Question Speculation: RDNA2 + CDNA Architectures thread

Page 79 - 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
 

Glo.

Diamond Member
Apr 25, 2015
5,803
4,777
136
According to VideoCardz:

  1. Navi 21: GFX1030 - RX 6900 XT Sienna Cichlid (I hate typing that)
  2. Navi 22: GFX1031 (Mac Only, RX 6600M)
  3. Navi 23: GFX 1032: RX 6500 series
  4. Van Gogh: GFX 1033 (Zen2, RDNA2)
The information VideoCardz has could be inaccurate, but if they are right, Navi24 -> GFX1034 -> RX 6400/RX 6300.

According to the drivers, right now everything is 1030.
Considering that they marked N22 as Mac only, I'd say pretty much is wrong.

As I have said.

80 CU die - X900 SKUs
60 CU die - X800 SKUs
40 CU die - X700/X600 SKUs
smaller than that - X500 SKUs.
 

eek2121

Diamond Member
Aug 2, 2005
3,100
4,398
136
Them both being marked gfx1030 is very likely just simplicity. Note that this is just the "CPU name" for LLVM. If both GPUs have the exact same instruction set, why bother calling out all the revisions? Just do the simplest thing that works, which is reusing gfx1030. VanGogh is a clear example of this.

Well, they would identify the specific revisions for cases that involve chip specific changes. If the chip behaves exactly the same and needs no further optimization, sure, it is fine to identify them all as the same.

Also, an interesting comment in a PR submitted for review today:

/* It's wasteful to enable all CUs for PS if shader arrays have a different
* number of CUs. The reason is that the hardware sends the same number of PS
* waves to each shader array, so the slowest shader array limits the performance.
* Disable the extra CUs for PS in other shader arrays to save power and thus
* increase clocks for busy CUs. In the future, we might disable or enable this
* tweak only for certain apps.
*/

So it sounds like they are more aggressively power gating to allow for higher clocks. I believe I mentioned some time ago that we'd see this.
 
  • Like
Reactions: lightmanek

eek2121

Diamond Member
Aug 2, 2005
3,100
4,398
136
Considering that they marked N22 as Mac only, I'd say pretty much is wrong.

As I have said.

80 CU die - X900 SKUs
60 CU die - X800 SKUs
40 CU die - X700/X600 SKUs
smaller than that - X500 SKUs.

We'll see. You are basically saying that the 6700XT is the leaked card today and will feature no performance improvements at all. EDIT: Unless a 2.3-2.4 GHz clock is enough to overcome the limitations of a 192 bit bus size for memory. The RX 5600XT did happen to scale really well with it's bus, after all.
 

Glo.

Diamond Member
Apr 25, 2015
5,803
4,777
136
We'll see. You are basically saying that the 6700XT is the leaked card today and will feature no performance improvements at all. EDIT: Unless a 2.3-2.4 GHz clock is enough to overcome the limitations of a 192 bit bus size for memory. The RX 5600XT did happen to scale really well with it's bus, after all.
That 40 CU die will be formidable opponent for RTX 3070. VERY formidable ;).
 

eek2121

Diamond Member
Aug 2, 2005
3,100
4,398
136
That 40 CU die will be formidable opponent for RTX 3070. VERY formidable ;).

Not with a 192-bit bus it won't. At least definitely not at 4K. NVIDIA is using a smaller bus size, but they have GDDR6X to make up for that. By all reports AMD is still using GDDR6. By comparison, the 1080ti and 2080ti both used a 352-bit bus.

According to NVIDIA:

1080ti -> 11gbps
2080ti -> 14gbps
3070 -> 16 gbps

AMD? 8gbps?

EDIT The RX 5700 XT apparently had 14gbps memory, so the 40CU RX 6000 equivalent will have 10 or 12 gbps.
 

Konan

Senior member
Jul 28, 2017
360
291
106
Navi 23 - RTX 2080 Super performance levels +10%, @150W TBP.
Navi 22 - 10-20% above RTX 2080 Ti performance levels @225 TBP.
Navi 21 - 40-50% above RTX 2080 Ti performance @ 275W TBP.

Those were performance targets for RDNA2, that I got hint few weeks ago.

What is the CU count next to each of these?
 

Glo.

Diamond Member
Apr 25, 2015
5,803
4,777
136
Navi 23 - RTX 2080 Super performance levels +10%, @150W TBP.
Navi 22 - 10-20% above RTX 2080 Ti performance levels @225 TBP.
Navi 21 - 40-50% above RTX 2080 Ti performance @ 275W TBP.

Those were performance targets for RDNA2, that I got hint few weeks ago.
What is the CU count next to each of these?
N21 - 80 CU
N22 - 60 CU
N23 - 40 CU.

I should also specify that I was specifically told the CU count/performance targets, and GPU die names are my addition, based on what I knew at that time.

I can also add what I have been told just today.

Quote, and read it as I write: "RX 5500 XT replacement will be roughly 60% faster than that GPU".
 

Kepler_L2

Senior member
Sep 6, 2020
537
2,198
136
Wild speculation: AMD might have actually went out of their way to troll/mislead leakers.

I believe that Navi23 (Dimgrey Cavefish) is actually the 80 CU, 505mm², 384b model we were expecting.

Now there is one main problem with this theory and that is Sienna Cichlid (80 CU, HBM2?) which most people assumed was Navi 21.

However it's common for some products to show up in drivers and never be released. Alternatively, Sienna could be a spin off of Navi 23 for the professional market.

Evidence of this theory include the Navi 21 die circulating on AIBs being much smaller than expected Navi 21 being 256b and Navi 22 192b (https://twitter.com/_rogame/status/1306655000454725636) and finally the fact that Dimgrey hexcode is #696969 (https://www.colorhexa.com/696969) because obviously Navi 23 is the 6900XT! ;)
 

jamescox

Senior member
Nov 11, 2009
644
1,105
136
Not with a 192-bit bus it won't. At least definitely not at 4K. NVIDIA is using a smaller bus size, but they have GDDR6X to make up for that. By all reports AMD is still using GDDR6. By comparison, the 1080ti and 2080ti both used a 352-bit bus.

According to NVIDIA:

1080ti -> 11gbps
2080ti -> 14gbps
3070 -> 16 gbps

AMD? 8gbps?

EDIT The RX 5700 XT apparently had 14gbps memory, so the 40CU RX 6000 equivalent will have 10 or 12 gbps.

That is where the “infinity cache“ makes up for the lower bandwidth, right? The statements about disrupting 4K gaming seem to indicate that they expect to do very well at 4K.
 
  • Like
Reactions: lightmanek

Ajay

Lifer
Jan 8, 2001
16,094
8,109
136
Wild speculation: AMD might have actually went out of their way to troll/mislead leakers.

I believe that Navi23 (Dimgrey Cavefish) is actually the 80 CU, 505mm², 384b model we were expecting.

Now there is one main problem with this theory and that is Sienna Cichlid (80 CU, HBM2?) which most people assumed was Navi 21.

However it's common for some products to show up in drivers and never be released. Alternatively, Sienna could be a spin off of Navi 23 for the professional market.

Evidence of this theory include the Navi 21 die circulating on AIBs being much smaller than expected Navi 21 being 256b and Navi 22 192b (https://twitter.com/_rogame/status/1306655000454725636) and finally the fact that Dimgrey hexcode is #696969 (https://www.colorhexa.com/696969) because obviously Navi 23 is the 6900XT! ;)
Wow, having a top card (N21) with only a 256b wide dram interface seems nuts; especially so if AMD isn't using GDDR6X.
 

CastleBravo

Member
Dec 6, 2019
120
271
136
With all the old rumors of Navi21 having both a GDDR6 controller and HBM IO built in, would it make sense to have mixed VRAM? Could they do 8/16GB 256bit GDDR6, and 4/8GB of HBM2E on the same card?
 

Macros96

Junior Member
Apr 27, 2017
6
6
81
Wow, having a top card (N21) with only a 256b wide dram interface seems nuts; especially so if AMD isn't using GDDR6X.
If we know this is nuts, then AMD knows this is nuts. So there are only so many scenarios:

1.) They are ceding the high end yet again. :rolleyes:
2.) They have some secret sauce compression or caching scheme up their sleeve.
3.) They have successfully avoided accurate leaks and there is a big-die plus HBM card at the top of the stack.

I'm fervently hoping for #3, but know it will be expensive.