Question Speculation: RDNA2 + CDNA Architectures thread

uzzi38 · Apr 28, 2020

All die sizes are within 5mm^2. The poster here has been right on some things in the past afaik, and to his credit was the first to saying 505mm^2 for Navi21, which other people have backed up. Even still though, take the following with a pich of salt.

Navi21 - 505mm^2

Navi22 - 340mm^2

Navi23 - 240mm^2

Source is the following post: https://www.ptt.cc/bbs/PC_Shopping/M.1588075782.A.C1E.html

DDH · Oct 23, 2020

Konan said:
Possibly. Apparently these scores are from the XTX ES on beta driver (according to KittyCorgi) and limited in TS and TSE due to shader.

The user said they are from an ES of an XT.

Stuka87 · Oct 23, 2020

Glo. said:
I would presume so. Aren't those the halo SKUs from Sapphire, and have highest priority in development?

Sapphire currently makes Nitro+ and Pulse designs. Nitro+ for the 5x00 series has big the OC model with the big cooler. The pulse is the more cost effective model with a dual fan config.

DDH · Oct 23, 2020

Glo. said:
AMD Radeon RX 6900XT to feature Navi 21 XTX GPU with 80 CUs - VideoCardz.com

AMD Radeon RX 6000 series, what we know so far AMD Radeon RX 6000 series specifications have been known to us since Monday. This week AMD provided a lot of details to AIBs, including new (working) drivers, final SKU names, and so on. We have been able to cross-check parts of the specs and we […]

videocardz.com

This is the most believable leak, because its based on what AIB's have got from AMD.

So if its 6800XT - according to this has 72 CUs.

Says info was provided by aibs who were given by amd. Could be true, but also might not be.

Still doesn't make sense as to why the 6800xt is a cut down die, why AIB partners would only get a cut down die when most dies produced would be full 80cu dies, why there is no 6900 non xt and why there would be a 20% cut down die(64cu). Also historically a xxXT die is the full die with XTX being a higher binned one. Why change now, when 21xt should be 80cu, 21xtx higher binned 80cu and xl(xe) should be 72cu

Konan · Oct 23, 2020

DDH said:
The user said they are from an ES of an XT.

Yes but No? 😉
sounds like the best card results on a ES with beta driver though

https://twitter.com/x/status/1319506032712806400

Kuiva maa · Oct 23, 2020

Glo. said:
6800XT is the 72 CU part.

If that's the case and AMD is not releasing the full 80CU for custom designs, I can think of only one outcome, and we have seen it before. This is a repeat of Fiji launch which had no custom 64CU models. All full Fiji chips were either AIO Fury-X or single fan Nano. Assuming the bus info and the lack of HBM2 models is true, we will not be having a nano this time. What's left is a triple axial reference 80CU XT model and maybe a premium AIO XTX.

DDH · Oct 23, 2020

Konan said:
No.

https://twitter.com/x/status/1319506032712806400

While he said that i don't think it's likely. That leak would have to come from AMD direct and considering how tight lipped they've been i doubt they will leak numbers 4 days away from reveal. The numbers are too well rounded too, could just be estimates

Konan · Oct 23, 2020

Well they have an excellent track record.
I’m hopeful.

80CU + XTX = 6900XT

AMD Radeon RX 6900 XT – Navi 21 XTX AMD’s Big Navi will feature 80 Compute Units (5120 Stream Processors). This card will be (for now) AMD exclusive. The Radeon RX 6900XT will be the AMD flagship series, which according to our sources will be in limited quantity. The card would allegedly feature 16GB of GDDR6 memory across a 256-bit bus. The graphics card would feature a 2040 MHz game clock and a 2330 MHz boost clock

kurosaki · Oct 23, 2020

DDH said:
Is it though. That really doesn't make a whole lot of sense

Why?

Glo. · Oct 23, 2020

DDH said:
Is it though. That really doesn't make a whole lot of sense

I agree. IF 72 CU chip is getting 13k PTS in FireStrike, RDNA2 is going to run circles around Ampere. And we already have seen that AIB's XT(supposedly) is getting around 11500 pts in FSU.

So those scores are either for full die, with 80 CUs, or, 72 CUs, which would make the already leaked 11500 pts score a 64 CU version.

Its a complete mess...

Timorous · Oct 23, 2020

Glo. said:
I agree. IF 72 CU chip is getting 13k PTS in FireStrike, RDNA2 is going to run circles around Ampere. And we already have seen that AIB's XT(supposedly) is getting around 11500 pts in FSU.

So those scores are either for full die, with 80 CUs, or, 72 CUs, which would make the already leaked 11500 pts score a 64 CU version.

Its a complete mess...

Can't they be at different clocks and power limits?

What is the spread on other cards between stock and OC variants?

CastleBravo · Oct 23, 2020

Glo. said:
I agree. IF 72 CU chip is getting 13k PTS in FireStrike, RDNA2 is going to run circles around Ampere. And we already have seen that AIB's XT(supposedly) is getting around 11500 pts in FSU.

