Question Speculation: RDNA3 + CDNA2 Architectures Thread

[uzzi38]

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 is much bigger than from RDNA1 to RDNA2. We should expect many big improvements in GFX11. 🤔" / Twitter

 

[TESKATLIPOKA]

Please ... MarkPC ? Fake channel with fake videos?

BTW MLID pulls 95% of his info from his azz.

I never said those are legit or not, I just reposted what TPU wrote.
BTW what MarkPC? Don't understand what you mean.

 

[Karnak]

AMD has been quite honest with their performance when they launched their CPU's. So when they say 54% increase performance/watt going from RDNA2 to RDNA3 we no longer believe this to be the case?

It def wont be the case for their real products because there wont be a 300W 7900XTX which they did use for the perf/watt claim. Pushing it all the way up to 355W means it'll be less efficient, same goes for the 7900XT since it's obviously slower @300W than a XTX.

 

[exquisitechar]

Nvidia is simply a strong competitor in comparison to Intel.

In addition. AMD being on 7nm when Nvidia was on Samsungs 8nm garbage created an illusion of AMD catching up but since maxwell, AMD has generally been behind.

Add in the high development of finfet chips, AMD generally CPU first spending for R and D and we get the typically AMD being behind in the GPU space.

What kind of magnifies this disappointment however is the hype train that AMD's guerilla marketing team creates for every launch. Fake leakers that mysteriously disappear after launches(the most recent is greymon). The use and then subsequent destruction of youtubers/influencers who use leaks for their platform. Eg. s semiaccurate, adoredtv, MLID, redgamingtech. After each of these hype trains crashes, typically these influencers have faded into obscurity. The next person awaiting that fate is likely MLID. AMD used to do it more transparently with people on forums and AMD representatives, but after similar hype crashes, AMD reps and subsequently own reputation AMD took a hit.

Don't think most of those have anything to do with AMD. Greymon probably disappeared because his leaks were a flop and people were going to stop taking him seriously or would just make fun of him if he kept going. AdoredTV also just had a meltdown after his terrible Zen 2 and Navi leaks. Charlie is just Charlie, he hates Nvidia.

AMD isn't really terribly behind in PPA even with the current N31, only the RT performance is dismal. If there's a bug that affects clock speeds as badly as rumored, the parts without it will be a slam dunk in terms of PPA. We'll see when N32 or the supposed fixed N31 releases.

 

[Tup3x]

Here are some test results from TPU. First Alleged RX 7900-series Benchmarks Leaked
Original source is @Vitamin4Dz
1. God of War
2. Cyberpunk 2077
3. Assassin's Creed Valhalla
4. Watch Dogs: Legion
5. Red Dead Redemption 2
6. Doom Eternal
7. Horizon Zero Dawn

gdansk: For that price yes, but originally It should have cost more and performed better.

Source is this video (MarkPC, deliberately added the spaces):
ww w.you tube.co m/watch?v=kKYBrANLbB0
Me, you and even neighbours dog knows that these are fake benchmarks. It's impressive that they even posted this as news...

That tweet just had screenshots from that fake video.

 

[Glo.]

Don't think most of those have anything to do with AMD. Greymon probably disappeared because his leaks were a flop and people were going to stop taking him seriously or would just make fun of him if he kept going. AdoredTV also just had a meltdown after his terrible Zen 2 and Navi leaks. Charlie is just Charlie, he hates Nvidia.

AMD isn't really terribly behind in PPA even with the current N31, only the RT performance is dismal. If there's a bug that affects clock speeds as badly as rumored, the parts without it will be a slam dunk in terms of PPA. We'll see when N32 or the supposed fixed N31 releases.

And the funniest part is that the rumors about AMD didn't came from People working for AMD, but... their competitiors.

 

[biostud]

It def wont be the case for their real products because there wont be a 300W 7900XTX which they did use for the perf/watt claim. Pushing it all the way up to 355W means it'll be less efficient, same goes for the 7900XT since it's obviously slower @300W than a XTX.

That might be so, but if a 300W 7900XTX is 54% faster than a 300W 6900XT, then a 355W 7900XTX is not going to be slower than a 300W 7900XTX. Then we should be looking at ~60% faster than a 6900XT.

 

[Karnak]

That might be so, but if a 300W 7900XTX is 54% faster than a 300W 6900XT, then a 355W 7900XTX is not going to be slower than a 300W 7900XTX. Then we should be looking at ~60% faster than a 6900XT.

That'd be still less than AMDs claimed +54% perf/watt because at 355W the XTX also consumes 18% more power while it won't be 18% faster.

ofc it's not by a huge amount but AMD knows that there won't be a real RDNA3 SKU (well at least for now) hitting their +54% marketing perf/watt claim. For consumers such comparisons are always a bit disappointing because they get less than what was "promised".

