Question AMD Rembrandt/Zen 3+ APU Speculation and Discussion

Page 34 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

izaic3

Member
Nov 19, 2019
61
96
61
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:

DisEnchantment

Golden Member
Mar 3, 2017
1,590
5,722
136
If desktop Rembrandt is due for early H2 2022 (as with Cezanne) then it's normal to see some early platform sample. I think it could be entirely possible that AMD wants to "polish" AM5 platform before launching Zen4.
Lilac RMB/FP7r2 with DDR5 is already showing up for a bit now being tested on Ubuntu 20.4 LTS

I think the Durango cIOD has more in common with Rembrandt than with a cut up Floyd, not to mentioned being fabbed on same process.
Many parts of the supposed Durango cIOD will already see action on RMB first, like the new DDR5 MC, iGPU/Yellow Carp, SFH, Thunderbolt, etc.
You can already see these in the dmesg log and the lspci output

What I have not been able to find are the C5/Q6 DSP. Don't know how they are exposed to the Kernel. (For example, Movidus VPU is exposed as PCIe endpoint)
On Android they are exposed via the NN API not sure how it will be on Linux and Windows.

Similarly, support for the ISP is already there in the amd-pmc driver for YC but not sure which device is it.
The AOP would probably come via the amd-pmc driver too but have not seen anything yet.

It is likely that Raphael will see a lot of issues already ironed out before it launches within Rembrandt bring up work itself.
Microcode will be a lot more different though

Update:
Seems YC SMU is missing the ISP, CVML and A55 block feature flags although CVML was mentioned in a leak by VCZ
Let's see if there is a change in the code in the future otherwise RMB might not have these features
VGH is indeed a custom mobile oriented APU unlike RMB if that's the case.
 
Last edited:

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136

If the MX550 is GTX1660 level there is no way for RMB IGP to land anywere near, specially in mobile. Unless 12CU RDNA2 with just 112GB/s perform waaaaay beyond expected.

We are talking about GTX1050TI level as realistic, GTX1650 non-super level as optimistic and RX570 8GB level as the best case scenario. GTX1660 is well past all that.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
If the MX550 is GTX1660 level there is no way for RMB IGP to land anywere near, specially in mobile. Unless 12CU RDNA2 with just 112GB/s perform waaaaay beyond expected.

You mean the 2050/mx570. The mx550 should easily be slower than Rembrandt, that's meant to be an upgrade for Alder Lake. Bandwidth should be roughly around 112 GB/sec so it's intentionally bandwidth starved as well.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,572
146
If the MX550 is GTX1660 level there is no way for RMB IGP to land anywere near, specially in mobile. Unless 12CU RDNA2 with just 112GB/s perform waaaaay beyond expected.

We are talking about GTX1050TI level as realistic, GTX1650 non-super level as optimistic and RX570 8GB level as the best case scenario. GTX1660 is well past all that.
Pretty sure that's a typo. The score was 2500pts.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,572
146
You mean the 2050/mx570. The mx550 should easily be slower than Rembrandt, that's meant to be an upgrade for Alder Lake. Bandwidth should be roughly around 112 GB/sec so it's intentionally bandwidth starved as well.
They all have the same memory bandwidth don't they?. And btw 112GB/sec is more than RMB has available to it. Pretty sure RMB is sec, LPDDR5-5500 support.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
You mean the 2050/mx570. The mx550 should easily be slower than Rembrandt, that's meant to be an upgrade for Alder Lake. Bandwidth should be roughly around 112 GB/sec so it's intentionally bandwidth starved as well.
I wouldn't be so sure about MX550 being slower, but It shouldn't be much faster.
MX550 : TU117, 1024 CUDA, 64bit 2G GD6 12Gbps, that's basically a bandwidth starved mobile 1650 with limited TDP.
Mobile Rembrandt would need to have 1/3 higher clocks than this to be on par in GFLOPs, and that still doesn't mean actual performance would be the same.
So If MX550 boosts to 1.5GHz then Rembrandt would need 2GHz boost for example.
RDNA2 is more efficient than Vega, so a 2GHz IGP boost for 45W Rembrandt shouldn't be impossible If the architecture is at least 50% more efficient, considering 45W Cezanne is boosting up to 2.1GHz.
 
Last edited:
  • Like
Reactions: Tlh97

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
RDNA2 is more efficient than Vega, so a 2GHz IGP boost for 45W Rembrandt shouldn't be impossible If the architecture is at least 50% more efficient, considering 45W Cezanne is boosting up to 2.1GHz.

768 RDNA2 SPs shouldnt exceed 25W@2GHz if we are to extrapolate from GPUs like the RX6600.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
They all have the same memory bandwidth don't they?. And btw 112GB/sec is more than RMB has available to it. Pretty sure RMB is sec, LPDDR5-5500 support.

