMD would most likely offer RX7600XT with 16GB at current RX7600's pricing. We should expect RX7600 to drop to $199 price point also. Now I know what the new Polaris meant for N44: with lower TDP, doubling RAM and lower price point; this is AMD's once again attack at nVidia's RTX4060.
• Don't expect much faster rasterization performance but at least they would be slightly faster than RTX4060.
• Ray tracing wise: 7600XT should perform 40% better than RX7600, unfortunately still far slower than RTX4060, I wish I am wrong on these figures but if you click the link above, you will know why...NV really shines on RT's performance...

1.) RX 7600 8GB -> release price is $269, current price is likely lower.
2.) RX 7600XT at $249-269 is already very expensive compared to RX 7600 8GB at $199, when you get only 8GB extra Vram and 5% extra performance.
3.) RX 7600XT OC is unsellable at that price.
4.) 160mm2 at N4P vs 204mm2 N6. It's clear which one is cheaper -> N6. There is no reason making a similarly performing chip based on RDNA4, which is costlier to make.
5.) 130W vs 165W is only 21% lower, there is no reason for lower boost If It was using a better process.
6.) Your expected RT performance is based on nothing.
7.) Your claim about RX 7600XT using RDNA4 is based on nothing.
8.) RX 7600XT using a better process is also based on nothing.

Thanks to reading your table, I know It's just pure speculation with many contradicting points and without any basis, just your expectations.
Yet you make It sound like It's a fact or you know some insider info.

 

[Saylick]

lol no.

AMD's best next gen will beat NVIDIA's best this gen. RNDA4 provides a solid uplift. Looking forward to the release.

EDIT: Also, no, you can't compare memory bandwidth numbers cross-gen due to architectural differences, and RDNA3 wasn't bandwidth limited anyway. RNDA4 won't be a miracle, but it will be better than RDNA3 in every way. There is a reason they elected not to go bigger, and it wasn't just $$$.

I find it hard to believe that N44 will beat a RTX 4090, unless I misunderstood what you meant. Did you mean that a hypothetical big RDNA 4 will beat AD102?

 

[linkgoron]

it didn't blow up in anyone's face.
They just focus on raw execution.

Obviously they had to lie about Navi 31's performance, had to release Navi 32 extremely late, cancel Navi 41/42 and will probably not release any chiplet based consumer GPU until at least 2026 because it's such a success.

 

[TESKATLIPOKA]

They'll just increase effective bandwidth.
I think AMD and NVidia calculate it like that:
(LLC bandwidth * LLC Hit rate) + VRAM speed

I think that effective BW is just marketing BS.

Averaging It is just nonsense, either you have the data in cache or you don't.
If you have It, then you have the maximum ~1940 GB/s BW in case of RDNA2 N21 or ~4470 5300 GB/s for RDNA3 N31.
If you don't then It leaves you with only GDDR6/7 BW.

P.S. I wonder how much IC would be needed for 90% hitrate at 4K. 1GB?
I think 4 stacks of Hynix HBM3E with 4TB/s(1TB/s per Stack) and 64-96GB Vram(16-24GB per Stack) could end up cheaper to make.

Edit: It looks like IC size matters for total BW.
I think what Locuza wrote as theoretical Iinfinity Cache BW is wrong for N23/24. N22 is 0.75 of N21 If we exclude hitrate, but N23 and N24 are not 1/4 and 1/8 of N21.
Then 1GB IC BW exluding hitrate would be 10.67x higher than N31 has?

 

[SteinFG]

I think that effective BW is just marketing BS.

Averaging It is just nonsense, either you have the data in cache or you don't.
If you have It, then you have the maximum ~1940 GB/s BW in case of RDNA2 N21 or ~4470 GB/s for RDNA3 N31.
If you don't then It leaves you with only GDDR6/7 BW.

P.S. I wonder how much IC would be needed for 90% hitrate at 4K. 1GB?
I think 4 stacks of Hynix HBM3E with 4TB/s(1TB/s per Stack) and 64-96GB Vram(16-24GB per Stack) could end up cheaper to make.

Edit: It looks like IC size matters for total BW.
I think what Locuza wrote as theoretical Iinfinity Cache BW is wrong for N23/24. N22 is 0.75 of N21 If we exclude hitrate, but N23 and N24 are not 1/4 and 1/8 of N21.
Then 1GB IC BW exluding hitrate would be 10.67x higher than N31 has?

If I'll take AMD's numbers, there's an easy way to calculate the L3 hitrate they're expecting. And about N23/N24 - yes, it's ~1/4 and ~1/8, otherwise numbers don't add up at all.

 

[GodisanAtheist]

I can't believe AMD executives actually have the balls to say "We could have completed but we didn't [cause our tech wasn't performing to snuff]" on record where people can hear them.

 

[GodisanAtheist]

Again, random forums are whatever, mainstream perception matters and your mainstream plebbitor expects anudda Vega.

-There is no mainstream perception. The mainstream does not theory-craft future products. The mainstream buys whatever is in front of them within their budget so long as it is branded Nvidia at any given time and are happy with it.

There is industry/insider and enthusiast perception, and that's what matters in this case.

 

[Panino Manino]

AMD fans always create such a massive amount of hype around their future products that even if the card turns out to be 10% on top for 75% of the price they'd still be disappointed because it doesn't do their taxes and make them a ham sandwich afterwards.

I think what people expect from AMD is to launch a card that it's faster than Nvidia's top competing card.
What we expect that is that the new card will be faster than the older card, with each competing brands alternating at the top. But for how many years AMD don't have a card that is faster than Nvidia's top? Since the days of the 980ti?