 

[biostud]

That'd be still less than AMDs claimed +54% perf/watt because at 355W the XTX also consumes 18% more power while it won't be 18% faster.

ofc it's not by a huge amount but AMD knows that there won't be a real RDNA3 SKU (well at least for now) hitting their +54% marketing perf/watt claim. For consumers such comparisons are always a bit disappointing because they get less than what was "promised".

If we get 60% better than 6900XT @ 45% better performance/watt, I don't think you will see many complaining.

 

[moinmoin]

Just to recollect RDNA1 carried these hazards and lots of bugs (...)

None of these are in RDNA2

But actually, yeah not as bad as RDNA1

So RDNA3 is buggy again, maybe RDNA4 was originally planed as the fixed version again? If that ever was a real plan and AMD does do a fixed N31 refresh next year AMD should instead better stop doing those faulty gens. RDNA1 didn't get a fixed refresh, but RDNA2 came only 16 months after RDNA1. RDNA3 took 25 months with only a limited product range.

 

[SmokSmog]

BTW what MarkPC? Don't understand what you mean.

Name of the youtube channel from which TPU took those images.

 

[Stuka87]

So RDNA3 is buggy again, maybe RDNA4 was originally planed as the fixed version again? If that ever was a real plan and AMD does do a fixed N31 refresh next year AMD should instead better stop doing those faulty gens. RDNA1 didn't get a fixed refresh, but RDNA2 came only 16 months after RDNA1. RDNA3 took 25 months with only a limited product range.

What do you mean by "again"? RDNA 3 isn't even out yet.

As for the time between releases of RDNA 2 and and RDNA, its the same as nVidia's 3K to 4K series. There was no reason for either maker to release things earlier when every single card was selling out due to the mining craze.

 

[moinmoin]

What do you mean by "again"? RDNA 3 isn't even out yet.

I was responding to the already documented software workarounds in Mesa for hardware bugs in both RDNA1 and RDNA3 which @DisEnchantment posted.

As for the time between releases of RDNA 2 and and RDNA, its the same as nVidia's 3K to 4K series. There was no reason for either maker to release things earlier when every single card was selling out due to the mining craze.

So AMD delayed the tape out of RDNA3 due to the mining craze only to then find out about the hardware bugs which delays N32 and a fix for N31 for a further year? That doesn't sound right.

 

[Saylick]

Around 3 months for the wafer, then some more for the packaging/binning/verification, let's say 3 more months, so around half a year in total.

I'm sure it took time to diagnose the problem, come up with the solution, and then needing to reprint the mask set. When Fermi was respun, it was roughly 6 months between the GTX 480 and the GTX 580. I have my doubts that it's that fast today because nodes and silicon designs are more complicated than ever before, so I'm thinking it's closer to a year. That is, unless AMD intentionally delayed N31 launch and was mad scrambling behind the scenes to get that respin out. But I doubt they would intentionally delay.

 

[Kepler_L2]

So RDNA3 is buggy again, maybe RDNA4 was originally planed as the fixed version again? If that ever was a real plan and AMD does do a fixed N31 refresh next year AMD should instead better stop doing those faulty gens. RDNA1 didn't get a fixed refresh, but RDNA2 came only 16 months after RDNA1. RDNA3 took 25 months with only a limited product range.

gfx11.5 exists for a reason, though that is a little bit more than just "fixed RDNA3".

 

[DisEnchantment]

So RDNA3 is buggy again, maybe RDNA4 was originally planed as the fixed version again? If that ever was a real plan and AMD does do a fixed N31 refresh next year AMD should instead better stop doing those faulty gens. RDNA1 didn't get a fixed refresh, but RDNA2 came only 16 months after RDNA1. RDNA3 took 25 months with only a limited product range.

Actually GFX10 lasted for 3.5 years (GFX 10_1/10_0/10_3), over two product cycles, RDNA 1 & 2 (2019-2022). RDNA4 is supposedly GFX12.

gfx11.5 exists for a reason, though that is a little bit more than just "fixed RDNA3".

Interesting, first time I heard of GFX11.5. GFX11 will have a shorter lifespan but another significant revision in GFX11, that is quite aggressive from RTG.
Keeping my eyes peeled for patches, or perhaps you got them already?
Strix patches perhaps could probably carry it since it says RDNA3+, but not sure if RDNA3+ is GFX11_5

 

[Kepler_L2]

Actually GFX10 lasted for 3.5 years (GFX 10_1/10_0/10_3), over two product cycles, RDNA 1 & 2 (2019-2022). RDNA4 is supposedly GFX12.