Oh I missed something. DC DDR5 4800 is 77 GB/sec, the mx570 and mx550 is max 96 GB/sec although I wonder if most will even use 12 Gbps. There's going to be titles where the 2 GB of GDDR6 is simply not going to be enough for 1080p min. Something Rembrandt should have less issues with.
 

andermans

Member
Sep 11, 2020
151
153
76
I wouldn't be so sure about MX550 being slower, but It shouldn't be much faster.
MX550 : TU117, 1024 CUDA, 64bit 2G GD6 12Gbps, that's basically a bandwidth starved mobile 1650 with limited TDP.
Mobile Rembrandt would need to have 1/3 higher clocks than this to be on par in GFLOPs, and that still doesn't mean actual performance would be the same.
So If MX550 boosts to 1.5GHz then Rembrandt would need 2GHz boost for example.
RDNA2 is more efficient than Vega, so a 2GHz IGP boost for 45W Rembrandt shouldn't be impossible If the architecture is at least 50% more efficient, considering 45W Cezanne is boosting up to 2.1GHz.

So based on a sheet by uzzi38 (https://docs.google.com/spreadsheets/d/1R57Xfp0pES5AFwbT-zAf8FXfE_iIRmkY7iKN0-Azbdc/edit#gid=0) it looks like 2 GHz might be reachable at 25W already for games that don't use a lot of CPU (I think many games from the PS4 era takes ways less CPU power than the 9W that is listed as for moderate CPU load). Open question for 35W/45W is what the top clocks for the GPU are going to be.

As an aside these days 2G is barely enough for anything. I think that will strongly limit the bandwidth benefit from having dedicated VRAM, as lots of assets will be stored in system RAM, and the discrete GPU might actually be hurt from the PCI Express bus being slower than the direct system RAM access that an integrated GPU has. Not sure about the magnitude of the effect though.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,708
3,554
136
I'll wait to see if Rembrandt is faster than even an MX450, let alone an MX550, in actual games. Timespy scores mean very little. 2GB is not as bad as it sounds, though 4GB would obviously be nice to have, since with the MX lineup you're going to be playing at resolutions like 720p or 768p, which isn't as bad as it sounds on a 14-inch display.

I can play something like AC:Unity at 1366x768 resolution with high settings and HBAO+ and get around 45-60 FPS before TDP limits kick in on an MX450.

EDIT:
 
Last edited:

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
I'll wait to see if Rembrandt is faster than even an MX450, let alone an MX550, in actual games. Timespy scores mean very little. 2GB is not as bad as it sounds, though 4GB would obviously be nice to have, since with the MX lineup you're going to be playing at resolutions like 720p or 768p, which isn't as bad as it sounds on a 14-inch display.

I can play something like AC:Unity at 1366x768 resolution with high settings and HBAO+ and get around 45-60 FPS before TDP limits kick in on an MX450.
MX550 won't be that much faster than MX450, at best It will be ~15% faster. But you are right, we need actual gaming tests.

A laptop with MX550 will have at least a FullHD 13.3-15.6 inch display, so playing games at only 720p or 768p is pretty bad in my opinion.
Why would someone bother buying such a laptop for playing games?
They won't be particularly cheap either, based on the current MX450 laptop prices, where the cheapest MX450 laptop is Lenovo IdeaPad 5 Pro for 869 €.
 
Last edited:
  • Like
Reactions: Tlh97

tamz_msc

Diamond Member
Jan 5, 2017
3,708
3,554
136
A laptop with MX550 will have at least a FullHD 13.3-15.6 inch display, so playing games at 720p or 768p is pretty bad in my opinion.
If you're going to be aiming for 60 FPS or higher fidelity (at least medium-high) then 720p is a great resolution for most of the newer games. It doesn't look bad at all on a 14" screen, even if the screen itself is 1080p. MX and iGPU gaming is all about compromises, and this is the compromise many would make to play newer games.
 

ahimsa42

Senior member
Jul 16, 2016
225
194
116
If you're going to be aiming for 60 FPS or higher fidelity (at least medium-high) then 720p is a great resolution for most of the newer games. It doesn't look bad at all on a 14" screen, even if the screen itself is 1080p. MX and iGPU gaming is all about compromises, and this is the compromise many would make to play newer games.
depends entirely on the game-perhaps this may be true for FPS but not for TBS,RTS & RPG for which RMB should work great at 1080p.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,708
3,554
136
nVidia believes otherwise. Hence why there's the mx570 and 2050 laptop.
The reason why the MX550 exists is because the MX450 is more often than not paired with GDDR5 instead of GDDR6. There's like one Lenovo model which comes with GDDR6 and most of the rest are all GDDR5. I believe that NVIDIA wants to push a TU117 based product with GDDR6 because there is a possibility that the MX450 GDDR5 might lose to Rembrandt.

