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]

I hear kopite77valtteri posted that N32 would be 72CU with a 3.5GHz base and 4.4GHz game clock. Also all AIB cards are required to be boldly emblazoned with a picture of Ruby cutting a 4090 in half with her sword.

He already deleted that post. I think he realized his source was joking.

 

[eek2121]

More of the card is engaged for gaming workloads.

I am curious, however, if a simpler gaming workload (maybe just draw a single triangle) would require reducing clocks.

 

[Panino Manino]

He already deleted that post. I think he realized his source was joking.

Wait, that was not a joke, the tweet (with 4.4GHz game clock) was real?

 

[maddie]

Oh boy, here we go.

 

[TESKATLIPOKA]

Wait, that was not a joke, the tweet (with 4.4GHz game clock) was real?

There was no such tweet. Don't take It seriously. MrTeal was joking, just reread his post.

Also all AIB cards are required to be boldly emblazoned with a picture of Ruby cutting a 4090 in half with her sword.

 

[biostud]

Obviously overall performance will be taken into consideration, but the fact that AMD has cards in the $600 range with 16 Gb vram, vs 8-10 with nvidia is something that I also will consider.

 

[biostud]

https://twitter.com/x/status/1603984304052523009

 

[TESKATLIPOKA]

Obviously overall performance will be taken into consideration, but the fact that AMD has cards in the $600 range with 16 Gb vram, vs 8-10 with nvidia is something that I also will consider.

Will N32 really be in that range? If we based It on N31, then It's possible, because of lower performance, but according to leaks N32 is not that buggy.
It is possible that performance will be higher than 65-70% of RX 7900XTX, then It will cost more than $699 in my opinion. I see RTX 4070Ti as Its main competitor in raster.
16GB Vram would be a nice bonus along with a decent TBP.
The next big adrenaline update from AMD will be interesting.

P.S. I want to see some RDNA3+ GPUs, to see how much performance was actually lost with RDNA3.

 

[Kepler_L2]

Will N32 really be in that range? If we based It on N31, then It's possible, because of lower performance, but according to leaks N32 is not that buggy.
It is possible that performance will be higher than 65-70% of RX 7900XTX, then It will cost more than $699 in my opinion. I see RTX 4070Ti as Its main competitor in raster.
16GB Vram would be a nice bonus along with a decent TBP.
The next big adrenaline update from AMD will be interesting.

P.S. I want to see some RDNA3+ GPUs, to see how much performance was actually lost with RDNA3.

At the moment RDNA3+ looks to be only for APUs.

 

[TESKATLIPOKA]

At the moment RDNA3+ looks to be only for APUs.

That's ok, I am just curious.
Hopefully AMD will take this as a lesson and won't make the same mistake with RDNA4 being overly aggressive with design goals.
I personally think N31 ended up worse than even Vega, yet the architecture looks pretty good.

 

[biostud]

Will N32 really be in that range? If we based It on N31, then It's possible, because of lower performance, but according to leaks N32 is not that buggy.
It is possible that performance will be higher than 65-70% of RX 7900XTX, then It will cost more than $699 in my opinion. I see RTX 4070Ti as Its main competitor in raster.
16GB Vram would be a nice bonus along with a decent TBP.
The next big adrenaline update from AMD will be interesting.

P.S. I want to see some RDNA3+ GPUs, to see how much performance was actually lost with RDNA3.

If N32 is not, then there will be absolutely no reason to buy it over N21 cards, price/performance wise.

 

[scineram]

Lmfao and how exactly do you plan to do that?

 

[biostud]

Lmfao and how exactly do you plan to do that?

Do what?

 

[TESKATLIPOKA]

If someone is interested, then here is a link to RDNA3 ISA reference guide.

BTW even that rumor about 4GHz wasn't a total lie. It just wasn't true for gaming.

 

[scineram]

Do what?

Buy Navi 21 in 6 months.

 

[biostud]

Buy Navi 21 in 6 months.

If not new, then used.

 

[scineram]

That's an entirely separate goalpost. Nobody prices their new hardware to some used market. Many are even apprehensive about buying used.

 

[coercitiv]

Did I miss the fireworks?
https://www.tomshardware.com/news/amd-addresses-controversy-rdna-3-shader-pre-fetching-works-fine

 

[Bigos]

If the twitter gang cannot even properly read facts (public source code) I would not trust anything they say.

 

[RnR_au]

BTW even that rumor about 4GHz wasn't a total lie. It just wasn't true for gaming.
...

A screeny from another twitter poster;

https://twitter.com/x/status/1603848194919636993

- from a XFX 7900XTX Merc

Its an odd story... with a messy narrative, not helped by the #silicongang I guess.

 

[Ajay]

Hmm..

https://twitter.com/x/status/1604135525648158720

 

[Panino Manino]

A screeny from another twitter poster;

https://twitter.com/x/status/1603848194919636993

- from a XFX 7900XTX Merc

Its an odd story... with a messy narrative, not helped by the #silicongang I guess.

What is worse?
N31 not reaching 3GHz, or N31 actually achieving even 4GHz but despie this not delivering enough actual performance to compete with Nvidia's best like expected?

 

[Mopetar]

Who knows what's really going on right now. I'm sure we'll learn more as the months go on and assuming we do eventually get an A1 stepping for Navi 31 we'll have a clearer picture still. It's certainly odd and I hope we get a full picture someday.

 

[moinmoin]

I guess the pre-launch roller coaster just didn't stop at launch.

Now N31 is an impeccable A0 silicon we should marvel about?

The one part I'm still waiting on is AMD explaining where the promised >50% perf/watt increase went. Some reviews like Computerbase even reported worse perf/watt gen over gen in some cases.

 

[RnR_au]

What is worse?
N31 not reaching 3GHz, or N31 actually achieving even 4GHz but despie this not delivering enough actual performance to compete with Nvidia's best like expected?

It doesn't hit 4GHz in games. Just compute tasks. But it means that some parts of the arch can clock really high under certain circumstances. This is probably where the RDNA 3 = 4GHz rumour came from. Right number, but without the caveats.