AMD RDNA3 unveiling event

Besty · 24 Jan 2023 at 09:03

Grim5 said:
RDNA3 will be fixed with drivers

RDNA3 will be fixed with refresh

Wait for RDNA 4

https://twitter.com/i/web/status/1617195441728245762

this guy creates his own narrative by listing a number of hardware bugs listed in the driver notes, some of which are probably not even bugs. Says there will be a re-tape then, there is no re-tape.

people quote this guys tweets in almost every tech forum going.

Zarax · 24 Jan 2023 at 09:08

Is he even remotely reliable?

delta0 · 24 Jan 2023 at 09:28

Zarax said:
Is he even remotely reliable?

Yes Twitter is the most reliable peer reviewed publication going

Zarax · 24 Jan 2023 at 09:35

delta0 said:
Yes Twitter is the most reliable peer reviewed publication going

I was talking in relative terms. There are a few leakers (like kopite7) who seem to have internal sources and publish semi-reliable info, while others just pull stuff from their nether regions and hope something sticks. I'm wondering which category this guy falls in.

Besty · 24 Jan 2023 at 09:40

Zarax said:
I was talking in relative terms. There are a few leakers (like kopite7) who seem to have internal sources and publish semi-reliable info, while others just pull stuff from their nether regions and hope something sticks. I'm wondering which category this guy falls in.

Kopite generally talks about GPU AIB level leaks, I think he is more reliable than not. This other guy (all the watts) probably has a contact in AIB's too but he gives different information about CPU's. When it comes to GPU's his information seems to be coming from outside the AIBs and does not seem as reliable.

I am not saying any silicon is perfect (AMD, Nvidia, Intel etc.) but, there is a lack of collaboration between twitter posters saying 'i saw stuff in the driver notes, therefore - broken' and software developers that address hardware functions via compilers saying 'yes, I have tried this' and a) its a new hardware change not supported by the compiler and therefore its never going to work until my compiler is updated to support the new/changed function in the hardware b) its a new hardware change supported by the compiler, I have used the updated compiler but the function that AMD have designed in is not working as intended and therefore is genuine silicon 'errata'.

Zarax · 24 Jan 2023 at 09:44

Besty said:
Kopite generally talks about GPU AIB level leaks. This other guy (all the watts) probably has a contact in AIB's too but he gives different information.

I am not saying any silicon is perfect (AMD, Nvidia, Intel etc.) but, there is a lack of collaboration between twitter posters saying 'i saw stuff in the driver notes, therefore - broken' and software developers that address hardware functions via compilers saying 'yes, I have tried this' and a) its a new hardware change not supported by the compiler and therefore its never going to work until my compiler is updated to support the new/changed function in the hardware b) its a new hardware change supported by the compiler, I have used the updated compiler but the function that AMD have designed in is not working as intended and therefore is genuine silicon 'errata'.

Guess we will see which one it is when Navi 32 is released.

Andorphine · 24 Jan 2023 at 12:22

Zarax said:
Guess we will see which one it is when Navi 32 is released.

If that ever happens...

kieran_read · 24 Jan 2023 at 12:52

Zarax said:
Is he even remotely reliable?

If I've learnt anything from the internet...

If they agree with your point of view they are the most reliable person in the world.
If they disagree with your point of view the speak nothing but hot garbage.

There is no in between.

kieran_read · 24 Jan 2023 at 12:54

Zarax said:
I was talking in relative terms. There are a few leakers (like kopite7) who seem to have internal sources and publish semi-reliable info, while others just pull stuff from their nether regions and hope something sticks. I'm wondering which category this guy falls in.

What is the overall summary atm?

So I don't have to read through countless pages not been keeping up to date the last few weeks.

Still keen on 7900xtx but considering the Nitro already has 100 off in the sale I think I'll wait until either the 7900xtx or 4080 AIBs drop below £1k

Andorphine · 24 Jan 2023 at 13:14

kieran_read said:
What is the overall summary atm?

So I don't have to read through countless pages not been keeping up to date the last few weeks.

Still keen on 7900xtx but considering the Nitro already has 100 off in the sale I think I'll wait until either the 7900xtx or 4080 AIBs drop below £1k

The rumors so far were that N32 top die (7800XT) would come fairly close to the 7900XT, because AMD would fix a bug that makes N31 not clock/perform properly. The newest rumors suggest that N32 will be worse than expected. I'd expect 6950XT performance in rasterization.

kieran_read · 24 Jan 2023 at 13:28

Andorphine said:
The rumors so far were that N32 top die (7800XT) would come fairly close to the 7900XT, because AMD would fix a bug that makes N31 not clock/perform properly. The newest rumors suggest that N32 will be worse than expected. I'd expect 6950XT performance in rasterization.