Interesting, first time I heard of GFX11.5. GFX11 will have a shorter lifespan but another significant revision in GFX11, that is quite aggressive from RTG.
Keeping my eyes peeled for patches, or perhaps you got them already?
Strix patches perhaps could probably carry it since it says RDNA3+, but not sure if RDNA3+ is GFX11_5

Yes gfx11.5 = RDNA3+ and gfx12 = RDNA4. gfx11.5 is mostly bug fixes but it has parts of the new SALU from RDNA4.

 

[Stuka87]

I'm sure it took time to diagnose the problem, come up with the solution, and then needing to reprint the mask set. When Fermi was respun, it was roughly 6 months between the GTX 480 and the GTX 580. I have my doubts that it's that fast today because nodes and silicon designs are more complicated than ever before, so I'm thinking it's closer to a year. That is, unless AMD intentionally delayed N31 launch and was mad scrambling behind the scenes to get that respin out. But I doubt they would intentionally delay.

AMD could have discovered this months ago.

Also, doing a respin means only changing the GCD (unless the issue involves the MCD). AMD themselves said an advantage to chiplets is that it’s far less work to respin.


Sent from my iPhone using Tapatalk

 

[scineram]

What product will be using gfx11.5? Is that just Strix Point?
Navi 31 might get a fixed new stepping, but not an architectural upgrade.

 

[scineram]

Kyle is giving me some hope after all the bad rumours at least.

https://twitter.com/x/status/1601997050580697088

 

[Kaluan]

Well, GCN v5.1 (Radeon VII/Pro/Instinct, Renoir/Lucienne, Cezanne/Barcelo/Barcelo-R) was GCN's final revision (and comprehensive full node shrink to 7nm) and was used in both high-end workstation/high-end gaming AND APUs.

So maybe we'll see the same with RDNA3+.
Don't expect big node changes outside of ground-up designs (such as the hypothetical N34). But the fact that Strix Point says "Advanced Node" is curious. Will it be N3E Zen5 + N5/N4 RDNA3+ or monolithic N3E (or something else)?

I don't think current RDNA3 being broken is a last moment realization by AMD, but they had to release it regardless.

 

[Semi-fiend]

I haven't seen anywhere, but does anyone know the exact time of date (including time and timezone) that the review embargo lifts?

 

[Stuka87]

I haven't seen anywhere, but does anyone know the exact time of date (including time and timezone) that the review embargo lifts?

I have not seen an exact time (Date is the 12th) but it’s normally like 9am EST. Which is tomorrow morning.


Sent from my iPhone using Tapatalk

 

[Timorous]

Well, GCN v5.1 (Radeon VII/Pro/Instinct, Renoir/Lucienne, Cezanne/Barcelo/Barcelo-R) was GCN's final revision (and comprehensive full node shrink to 7nm) and was used in both high-end workstation/high-end gaming AND APUs.

So maybe we'll see the same with RDNA3+.
Don't expect big node changes outside of ground-up designs (such as the hypothetical N34). But the fact that Strix Point says "Advanced Node" is curious. Will it be N3E Zen5 + N5/N4 RDNA3+ or monolithic N3E (or something else)?

I don't think current RDNA3 being broken is a last moment realization by AMD, but they had to release it regardless.

Last moment, no.

They didn't have to release it though but I suspect they ran the numbers on how much they could sell a fixed version for sometime next year vs how much they can sell the current version for now and it worked out that selling the existing version now was more profitable and they can save the fixed version for the refresh to have a stronger uplift there.

 

[leoneazzurro]

If the performance of the broken chip is indeed between 4080 and 4090 in raster, and around 3090 in RT, it is a good value proposition (looking at the market) for 999$ while having a BOM way below that. That is, it may have its market and as such, it would help AMD to recover part of the investment while waiting for the fixed version. Not selling would have meant getting only the losses for the chip development until the new stepping comes out, and leave the whole high end of the market to Nvidia (Not being able to compete at the same level with the 4090 does not mean the 7900 cards are not high end).

 

[eek2121]

If the performance of the broken chip is indeed between 4080 and 4090 in raster, and around 3090 in RT, it is a good value proposition (looking at the market) for 999$ while having a BOM way below that. That is, it may have its market and as such, it would help AMD to recover part of the investment while waiting for the fixed version. Not selling would have meant getting only the losses for the chip development until the new stepping comes out, and leave the whole high end of the market to Nvidia (Not being able to compete at the same level with the 4090 does not mean the 7900 cards are not high end).

While AMD would never do it (they try to roughly maintain perf/$ with the competition), imagine what would have happened if they had released for $699 for the 7900XT and $799 for the XTX.