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

 

[uzzi38]

Not really. Their next step (which surely already happened) is to determine if it is financial viable to fix it at all. Correct me if I'm wrong but a respin will require new masks which are in millions likley double digit millions. Then it will take at least half a year. So it depends on how far RDNA 4 is. the new respinned product will require a certain time on market to make it viable.

Just a reminder that AMD even did a respin for Navi10, which was a pretty sub-par product overall.

 

[Timorous]

...no dual GCD products will come to market either

So quad GCD then?

 

[DisEnchantment]

So quad GCD then?

Checkmate, AMD moved all the CUs to the MCD. 6x MCD has 6x SEs
The GCD is just dummy silicon, jebaited!!

 

[biostud]

Sometimes the performance goals might not be reached, sometimes the competitor launch a product that forces you to change your product stack.

Many has speculated that nvidia hadn't intended the regular 3080 to be a GA102 part, but were forced to move it up to compete with the RX6800XT. Now it could be that AMD would have to lower the price/segment simply because they couldn't hit the performance targets they had aimed for. If any of these speculations holds any merits we may never know, but if we didn't have tech gossip and speculations why would we even hav thread with 126 pages

 

[Stuka87]

Just a reminder that AMD even did a respin for Navi10, which was a pretty sub-par product overall.

My memory must be fading. Navi 10 was the 5600XT/5700/5700XT. I don't recall a respin for any of these. And those cards were certainly not sub-par.

 

[uzzi38]

My memory must be fading. Navi 10 was the 5600XT/5700/5700XT. I don't recall a respin for any of these. And those cards were certainly not sub-par.

Around the same time as the 5600XT launched it came to market. No differences to the original silicon aside from it seemingly fixing issues with the display engine afaik. Those cards crashed less than the original silicon if nothing else.

 

[blckgrffn]

My memory must be fading. Navi 10 was the 5600XT/5700/5700XT. I don't recall a respin for any of these. And those cards were certainly not sub-par.

The cards were fine but the volume was always meant to be limited, imo, and the focus was on getting to RDNA2 asap. Not only was it the DX12 Ultimate compliant part but baked in issues (like what Intel has discussed with their latest GPUs in frank terms) with internal chip performance and bottlenecks meant they knew RDNA2 was the part that was going to be most competitive from an absolute and perf/watt perspective.

I bought a RDNA part (multiple, actually) thinking they were just the forerunner for the drivers, architecture and usage of TSMC silicon.

In retrospect I don’t think I was that far off.

 

[MrTeal]

Outside of clocks being lower than people expected, are there any solid leaks from AMD itself or typical leakers like kopite7kimi that there actually is a silicon issue that's caused the card to clock lower than expected.
Maybe I haven't been following along closely enough, but it seems almost taken as truth that there is an issue, it's identified and a respin could be possible when it doesn't seem like the original premise has really been proven.

 

[Kaluan]

Outside of clocks being lower than people expected, are there any solid leaks from AMD itself or typical leakers like kopite7kimi that there actually is a silicon issue that's caused the card to clock lower than expected.
Maybe I haven't been following along closely enough, but it seems almost taken as truth that there is an issue, it's identified and a respin could be possible when it doesn't seem like the original premise has really been proven.

How would one go about proving that tho?
It's just that it seems very plausible, something possibly happened and they didn't get the v/f curve they expected. Look at their other N5 design, Zen4/Raphael, it seems their clocking ambitions worked out very well there.

The main questions remain: "is that really the case?", followed by "and can it be fixed?"

Edit: I think Zen4/N5's meteoric clock bump over Zen3/N7 silicon also fed into the 3GHz++ RDNA3 hype. As well as the plausible explanation that they derped somewhere with the current design.

Either way, 13th of December can't come soon enough, we may get some more concrete answers to that then. Or at least more educated guesses.

Anyone remember the 3,72GHz fmax/cap leak? Is that still at play with the current 7900XT/XTX's firmware? Or was the info false?
Also is that different than 6900XT's initial 3GHz hard cap or the same deal?

 

[jpiniero]

Anyone remember the 3,72GHz fmax leak? Is that still at play with the current 7900XT/XTX's firmware? Or was the info false?

That's probally just something in the GPU's BIOS set as a max allowed frequency.

 

[Kaluan]

That's probally just something in the GPU's BIOS set as a max allowed frequency.