Hmm

Well I don't consider 6950xt performance bad but it needs to be priced right.

Considering OCUK were selling. 6900xt for £700 (the model that was close to 6950xt performance)

I would expect the 7800xt to be under £700 ideally 5-600 with 6-700 being the premium AIB cards

Any higher than that then well ... 0 generational performance increase based on what we can buy last gen cards for atm.

I expect the starting price to be 700 though tbh.

gonna be a bad card of true, would have been better buying previous gen.

Joxeon · 24 Jan 2023 at 13:38

Andorphine said:
The rumors so far were that N32 top die (7800XT) would come fairly close to the 7900XT, because AMD would fix a bug that makes N31 not clock/perform properly. The newest rumors suggest that N32 will be worse than expected. I'd expect 6950XT performance in rasterization.

I’m expecting it’ll be 0-10% better than 6800XT and cost $700

humbug · 24 Jan 2023 at 13:51

Joxeon said:
I’m expecting it’ll be 0-10% better than 6800XT and cost $700

Per shader RDNA3 is no better than RDNA2, Navi 32 has the same shader count as the RX 6800, the 6800XT is 15% faster than that so they would have to push the clocks up 20% just to match the 6800XT. that would be 2550Mhz, that's 150Mhz higher than the 7900XT, to get it to 6800XT +10% it would have to be running at about 2900Mhz.

Its not looking good, its under performing and even that running at 2.9Ghz for 6800XT + 10% (which is optimistic) at $699 is DOA.

Zarax · 24 Jan 2023 at 13:57

humbug said:
Per shader RDNA3 is no better than RDNA2, Navi 32 has the same shader count as the RX 6800, the 6800XT is 15% faster than that so they would have to push the clocks up 20% just to match the 6800XT. that would be 2550Mhz, that's 150Mhz higher than the 7900XT, to get it to 6800XT +10% it would have to be running at about 2900Mhz.

Its not looking good, its under performing and even that running at 2.9Ghz for 6800XT + 10% (which is optimistic) at $699 is DOA.

We can only hope N31 had buggy silicon and it's fixed in N32.

humbug · 24 Jan 2023 at 14:02

Zarax said:
We can only hope N31 had buggy silicon and it's fixed in N32.

A silicon re-tape? Changing the transistor architecture of the silicon and then re-taping it is not cheap or that quick, but if they can't fix it in drivers they will have to, or just sell them for what they are worth, there is nothing wrong with the GPU's, they have a value for which they will sell at.

For the 7900XT that's $749 and the 7800XT its $549, i'm happy to buy the 7800XT for £550.

Joxeon · 24 Jan 2023 at 14:04

humbug said:
Per shader RDNA3 is no better than RDNA2, Navi 32 has the same shader count as the RX 6800, the 6800XT is 15% faster than that so they would have to push the clocks up 20% just to match the 6800XT. that would be 2550Mhz, that's 150Mhz higher than the 7900XT, to get it to 6800XT +10% it would have to be running at about 2900Mhz.

Its not looking good, its under performing and even that running at 2.9Ghz for 6800XT + 10% (which is optimistic) at $699 is DOA.

AMD have made a right dogs breakfast out of RDNA3, all they had to do was release the 7900XT as a 7800XT at $700 then the chip that’s now going to be a 7800XT as a 7700XT at $500 and it would have looked like a solid generation.

Darujhistan · 24 Jan 2023 at 14:04

Besty said:
this guy creates his own narrative by listing a number of hardware bugs listed in the driver notes, some of which are probably not even bugs. Says there will be a re-tape then, there is no re-tape.

people quote this guys tweets in almost every tech forum going.

Bencher or Nexus with the OTT bad English to throw people off the scent?

humbug · 24 Jan 2023 at 14:05

Joxeon said:
AMD have made a right dogs breakfast out of RDNA3, all they had to do was release the 7900XT as a 7800XT at $700 then the chip that’s now going to be a 7800XT as a 7700XT at $500 and it would have looked like a solid generation.

Yes. Exactly.

Joxeon · 24 Jan 2023 at 14:09

humbug said:
Yes. Exactly.

It makes you wonder what they would have done if Nvidia had released the recent 4080 for $700 and a 4070 for $500.

Darujhistan · 24 Jan 2023 at 14:09

Joxeon said:
AMD have made a right dogs breakfast out of RDNA3, all they had to do was release the 7900XT as a 7800XT at $700 then the chip that’s now going to be a 7800XT as a 7700XT at $500 and it would have looked like a solid generation.

This gen is garbage all around.