Question AMD Rembrandt/Zen 3+ APU Speculation and Discussion

Page 15 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
I

izaic3

Member
Nov 19, 2019
61
96
91
Alright, so we've had some leaks so far. I don't know if any of it's been confirmed yet, as it's pretty early, but here is what I've surmised so far (massive grain of salt of course):

If if turns out to have RDNA 2 and 12 CU, I could see iGPU performance potentially almost doubling over Cezanne.

If I've made any mistakes or gotten anything wrong, please let me know. I'd also love to hear more knowledgeable people weigh in on their expectations.
 
Last edited:
  • Like
Reactions: Kaluan, Tlh97, Vattila and 4 others
Sort by date Sort by votes
K

Kepler_L2

Senior member
Sep 6, 2020
331
1,162
106
Spicy said:
Really? What makes you say that?
PS5 has Navi 2, not Navi 1, and this driver is very recent (committed some mounths ago).
Notebookcheck has the same intuition.
Maybe it won't be Cyan Skillfish, maybe RDNA2, but I still think a light version of it (few CUs).
Click to expand...
If you check the Linux SMU patches, you will see it's an 8 Core Zen 2 APU. Also LLVM shows gfx1013 (Cyan skillfish) has RT support.
 
  • Like
Reactions: uzzi38, Tlh97 and scineram
Upvote 0 Downvote
D

Doug S

Platinum Member
Feb 8, 2020
2,254
3,485
136
If anyone could benefit from stacked cache it would seem to be Ampere. Maybe they don't care if the small SLC that hurts them in some benchmarks doesn't really impact their target cloud market. But stick a 7nm die the size of the 128 core chip that's entirely SRAM and that SLC goes from 16MB to 1GB for only a modest increase in latency.
 
Upvote 0 Downvote
uzzi38

uzzi38

Platinum Member
Oct 16, 2019
2,625
5,894
146
Doug S said:
If anyone could benefit from stacked cache it would seem to be Ampere. Maybe they don't care if the small SLC that hurts them in some benchmarks doesn't really impact their target cloud market. But stick a 7nm die the size of the 128 core chip that's entirely SRAM and that SLC goes from 16MB to 1GB for only a modest increase in latency.
Click to expand...

They don't care at all. Their focus is not per-core performance - you'll never see them do that frankly.
 
Upvote 0 Downvote
L

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
Let's not forget: The desktop 3400g had 11 CUs and could routinely clock north of 1600mhz. 12 CUs of RDNA 2, while being more efficient and likely clocked above 2Ghz in some parts, isn't a massive uplift. Assuming that the DDR5 can keep it fed, I'd still be shocked if it consistently reached RX560/1050 levels of in game performance.
 
  • Like
Reactions: lightmanek
Upvote 0 Downvote
T

TESKATLIPOKA

Platinum Member
May 1, 2020
2,355
2,848
106
LightningZ71 said:
Let's not forget: The desktop 3400g had 11 CUs and could routinely clock north of 1600mhz. 12 CUs of RDNA 2, while being more efficient and likely clocked above 2Ghz in some parts, isn't a massive uplift. Assuming that the DDR5 can keep it fed, I'd still be shocked if it consistently reached RX560/1050 levels of in game performance.
Click to expand...
RDNA2 vs Polaris(Vega) is 33-40% better per clock at the same number of CU compared to Polaris(Vega).
A desktop 12CU RDNA2 can clock easily to at least 2.5GHz.
This all adds to more than 2x better performance. If this is not a massive uplift then I don't know what is.
Of course bandwidth could be a problem, but If they couldn't feed It properly, then they wouldn't put 12CU inside Rembrandt.

Can It beat RX560/1050 levels of game performance?
3400g = 11CU(704SP) at 1600Mhz -> 2250 GFLOPs
RX560 = 16CU(1024:64:16) at 1175(1275) MHz -> 2406(2611) GFLOPs
GTX 1050 = 5SM (640:40:32 ) at 1354 (1455 ) MHz -> 1733 (1862) GFLOPs
Rembrandt = 12CU(768SP) at 2.5GHz -> 3840 GFLOPs
If It can't beat It, I would be surprised.
 
