Question Speculation: RDNA2 + CDNA Architectures thread

Page 143 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

uzzi38

Platinum Member
Oct 16, 2019
2,705
6,427
146
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
 

Veradun

Senior member
Jul 29, 2016
564
780
136
The announcement should at least clear up some of the confusion. As far as performance goes you'll want to wait for independent review, but I doubt AMD will get up on stage and stretch the truth beyond reality. Not sure when independent reviews will drop. Wasn't Nov 5th for Zen 3's reviews and availability?
"Remember, remember, the fifth of November"
 

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,249
136
Double crown?

In GPU Efficiency? Sure, I can believe that AMD will have more efficient architecture. But performance? Thats way tougher.

Double crown = trumping Intel's gaming dominance and Nvidia's for the record....The way I meant it.

It's pretty much unthinkable that they'd pull off a double whammy in the same year, but it's not like 2020 has stayed true to the norm.

The internet forums and article comment sections would go heavily into rage mode!
 

Glo.

Diamond Member
Apr 25, 2015
5,803
4,777
136
Double crown = trumping Intel's gaming dominance and Nvidia's for the record....The way I meant it.

It's pretty much unthinkable that they'd pull off a double whammy in the same year, but it's not like 2020 has stayed true to the norm.

The internet forums and article comment sections would go heavily into rage mode!
I knew perfectly well what you meant, just made a Thought Shortcut.

Of course. AMD winning in both: CPU and GPU world are not out of the equation, and it has absolutely nothing to do with the year we are in ;). I'd love to see that happen. Good story of a Tech Underdog kicking butts of Tech Giants.
 

DDH

Member
May 30, 2015
168
168
111
So theres something Im having doubts about and that is that the XTX die, AMD exclusive, is either the only full die (80CUS) or a higher binned full die.

If its a higher binned die, the performance delta over the remaining full dies wont be that large I imagine. The 5700xt lisa su edition was clocked higher out of the box but ultimately once overclocked it was the same as every other 5700xt. This leads me to believe that a higher binned full 80cu die would be similar, in that aib models might well clock and perform beyond it. and if thats the case why bother with the subterfuge of not letting aibs know about it.

If its the only full 80cu die and the partner cards are getting the 72cu cut down version, there would be far fewer cut down dies than full. Almost 2:1 based on TSMC's reported defect density of 0.09 and coreteks die size estimate. AMD arnt going to send full dies with limited bioses to meet demand. And the AIB would know its not a fully functional die. Again why bother with the subterfuge when it would probably be obvious. Also, with regards to the naming scheme, AMD have always historically called the full die of a segment the XT, and the cut down version the non-xt. Why would a cut down die be the 6800xt. This supply limitation would also be exasperated by needing to have dies cut down by 20% (64cu) to fit a 6800

Really the better solution would be to have a 64cu full die, with a 56 cut down die, and a 80cu die with a 72cu cut down die. This leads me to believe that the board partner cards are the 64cu 6800xt full die. The 56cu die may not yet have been sampled to board partners. This is also probably the 2 fan design weve seen

The 80cu and 72cu cut down are the amd exclusive, the 6900xt and 6900. That is the only thing that makes sense with regards to AMD not letting board partners have the chips. This would be the 3 fan design


Some may point to lisa su holding on stage a 3 slot card, calling it big navi and showing benches that are below to on par with the 3080. It may have been the 72cu version the benches were from. But big navi is just a bigger chip than navi10 die, 251mm2

Also I would add that the rumored size of navi22, 340mm2 is much too large for a 40cu die, infact it is more unaccounted for space than its larger sibling the 80cu 500mm2

One more point Id like to add, and that is the benches show at the zen3 event. I would suggest they are by a 64cu card running at 2.2ghz. We know the XBSX benches in gears5 10% slower than a 2080ti. A 64cu die would have 23% more cu's than the XBSX, and at 2.2 ghz it would have 20% higher clocks. So assuming 2080ti = 100, XBSX = 90, then 64cu 2.2ghz card would be 132(90x1.23x1.2). Ballpark 3080 figures
 

Glo.