So those scores are either for full die, with 80 CUs, or, 72 CUs, which would make the already leaked 11500 pts score a 64 CU version.

Its a complete mess...

At the very least, there shouldn't be any more claims of Navi21=3070.

Zstream · Oct 23, 2020

Glo. said:
I agree. IF 72 CU chip is getting 13k PTS in FireStrike, RDNA2 is going to run circles around Ampere. And we already have seen that AIB's XT(supposedly) is getting around 11500 pts in FSU.

So those scores are either for full die, with 80 CUs, or, 72 CUs, which would make the already leaked 11500 pts score a 64 CU version.

Its a complete mess...

That makes no sense. FireStrike is rarely bandwidth limited vs. Time Spy. It just shows that the 256 bit bus, if true, is a bottleneck.

Glo. · Oct 23, 2020

https://twitter.com/x/status/1319648356029923328

So yeah, this score of 11500 FSU was actually 6800XT AIB.

Timorous · Oct 23, 2020

Glo. said:
https://twitter.com/x/status/1319648356029923328

So yeah, this score of 11500 FSU was actually 6800XT AIB.

But at what clocks. Just doing a very quick look of 3080 scores over at Guru3D there is quite a spread between the likes of the Galax / Zotax models and the mild OC models like the MSI Gaming X Trio.

Glo. · Oct 23, 2020

Timorous said:
Can't they be at different clocks and power limits?

What is the spread on other cards between stock and OC variants?

Well, there is a chance, small, very small, that its due to newest drivers, that AMD dropped just recently.

Edit: Aaaand, according to latest Igor's info, that is just what happened. O_O

What is happening. This is a complete mess.

raghu78 · Oct 23, 2020

RGT who has been the most accurate of all leakers till now has been saying for now over a month that 6800XT competes with RTX 3080 and 6900XT with RTX 3090 in rasterization. In ray tracing benchmarks Ampere is ahead. But that is also a function of Nvidia RTX being out for 2 years while RDNA2 is a brand new RT implementation. Give it sometime and with the new game consoles using RDNA2 even on RT AMD will not be far behind. RDNA3 is when I think AMD will beat Nvidia across RT and take the GPU crown without a question.

Timorous · Oct 23, 2020

Glo. said:
Well, there is a chance, small, very small, that its due to newest drivers, that AMD dropped just recently.

Edit: Aaaand, according to latest Igor's info, that is just what happened. O_O

What is happening. This is a complete mess.

Fan meet poo.

Krteq · Oct 23, 2020

What a silly season

Glo. · Oct 23, 2020

So, this is getting crazier and crazier with each day.

According to Igor's info, AMD has given AIBs the latest driver, and that driver alone brought 10% more performance, and that 13k Score from Igor is from 6800XT with latest driver.

O_O This RTX 3090 levels of performance, isn't it?

WCCFTech got similar scores from one of AMD partner, to Igor: they've got 12,8k, so if the rumor about latest driver is true, it might have been with latest driver.

Glo. · Oct 23, 2020

Krteq said:
What a silly season

Its getting way too crazy to be believable. I would understand this if this was coming from random noname-twitter accounts.

But this is from very reputable sources, that have AIB connections, and all of their info is coming from the AIBs.

Krteq · Oct 23, 2020

Glo. said:
According to Igor's info, AMD has given AIBs the latest driver, and that driver alone brought 10% more performance, and that 13k Score from Igor is from 6800XT with latest driver.

Any link?

dzoni2k2 · Oct 23, 2020

It's 3DMarks though. It's gonna be slightly worse in actual games.

Glo. · Oct 23, 2020

Krteq said:
Any link?

Yeah, its German.

Vope45 · Oct 23, 2020

raghu78 said:
RGT who has been the most accurate of all leakers till now has been saying for now over a month that 6800XT competes with RTX 3080 and 6900XT with RTX 3090 in rasterization. In ray tracing benchmarks Ampere is ahead. But that is also a function of Nvidia RTX being out for 2 years while RDNA2 is a brand new RT implementation. Give it sometime and with the new game consoles using RDNA2 even on RT AMD will not be far behind. RDNA3 is when I think AMD will beat Nvidia across RT and take the GPU crown without a question.

At this point I think we all know who leaks infos to RGT.

dzoni2k2 · Oct 23, 2020

Vope45 said:
At this point I think we all know who leaks infos to RGT.

Sapphire?

Question Speculation: RDNA2 + CDNA Architectures thread

Platinum Member

Member

Diamond Member

Member

Senior member

Member

Member

Senior member

Senior member

Diamond Member

Golden Member

Member

Diamond Member

Diamond Member

Golden Member

Diamond Member

Diamond Member

Golden Member

Golden Member

Diamond Member

Diamond Member

Golden Member

Member

Diamond Member

Member

Member