We see AMD "doing the impossible" against Intel with CPUs, we were expecting that with time and engineers the GPU site would improve, but besides "surviving" things feels pretty stagnant. They're not even trying to best Nvidia and the software side that with each generation becomes more important is still lagging.

Some times AMD gives some hope that things can change, but than a new arch like RDNA 3 falls bellow their own expectations, so of course some will be disappointed.

There wasn't any.
Everyone expected RDNA2-ish perf/w bump with more watts to get more goodies on the top end.

There was some hype as always, but I think this time the hype deflated before launch as soon as people know that AMD was going backwards and changing to dual issue.

 

[Panino Manino]

Ughhh have you seen MI300?
That's very GPU.

Things may be better on the markets CDNA is aimed at, but I'm talking about the consumer market and RDNA3.

This is merely a question of building a big enough config.

If they can't or don't want doesn't matter.

 

[Gideon]

OUCH. GG if true.

https://twitter.com/x/status/1687388515670056960

Navi4 lineup will not have any high-end GPUs

Think of it like RDNA1 or Polaris generation.

Judging by the predating @uzzi38 reaction this looks to be true:

https://twitter.com/x/status/1687379360276185088

It's almost public knowledge now (and probably will be eventually) but congratulations guys, you did it!

The way things are going soon there really will be only one desktop GPU manufacturer left. Just what you all wanted!

I can understand the RDNA1 generation as AMD was totally broke for most of the time it was under development, but ...

Frankly, I don't get why they did it. Are the economics really just not there? Or rather is it a sign that chiplet-based GPUs aren't really working out yet?

 

[CouncilorIrissa]

Isn’t Arc Alchemist (DG2) technically a second-gen product, with DG1 being the first? It’s just that DG1 sucked so bad that it was a no-go in terms of making an actual product. So Intel kind of buried it to make sure no one remembers it, much like they did with Cannon Lake?

My point is that Intel can’t afford to be this behind in PPA. If they’re not reasonably close with Celestial, it’s not out of the question for them to pull the plug on their dGPU program (again)

 

[igor_kavinski]

I mean for gods sake, Gen's head honcho (Blythe) is at AMD.

Ah. So David Blythe is the real genius behind ARC and XeSS. Must have been irked by Koduri's over exposure and decided to jump ship to escape the boisterous monkey trying to steal the limelight.

 

[GodisanAtheist]

Yes they can.
Lol.
Roadmaps don't exist in vacuum.

That's just naive.
RDNA4 had 5 parts, and so does RDNA5.

- "Yes AMD can do everything, but in the very same post I'm also saying they had to cut parts from RDNA 4 because they can't do everything"

That's some Level 9 Black Belt Logic-Fu there.

 

[gdansk]

Interesting. All the articles I have seen just say every N3 wafer they can produce, no process differentiation.
But hey that is main-stream news isn't it =)

It's from Wayne Ma originally who says Apple has "roughly a year" of 3 nanometer exclusivity. Article is more concerned about Apple getting a special deal for helping finance TSMC's bleeding edge processes.
What they don't say is that the exclusivity is because N3e - the non-bleeding process which everyone else will be using - is about a year after N3b (give or take a quarter).

 

[Joe NYC]

Unlikely, current GPGPU spending bump is mostly due to a massive pull-in by the likes of China in particular.

Interesting point. So if there is a pull in, of demand, to beat the sanctions, then, there is going to be a reduction from China, once all the stupid sanctions kick in.

In the meantime, Huawei is claiming to have an A100 level home grown card.

 

[Ajay]

Oh it is.
I'll give it 4-6Q before startups go mass extinction/exit scam.

Startups - yes. NV, don't think so. We'll see if any of the well funded companies like Alphabet, or Intel can do anything - probably not.
It's not a scam, just overblown. ML has a bright future, AI is the hype.

 

[TESKATLIPOKA]

No they have amazing GPU IP tied to best in the industry packaging engineering.

GPU IP could be excellent, but If they can't make a great product out of It then It's kinda pointless.
RDNA3 was not a good showing, hopefully we will see some frequency improvement with RDNA3.5.
If not, then It's kinda worrying.

 

[moinmoin]

TruForm was just an experiment.

Everything is an experiment until it isn't anymore.

 

[branch_suggestion]

AMD is launching a pro variant of cards that uses N32. N32 is doing just fine.

Well yeah, it functions fine and is still an improvement over RDNA2, but perf/power targets were missed badly.
RDNA3 is still cost competitive, but AMD had to take a chunk off the margins they actually wanted this gen.

 

[dr1337]

This really sounds like an unmedicated rambling.

If AMD markets zen 2 and zen 3 along side zen 4 as 7xxx series in laptops, they could start doing the same thing in GPUs. Fact of the matter is their speculation is as baseless as yours. And frankly you shouldn't be one to judge with your endless stream of vague and unhelpful single sentence quips.

I am just asking for model numbering, be it RX7000 or RX8000? Is that so hard for you to answer?

He doesn't know anything really, and nobody will until we're close to the announcement. AMD almost branded the 5700xt as the rx 690, sloppily leaving traces of this in their E3 2019 slides. I really doubt they're going to do the laptop thing to their GPU branding but things are technically still up in the air.

I definitely don't believe the MLID rumor millers that RDNA 4 is an eco round. But doing another rx 480 type launch isn't out of AMDs cards and with such an idea would also potentially come a branding reshuffle.

 

[GodisanAtheist]

What do you say to that, Mr. @adroc_thurston ?

-He says "short, vague, non-committal response that sounds confident but can be interpreted 12 ways from Sunday based on your own personal biases".