Diamond Member
Apr 25, 2015
5,803
4,777
136
Some may point to lisa su holding on stage a 3 slot card, calling it big navi and showing benches that are below to on par with the 3080. It may have been the 72cu version the benches were from. But big navi is just a bigger chip than navi10 die, 251mm2

Also I would add that the rumored size of navi22, 340mm2 is much too large for a 40cu die, infact it is more unaccounted for space than its larger sibling the 80cu 500mm2

One more point Id like to add, and that is the benches show at the zen3 event. I would suggest they are by a 64cu card running at 2.2ghz. We know the XBSX benches in gears5 10% slower than a 2080ti. A 64cu die would have 23% more cu's than the XBSX, and at 2.2 ghz it would have 20% higher clocks. So assuming 2080ti = 100, XBSX = 90, then 64cu 2.2ghz card would be 132(90x1.23x1.2). Ballpark 3080 figures
We don't know what is on the Dies hence its hard to understand the die sizes that are rumored/have been leaked, compared to the Navi 10 die sizes.
 

lightmanek

Senior member
Feb 19, 2017
429
916
136
I can imagine XT die to be full 80CUs and AMD exclusive XTX to be HBM version as most HBM2 cards were AMD reference models anyway, just with different box. Exception being Vega 10.

That way I can see it, XTX being still 80CU but with +15% performance thanks to faster memory and higher available GPU power coming from lower HBM versus GDDR6 draw.

7 days to find out ...
 
  • Like
Reactions: Tlh97

Krteq

Senior member
May 22, 2015
993
672
136


Those leaked photos seems to belongs to some AIB OC model (beefed up VRM, higher PCB etc.).

Photo of reference card for comparison
75367_02_if-these-new15kxy.jpg
 

uzzi38

Platinum Member
Oct 16, 2019
2,705
6,427
146
Double crown = trumping Intel's gaming dominance and Nvidia's for the record....The way I meant it.

It's pretty much unthinkable that they'd pull off a double whammy in the same year, but it's not like 2020 has stayed true to the norm.

The internet forums and article comment sections would go heavily into rage mode!

Us lot in the Twitter etc have been joking about it as the "gaymercide" for a while now.

Though tbh, the general consensus bar a couple is we expect a competitive AMD on GPUs, not a fully dominant one. That alone is enough to set a lot of people off though
 

dr1337

Senior member
May 25, 2020
417
691
136
It has those same two screw holes left and right of the package that people thought indicated HBM rather than GDDR.
It also doesn't have any of the GDDR pads populated but the rest of the board is, could be bait, but that would seriously imply it employs HBM as well. Its just so unlikely that anybody would run up a production sample board fully finished but without any DRAM
 

CastleBravo

Member
Dec 6, 2019
120
271
136
It also doesn't have any of the GDDR pads populated but the rest of the board is, could be bait, but that would seriously imply it employs HBM as well. Its just so unlikely that anybody would run up a production sample board fully finished but without any DRAM

Interesting... if you have both GDDR and HBM products using the same die, would it make sense to use the same PCB for both? I guess you would also want to put something like Kapton tape over the GDDR pads to avoid shorting stuff out.
 

Stuka87

Diamond Member
Dec 10, 2010
6,240
2,559
136
It also doesn't have any of the GDDR pads populated but the rest of the board is, could be bait, but that would seriously imply it employs HBM as well. Its just so unlikely that anybody would run up a production sample board fully finished but without any DRAM

That would mean the same board is used for more than one GPU though, which would be strange. The GPU is either HBM, or GDDR6. Because the package has to change.

This is most likely just a dev board that was being used to test something else, or they were going to hand place the RAM chips (which can be done).
 

PhoBoChai

Member
Oct 10, 2017
119
389
106
Well... Keller probably had the worst draw of the straw if one considers funding/resources or lack thereof (especially compared to Intel) but nevertheless Ryzen was/is considered a roaring success.

AMD prioritized Zen development in 2013 to 2017 period. With whatever little funding they had left. Right call too.
That has been coming for a long time, since Vega days? Glad it's getting somewhere now.

Informative response from the comments:

The whole point about NGG is to enable Primitive Shaders or Mesh Shaders to speed up geometry processing. NGG by itself doesn't speed up anything. It has to be used by devs!
 

PhoBoChai

Member
Oct 10, 2017
119
389
106
What I meant is N7 and N7+ share equipment, so moving a line to EUV removes capacity from the former to increase the latter, making zen3 and rdna2 battling for capacity anyway.

It's the same production line, just with additional EUV steps, so no its not going to create additional capacity. People aren't using the EUV step on 7N for some reason.. probably low yields?