Last edited:
  • Like
Reactions: Tlh97, scineram, moinmoin and 2 others
Upvote 0 Downvote
L

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
Your own chart works against your point. The RX 560, despite having a near 50% GFlop advantage over the 1050, is almost always beaten by the 1050 in game benchmarks, and its only wins are by very slim margins. The 3400G, despite having 90% of the GFlops of the RX 560, is almost never even close to meeting the 560 in game benchmarks. Bandwidth starvation is a big deal.

RDNA2 is definitely more efficient in using VRAM, but, is it good enough to overcome the limits imposed by dual-dual channel DDR5, the actual bandwidth of which won't be more than 50% greater than the 3200 DDR4, and the few LPDDR4X 4333 implementations that we've already seen? I hope it's something special, but I'm not holding my breath, especially since they chose not to include any infinity cache for it, which seems to be a large part of why RDNA2 does so well with limited bandwidth in other GPUs.
 
Upvote 0 Downvote
uzzi38

uzzi38

Platinum Member
Oct 16, 2019
2,625
5,894
146
It might not, idk. Early rumours make it sound rather meh, we'll have to see, it could be memory bandwidth is a much bigger issue than first expected.
 
Upvote 0 Downvote
T

TESKATLIPOKA

Platinum Member
May 1, 2020
2,355
2,848
106
LightningZ71 said:
Your own chart works against your point. The RX 560, despite having a near 50% GFlop advantage over the 1050, is almost always beaten by the 1050 in game benchmarks, and its only wins are by very slim margins. The 3400G, despite having 90% of the GFlops of the RX 560, is almost never even close to meeting the 560 in game benchmarks. Bandwidth starvation is a big deal.
Click to expand...
It doesn't work against my point, It just shows GFLOPs, but doesn't take into account the GPU architecture or bandwidth. For example as I already mentioned RDNA2 is 33-40% faster than Polaris at the same GFLOPs.
Pascal is also faster than Polaris at the same GFLOPs, not to mention Pascal clocks higher than what It's official boost suggests, just take a look at some old Pascal GPU reviews.
You mention 3400g having 90% GFLOPs of RX560, but that's not true. I used 1.6GHz, because you said It can clock that high, but officially It has only 1.4GHz or 1971GFLOPs and that's only 76% of RX 560 boost.

If I compare Rembrandt IGP vs the new RX 6600, then Rembrandt IGP would need 6000Mhz DDR5 to have the same GFLOPs/BW ratio, but that's at ~2.45GHz.
4800Mhz DDR5 is good enough for 1960Mhz IGP clockspeed. The problem is RX6600 has Infinity cache, but Rembrandt won't have any.
 
Last edited:
  • Like
Reactions: Tlh97 and BorisTheBlade82
Upvote 0 Downvote
E

eek2121

Platinum Member
Aug 2, 2005
2,930
4,026
136
FYI, I've been hearing that the upcoming Rembrandt release is OEM only and consumer parts (including AM5) will NOT be released.

EDIT: That is, consumer AM5 boards will be introduced with Zen 4.
 
Last edited:
  • Like
Reactions: BorisTheBlade82
Upvote 0 Downvote
A

ahimsa42

Senior member
Jul 16, 2016
225
194
116
eek2121 said:
FYI, I've been hearing that the upcoming Rembrandt release is OEM only and consumer parts (including AM5) will NOT be released.

EDIT: That is, consumer AM5 boards will be introduced with Zen 4.
Click to expand...

makes sense since Rembrandt is a mobile CPU so will be for laptops, all in one desktops & mini PC's.
 
Upvote 0 Downvote
T

TESKATLIPOKA

Platinum Member
May 1, 2020
2,355
2,848
106
ahimsa42 said:
the performance of the steam deck bodes well so hoping to see rembrandt U at worst = to rx 560, at best = to rx 570
Click to expand...
Rembrandt U will certainly not perform as RX 570.
Even a desktop part wouldn't perform as high.
Rannar could be pretty close with his estimate.
IGP performance crown will be once again held by AMD.
 
Upvote 0 Downvote
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom