Discussion AMD Cezanne/Zen 3 APU Speculation and Discussion

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

dr1337

Senior member
May 25, 2020
309
503
106
Fresh leak out today, not much is known but at least 8cu's is confirmed. Probably an engineering sample, core count is unknown and clocks may not be final.

This is very interesting to me because cezanne is seemingly 8cu only, and it seems unlikely to me that AMD could squeeze any more performance out of vega. A cpu only upgrade of renoir may be lackluster compared to tigerlake's quite large GPU.

What do you guys think? Will zen 3 be a large enough improvement in APU form? Will it have full cache? Are there more than 8cus? Has AMD truly evolved vega yet again or is it more like rdna?
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
If Rembrant is really 208mm2, then the known specs don't match. As Seronx mentioned, the SoC in Xbox series S is only 197mm2(8 Zen2 cores, 24CU RDNA2, 128bit GDDR6) and Renoir is only ~150mm2.
There could be more cores, which is very unlikely considering Zen3 has 8 cores per CCX and the size is small for 16 cores(2x CCX) or a lot more L3 cache. I don't think Zen3 core is so much bigger than Zen2 to explain the difference in size.
 
Last edited:

Kedas

Senior member
Dec 6, 2018
355
339
136

Hans de Vries

Senior member
May 2, 2008
321
1,018
136
www.chip-architect.com
From Twitter picked op by VideoCardz:


Cézanne and Lucienne specs:

AMD_5000_APU.JPG
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
Definitely sounds more like 12 CU than the 24 CU that Nosta insists to be the case.

That sounds correct to me, 24CU RDNA2 is designed for 128bit GDDR6, so 12CU for 128bit DDR5 sounds about right. Altrought it may be a little more, maybe 14CU if it really is 208mm2.

I think they are aiming for RX570 perf.
 

mikk

Diamond Member
May 15, 2012
4,111
2,105
136
So there are three Zen 2 SKUs and three Cezanne SKUs if this is true. This is super confusing and the 0815 user has no chance to differentiate between Zen 2 and Zen 3 devices when they all use the 5000U branding. And why do they need to mix between Zen 2 and Zen 3? Not enough capacity, fool the mainstream buyer or why?
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,570
146
Wonder if Cezanne is actually on N7+? Unlike Vermeer, it actually has improved base clocks over Zen 2 counterparts.

Could explain why Lucienne also exists. N7+ capacity requires time with the EUV machines, which will be shared with Apple.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
Wonder if Cezanne is actually on N7+? Unlike Vermeer, it actually has improved base clocks over Zen 2 counterparts.

Could explain why Lucienne also exists. N7+ capacity requires time with the EUV machines, which will be shared with Apple.
N7+ is at Fab 15B while N5 is at Fab 18A.
N7+ is shares its floor space with N10, N7, N7+, N6.

N5 will not share fab capacity with any node except cost-effective RTO N4 and for N3/N2 up to 0.9v DKs before they switch over to their own fabs.

Apple doesn't use Fab 15 EUV machines(N7+/N6), instead uses Fab 18 EUV machines(N5). Which they only are sharing with an unspecified HPC customer.

C. C. Wei - Q1 2020
"But all I can say is now in N5, we have customer from -- smartphone; customer from HPC-related area"

The only two products to be on N7+ is Ascend 910 and Kirin 990 5G. Everyone else has moved to 2nd wave 7nm family node: N6 and 1st wave 5nm family node: N5.
 
Last edited:

exquisitechar

Senior member
Apr 18, 2017
655
862
136
Wonder if Cezanne is actually on N7+? Unlike Vermeer, it actually has improved base clocks over Zen 2 counterparts.

Could explain why Lucienne also exists. N7+ capacity requires time with the EUV machines, which will be shared with Apple.
Interesting. I can see this being true. Why would Lucienne even exist as it does otherwise?
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
Maybe Lucienne and Cezanne are on the same base process, but use different track libraries. Cezanne uses denser libraries to fit it's increased transistor count into the same die space, but Lucienne uses a relaxed track library set with minor circuit and firmare tweaks. Also, with N7 being more mature, they are seeing better yield fractions, so they can have higher CU counts per performance level.

I think Ice Lake performing as well as it did has affected their binning tiers for the CUs. Tiger Lake was too recently uncovered to go into that calculus in my opinion.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
There is still a notable generational improvement at each SKU. The odd numbered SKUs gain: added boost clock, one more CU, and HT. The even numbered SKUs gain more L3 and the Zen3 core improvements. Going up from an even to the next odd, you gain one CU and two more cores, so at least an MP/MT improvement and an iGPU performance. The only place where it will be a bit situational is in single threaded benchmarks or things that are case sensitive between 4MB and 16 MB.
 

Gideon

Golden Member
Nov 27, 2007
1,608
3,570
136
Cezanne even has twice the L3 cache based on these leaks
Interesting, haven't noticed that, where was this mentioned?

EDIT: Never mind, opened the twitter post (preview truncates the L3 part).
16MB unified L3 from 4MB per CCX should improve ST perfomance a lot (on top of other Zen 3 gains). It also IMO hints at 7NP or 7N+ as otherwise such a cache increase could negatively affect battery life (especially as it also as higher base-clocks).
 
  • Like
Reactions: Tlh97

Olikan

Platinum Member
Sep 23, 2011
2,023
275
126
EDIT: Never mind, opened the twitter post (preview truncates the L3 part).
16MB unified L3 from 4MB per CCX should improve ST perfomance a lot (on top of other Zen 3 gains). It also IMO hints at 7NP or 7N+ as otherwise such a cache increase could negatively affect battery life (especially as it also as higher base-clocks).

An interesting review would measure the Igp performance by the bandwidth savings of the cpu cache...
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
I doubt that the enlarged cache I'd directly responsible for much of a change in iGPU performance. In essence, from what I've read and seen, the iGPU is responsible for 90+% of the memory reads on the controller while gaming. Texture churn is huge, which is why modern cards have such massive memory throughput. I'm firmly in the camp that it makes a big change for the cpu side of things, but not as much for the iGPU.
 
  • Like
Reactions: Tlh97 and moinmoin

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
If they can get better volume with these, maybe they can better fill out the bottom of the product stack this time? If we take the U numbers as a base of comparison, the iGPU should see another 5-10% performance uplift.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136