Yeah and we know some watercooled N21 could reach (the 3GHz hard cap). Where as 7900XT/XTX seem like a far cry from that. Guess we'll only know more after the launch.

 

[biostud]

Lets see when N32 is launched, and what clockspeeds it is able to achieve.

 

[amenx]

I wonder if its an inherent limitation of chiplet design. That N32 would not have this problem to begin with, that it only affects N31 and is not really an issue given its good performance already.

 

[GodisanAtheist]

Outside of clocks being lower than people expected, are there any solid leaks from AMD itself or typical leakers like kopite7kimi that there actually is a silicon issue that's caused the card to clock lower than expected.
Maybe I haven't been following along closely enough, but it seems almost taken as truth that there is an issue, it's identified and a respin could be possible when it doesn't seem like the original premise has really been proven.

- Its funny that the last go around with the 7900 series, I picked up an HD7950 that could overclock to the damn moon and back. Card came with stock 925 Mhz core and I could push the card to and have it hold at 1200Mhz core, yielding a buttload of extra performance with the help of the overprovisioned bandwidth on that gen. That was far from a unique experience, the HD7900's were known to be fantastic overclockers in general.

I thought it was odd then that AMD launched the card as is and had to relaunch a Ghz edition, didn't they know what their own silicon was capable of?

I wonder if we have a similar situation here. There seems to be a lot of conflicting info on what RDNA3 can and can't do and people seemed to have glommed on to the hardware bug explanation.

 

[adamge]

This forum won't even accept that the Intel ARC cards have a hardware bug. You really think they will accept that a company that has been making GPUs for 30 years would release a hardware bug?

 

[biostud]

This forum won't even accept that the Intel ARC cards have a hardware bug. You really think they will accept that a company that has been making GPUs for 30 years would release a hardware bug?

“This forum” is quite a broad statement, most probably don’t really care that much about the ARC cards.

 

[Joe NYC]

How can anyone believe rumors about how AMD is fixing the problem when all the leakers were blindsided by AMD even having the problem.

???

People are asking a questions if AMD is going to re-spin RDNA 3 or leaving it as is. And there are no official answers (AFAIK). So it is a puzzle, and the way you solve a puzzle is look for datapoints / hints.

 

[gdansk]

???

People are asking a questions if AMD is going to re-spin RDNA 3 or leaving as is. And there are no official answers (AFAIK). So it is a puzzle, and the way you solve a puzzle is look for datapoints / hints.

The entire "oh there's a bug and we need a respin" is still speculation. Spread by some of same people who were surprised by 2.3GHz.

The only thing that can confirm it, because AMD never will, is if they launch N31 again except with higher clocks.

 

[Joe NYC]

So quad GCD then?

If you had close captioning on during the RDNA (our DNA) presentation, it said AMD has triplets (chiplets).

 

[linkgoron]

The entire "oh there's a bug and we need a respin" is still speculation. Spread by some of same people who were surprised by 2.3GHz.

The only thing that can confirm it, because AMD never will, is if they launch N31 again except with higher clocks.

Or if N32 or N33 have much higher clocks.

 

[DeathReborn]

If you had close captioning on during the RDNA (our DNA) presentation, it said AMD has triplets (chiplets).

The return of the awesome X3 (RX2, RX3 & RX4).

 

[Kaluan]

We'll know a lot more on the 13th next month. And even more at CES 2023, when AMD will likely announce or at the very least tease RX 7000M (N32, N33).

 

[Joe NYC]

The entire "oh there's a bug and we need a respin" is still speculation. Spread by some of same people who were surprised by 2.3GHz.

The only thing that can confirm it, because AMD never will, is if they launch N31 again except with higher clocks.

There will be other RDNA3 products. Depending on their behavior, we will see if something held back N31.

 

[biostud]

The entire "oh there's a bug and we need a respin" is still speculation. Spread by some of same people who were surprised by 2.3GHz.

The only thing that can confirm it, because AMD never will, is if they launch N31 again except with higher clocks.

If the slide with the "designed for 3Ghz" is not a hoax, then it is a reasonable conclusion that something did not end up as expected.

 

[gdansk]

Or if N32 or N33 have much higher clocks.

How much higher? Base clocks between N21 and N22 were 500MHz apart and I doubt that was a fluke. For N32 to be 500MHz faster and hit 3GHz boost would not confirm anything, in my opinion.