Based on what I think about Rembrandt, given that it should be 2x faster in games than Cezanne, and that Cezanne and Iris Xe 96 EU are roughly equal, and coupled with my own testing on Iris Xe 96 EU and MX450 GDDR5, it should land in between the GDDR5 and GDDR6 MX450.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
There's not a lot of wiggle room for any of these products in relation to price. I can go on Amazon right now and find multiple models listed with 3050 video cards in the $750-$800 price range and a whole bunch with 4GB 1650 gpus in the $675-$750 range. That's just running a search on one site.

The only way that the mx550 and mx570 make sense is below that price range. For the difference in performance, they would need to be in the $600-650 range, where the current competition is the ryzen 5500u-5700u and Intel tiger lake i5 models.


As for the 2050, I struggle to see where it fits in, unless that's the replacement for the 1650 in the market.

(Prices in USD as of today)
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
If memory serves, those are preserved for possible embedded product offerings. I remember reading someone else posting that it didn't take up too much die space, offered a marginal profit opportunity, and they didn't have to pay for the IP included in the MAC unless the chip's package was enabled for it.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,572
146
The lower the power budget the more you need SP count to have decent perf as frequency is too much power demanding comparatively, so lower power parts are the SKUs of choice for maximal SP count.
Until you hit the voltage floor, that is, then it doesn't really matter. Which is relatively high on RDNA2, somewhere around 1200-1500MHz on the dGPU side, probably a tas higher fod Rembrandt given N6.
 

DisEnchantment

Golden Member
Mar 3, 2017
1,590
5,722
136
Rembrandt has integrated MACs: https://www.phoronix.com/scan.php?page=news_item&px=Linux-5.17-AMD-YC-Ethernet
Probably as useless as those 10G MACs that were already available since the original Zen.
They are very useful for the embedded SKUs.

Ryzen V3000 (looks RMB based) should be an enormous upgrade over the V2000 and V1000 series (Used in Tesla Model Y for instance).
In a Linux embedded environment, it should clobber many current high end embedded SoCs like Exynos Auto V9 or Snapdragon SA8xxx series, not to mention anybody can take a Linux distro and support comes out of the box for every peripheral. Zero binary blobs in the HLOS.

RMB is looking like an impressive overhaul to me (LP/DDR5 MC, RDNA 2 iGPU, SFH, Thunderbolt, USB4, new ACP6 Audio DSP, new XGBE, new Z State, new SMU, new Management Controller, updated HSP/TEE). Unfortunately I have not seen the ISP and DSP support in the system logs for RMB so far.
Should be interesting to see how the new Aya NEO (if equipped with RMB) will perform vs the Steam Deck (ignoring the price differences)

If Valve can do a custom VGH SoC with AMD, I hope MS can do one too for their surface devices, Octa Core Zen 4c with RDNA2 on N5 should be possible in around 10W (VGH on N5 would easily be under 7W).
Actually, Valve is also hinting about a Deck successor, during their dev stream when they mentioned a "multi generational product development roadmap"
 

izaic3

Member
Nov 19, 2019
61
96
61
They are very useful for the embedded SKUs.

Ryzen V3000 (looks RMB based) should be an enormous upgrade over the V2000 and V1000 series (Used in Tesla Model Y for instance).
In a Linux embedded environment, it should clobber many current high end embedded SoCs like Exynos Auto V9 or Snapdragon SA8xxx series, not to mention anybody can take a Linux distro and support comes out of the box for every peripheral. Zero binary blobs in the HLOS.

RMB is looking like an impressive overhaul to me (LP/DDR5 MC, RDNA 2 iGPU, SFH, Thunderbolt, USB4, new ACP6 Audio DSP, new XGBE, new Z State, new SMU, new Management Controller, updated HSP/TEE). Unfortunately I have not seen the ISP and DSP support in the system logs for RMB so far.
Should be interesting to see how the new Aya NEO (if equipped with RMB) will perform vs the Steam Deck (ignoring the price differences)

If Valve can do a custom VGH SoC with AMD, I hope MS can do one too for their surface devices, Octa Core Zen 4c with RDNA2 on N5 should be possible in around 10W (VGH on N5 would easily be under 7W).
Actually, Valve is also hinting about a Deck successor, during their dev stream when they mentioned a "multi generational product development roadmap"
Wait, is thunderbolt confirmed for Rembrandt?
 

Thibsie

Senior member
Apr 25, 2017
727
752
136
Wait, is thunderbolt confirmed for Rembrandt?

Well since USB4 is heavily based on TB3, no wonder.
To be confirmed but what I understood is any PC compatible with USB4 is compatible with TB3 AND compatible with USB 3.x and USB 2. USB 1(1.1) though is unkwown (unspecced ?)
That being said lower end devices will be 20Gbps, better oines will be 40Gbps and cables will cost an arm and a leg.