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

 

[Ajay]

Reading comprehension issue?

I said "Even 4K at typical living room distance and TV size is overkill"

If you are sitting 10+ feet away from a 55" TV, then you aren't seeing 4K. You are barely in the 1080p zone. My point is that distance matters. This was in response to someone claiming we needed 12K...

Distance matters.

We have a dedicated forum for discussing this sort of thing http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=forums/displays.72/

 

[SolidQ]

looks like some dudes also had similar confirmations about this mysterious next-gen RDNA GPU:

Seems 8700XT near 7900XT perfomance, which fine, but the price will decide everything

 

[igor_kavinski]

Seems 8700XT near 7900XT perfomance, which fine, but the price will decide everything

Should be $450 or worse, $600.

 

[TESKATLIPOKA]

Should be $450 or worse, $600.

Difference between 7800XT vs 7900XT is only 28%.
$600 would be a lot for a GPU weaker than 7900XT.

 

[SolidQ]

$600 would be a lot for a GPU weaker than 7900XT.

i think 500 max and big question, what he's meaning N31 is 7900XTX or whole N31

 

[jpiniero]

Seems 8700XT near 7900XT perfomance, which fine, but the price will decide everything

Therein lies the rub.

i think 500 max

Problem of course is that if it's N3E monolithic, the infinity cache becomes that much more expensive. Plus whatever the additional cost of GDDR7.

 

[maddie]

Therein lies the rub.



Problem of course is that if it's N3E monolithic, the infinity cache becomes that much more expensive. Plus whatever the additional cost of GDDR7.

Why should it be monolithic? Even monolithic N33 is on 6nm. Do we have any monolithic 5nm N3x chips? Now you're pushing this N3E monolithic idea. Everything to raise our price barriers to purchase. Why do you never surprise me?

 

[jpiniero]

Why should it be monolithic?

That's the rumor, that it's monolithic. As to it using N3E, well, that's just a guess. Could use N4P.

 

[maddie]

That's the rumor, that it's monolithic. As to it using N3E, well, that's just a guess. Could use N4P.

Could as well mean 1 GPU chiplet with associated Cache& I/O chiplets. The rumored problem was getting the multi-chiplet core chiplets to work well, not that it was monolithic completely.

If the cost delta between 6nm and 5nm was not enough for a chiplet low end Navi3, the greater cost delta between 6nm & N3 probably changes the balance to favor a chiplet low end this round.

 

[Kepler_L2]

Therein lies the rub.



Problem of course is that if it's N3E monolithic, the infinity cache becomes that much more expensive. Plus whatever the additional cost of GDDR7.

Both Navi4 dies are monolithic N4P.

 

[Tigerick]

Both Navi4 dies are monolithic N4P.

Yep, and the reason behind designing monolithic die but not chiplets because AMD is positioning N43 and N44 to slot in between N33 and N32. They are not replacing 7600/7700XT/7800XT; AMD won't throw them away just after one year. In fact, I can make simple price guess: N43 going to be priced $399 and N44 going to cost $299...

 

[Ajay]

looks like some dudes also had similar confirmations about this mysterious next-gen RDNA GPU:

View attachment 86862

https://twitter.com/x/status/1710832438321320067

edit: ouch nobody pointed out my mistake of posting wrong X(twitter) link lol. but yeah OTOH the X links seem don't work well since X changed the policy.

Seems 8700XT near 7900XT perfomance, which fine, but the price will decide everything

The comparison being made is GPU *size*. Given that this will be RDNA4, not have RDNA3's clock bugs, it'll have performance above the 7900 XTX. If any of this is true. It would support that comment by the AMD engineer who said he currently has a next gen high performance GPU in hand that he is working on. Not great news, but not terrible. We'll just all have to wait and see.

 

[Tigerick]

The comparison being made is GPU *size*. Given that this will be RDNA4, not have RDNA3's clock bugs, it'll have performance above the 7900 XTX. If any of this is true. It would support that comment by the AMD engineer who said he currently has a next gen high performance GPU in hand that he is working on. Not great news, but not terrible. We'll just all have to wait and see.

Yeah, it is refering to the die size, but because of difference in single die and chiplet, it should be like this:-
300 > N43 > 200

adhoc did hinted about die size of N43:- between 200 - 250 mm2, so it fits right in

 

[coercitiv]

Yeah, it is refering to the die size, but because of difference in single die and chiplet, it should be like this:-
300 > N43 > 200

adhoc did hinted about die size of N43:- between 200 - 250 mm2, so it fits right in

Ladies and gents, make (some) room for the new Polaris!

 

[Mopetar]

Ladies and gents, make (some) room for the new Polaris!

If it's priced like Polaris (particularly after the price cuts) I'll roll out the carpet and throw some rose petals.

I don't know what we'll see with RDNA 4, but it does make a certain amount of sense for the bottom part(s) to be monolithic since it makes for a smaller package which is important for many notebooks.

The issues with RDNA3 don't seem to be related to the use of chiplets of any kind as N33 had the same issues with clock speeds as N31 and it was a monolithic design. Granted it was on a different, older node that would have limited the clocks to some extent anyways.

 

[Tigerick]

Ladies and gents, make (some) room for the new Polaris!

And 8000M mobile GPU

 

[Ajay]

Yeah, it is refering to the die size, but because of difference in single die and chiplet, it should be like this:-
300 > N43 > 200

adhoc did hinted about die size of N43:- between 200 - 250 mm2, so it fits right in

He also said N43 wasn’t the correct ID for the die. N31 GCD is 350mm^2

 

[SolidQ]

Interesting news. Maybe FSR going finally to AI?

https://www.amd.com/en/newsroom/press-releases/2023-10-10-amd-to-acquire-open-source-ai-software-expert-nod-.html

 

[jpiniero]

He also said N43 wasn’t the correct ID for the die. N31 GCD is 350mm^2

TPU says N31 GCD is 304 mm2.

 

[TESKATLIPOKA]

Yep, and the reason behind designing monolithic die but not chiplets because AMD is positioning N43 and N44 to slot in between N33 and N32. They are not replacing 7600/7700XT/7800XT; AMD won't throw them away just after one year. In fact, I can make simple price guess: N43 going to be priced $399 and N44 going to cost $299...

You don't need 2 new GPU dies to slot between N33 and N32, one is more than enough.

4K performance from TPU
7800XT: 204%
7700XT: 167%
N44 full: 140%
N44 cut: 120%
7600: 100%

N44 to be ~40% faster than N33 would need 40CU and 3GHz, excluding any possible IPC gain.

With this performance, N44 at $399 would have worse performance/$ ratio than both N33 and N32

BTW, I don't even understand why AMD should keep N32 and N33.

 

[Ajay]

TPU says N31 GCD is 304 mm2.

Hmm, should have checked TPU. Oh well - just hoping RDNA4 isn’t as sad as it appears it will be.

 

[adroc_thurston]

You don't need 2 new GPU dies to slot between N33 and N32

They're not slotting anything there.

 

[TESKATLIPOKA]

They're not slotting anything there.

Tell that to @Tigerick, not me.

 

[Timorous]

Both Navi4 dies are monolithic N4P.

So either small or no infinity cache then which means GDDR7.

If these were lower end so 250mm ish then it would indicate 192 bit GDDR7 running 3GB modules for 18GB or vram on a 35B xtor or there abouts chip with ballpark 7900XT / 4070Ti performance as a baseline with some room for upside if the clock rate fixes don't cost more xtors. In which case there is potential for 3Ghz+ clocks and an extra maybe 20% over the baseline which is 7900XTX tier.

If AMD are sticking with GDDR6 then it will need IC and with how poor cache and logic scales it wouldn't make any sense to move away from an RDNA3 style layout even if they dropped their monster MI300 style dGPU because that config would have been halo tier and priced as such.

 

[Tigerick]

You don't need 2 new GPU dies to slot between N33 and N32, one is more than enough.

4K performance from TPU
7800XT: 204%
7700XT: 167%
N44 full: 140%
N44 cut: 120%
7600: 100%

N44 to be ~40% faster than N33 would need 40CU and 3GHz, excluding any possible IPC gain.

With this performance, N43 at $399 would have worse performance/$ ratio than both N33 and N3

BTW, I don't even understand why AMD should keep N32 and N33.

It is business decision, my friend. AMD is expecting N32 & N33 to have shelf life of two years. Especially with N32 is only available for desktop variant....

Anyhow, below is my speculation and price estimates of RDNA4 & RDNA5 lineups for reference:

SRP	2023	N5/N6	2024	N4P	2025	N3E
$1499?					Navi5c 32GB 512-bit GDDR7	270 CU
					28GB 448-bit GDDR7	?
$999	7900XTX 24GB 384-bit GDDR6	96 CU			N51 24GB 384-bit GDDR7	210 CU
$899	7900XT 20GB 320-bit GDDR6	84 CU			N51L 20GB 320-bit GDDR7	?
$699					N52 16GB 256-bit GDDR7	140 CU
$649	7900GRE 16GB 256-bit GDDR6	80 CU
$499	7800XT 16GB 256-bit GDDR6	60 CU			N53 12GB 128-bit GDDR7	70 CU
$449	7700XT 12GB 192-bit GDDR6	54 CU
$399?			7700 12GB 192-bit GDDR6	40 CU
$299?			7600XT 8GB 128-bit GDDR6	32 CU
$269	7600 8GB 128-bit GDDR6	32 CU

• I know it is pre-mature to list the prices, let alone WGP numbers, but we got to start somewhere. When AMD started new design, they will think about price, position, process node and memory bandwidth requirements. Here comes the table; I could be wrong but I could also modify later, that's how we learned.
• We don't know much about RDNA5 but at least we know they are going to support GDDR7. And since AMD going to employ chiplet design, the numbers of CU should be multiply of base number. That number is hard to know, but cause of 270/3 number appeared before, I could make a wild guess if that number is divided by 4, we get maximum "theoretical" CU numbers above. The CU number also aligns with my calculation of bandwidth's improvement that I listed on my frontpage of Blackwell series. Feel free to disagree with me if you have different numbers in mind.
• There is a wide gap between $500 and $900 without any GPU from AMD, I think I just found out how AMD going to fill it with N52, yes that's mean AMD going to price upcoming 8800XT with 16GB to $699 if my calculation is correct.
• If NV going to drop the price of RTX4080 to $999 next year and latest rumor is NV going to release RTX5000 series end of next year; we could see sharp price drop of N31 series, remember price cut of RX6950XT from $1,099 to $599??? Let's hope AMD won't drop down like this but if AMD is not releasing RDNA5 until end of 2025.......Screwup and delay do have consequences. Hmm, maybe that is the reason debaited guy is leaving????

PS: I am confused about AMD's RDNA4 lineup now, so let's focus on RDNA5