Question AMD Rembrandt/Zen 3+ APU Speculation and Discussion

Page 20 - 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
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:

Spicy

Member
Oct 5, 2021
46
48
51
It makes little business sense to make a giant APU, as much as everyone drools over the idea. AMD would rather sell a GPU and a CPU instead of just a CPU. I am not saying I agree with it.

AMD could consider selling motherboard OEMs radeon mobile chips that would be integrated directly onto the motherboard.

I wish they would get more creative and do something like that.
The good old days of nForce 2 motherboards, with fan above the chipsets :cool:

Seeing that the console makers have committed financially to GPU development, is there a possibility of a secret agreement preventing AMD from getting too close in performance with a retail client APU?
This was also my hypothesis, for this lack of RDNA1/2 before Rembrandt.
 
Last edited:

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
While I don't trust UserBenchmark any farther than I can throw the editors, if that benchmark bears out in actual games on modern APIs, it will neatly match the comparisone between the Vega56 and the 6600XT, even without the infinity cache being invalve. That would be impressive. I don't think it will quite be a 40% improvement across the board on games, it could be situationally impressive.
 

Joe NYC

Golden Member
Jun 26, 2021
1,935
2,272
106
do four channels of DRAM in four SODIMM slots, two on either side of the processor, or, have 16GB soldered on two channels, and have two SODIMM slots for the other two.

That's exactly what I was thinking. It would still give OEM the capability to create a garbage system by opting for no internal DRAM and populating only 1 channel of external.

But less moronic OEM would pick the one with fast internal memory and still give users capability to upgrade to more memory (and with it more bandwidth too).

Or, another OEM could just decide to go with an entry level system with just internal memory, no expandability and lowest cost still with good performance

2 internal channels with 6400 LPDDR5 would go quite a long way. Definitely farther than a garbage system floating around today in the rumormill, Rembrandt engineering sample with single channel of DDR5-4800:

AMD Ryzen 'Rembrandt' APU With 8 Cores, 16 Threads & RDNA 2 GPU Spotted, Running On Corsair Xenomorph AIO PC With DDR5-4800 Memory (wccftech.com)

That's MUCH cheaper than integrating a dGPU that will likely cost over $100, require routing for 128 bits of GRRD6, handle thermals for those packages and the dGPU, handle power for that dGPU, and deal with the UEFI oddities that it requires , and, instead, keep the revenue in AMD's pocket instead.

We aren't seeing many design wins for AMD mobile GPUs in the market. We've got, what, 6 different models out of the 6600M right now, base around 4 distinct chassis designs? Nothing higher in the x86 market, and a few 5500M designs knocking around in the low end? If AMD decided to invest in a SOC that could obviate the need for low end dGPUs, they could get FAR more market penetration and revenue from that market, revenue that they are currently not getting! Sure, they are getting money from low end gaming laptops with 4600/5600H processors, but they are giving away the GPU side to 1650s and 3050s. With a SOC that costs maybe $20 more per unit to make and package, they can achieve an additional $100 in realized revenue per unit in laptops that are price and performance competitive with 1650/3050 discrete designs that allow both the laptop manufacturer and themselves to realize a greater share of the revenue.

Agreed. It may be possible that AMD had a hope that it may be able to get on the gravy train of the dGPUs. But all the revenue is just going to NVidia, very little to AMD.

Rembrandt is a step in the right direction, and what would really kick ass if AMD sold one SKU with 16 GB of fast LPDDR5 in MCM.

If it can be done in the Steam Deck, probably with identical Memory Controller, maybe it could be done with Rembrandt.
 

Joe NYC

Golden Member
Jun 26, 2021
1,935
2,272
106
This was also my hypothesis, for this lack of RDNA1/2 before Rembrandt.

Imagine that, say, RDNA3 has a pure compute (GCD), no external IO, that is modular and reusable. Then, it would not have to take 12-18 months to integrate it into the APU, to spin different APU configurations...

That's probably where AMD would like to be in the future...
 

soresu

Platinum Member
Dec 19, 2014
2,657
1,858
136
Imagine that, say, RDNA3 has a pure compute (GCD), no external IO, that is modular and reusable. Then, it would not have to take 12-18 months to integrate it into the APU, to spin different APU configurations...
It's certainly a nice outcome in theory, but without some sort of common interface to bind it to with the TSVs such a possibility would still result in significant delays.

It's also possible that they can just create a dumb block of passive bridging silicon+metal layer that can be more easily designed for the purpose even if not in a standarised way.

In fact I'm pretty sure that something of the sort was found in their patents for 2.5D/3D chiplets a while back, something that sits in a recessed area of the interposer so that it adds nothing extra to the overall package height when the IOD and GCD are added to it.
 

scannall

Golden Member
Jan 1, 2012
1,946
1,638
136
OEM's can be a tough nut to crack. I've thought for a while that AMD should put together a reference laptop motherboard, and guarantee dimensions, cables and connections for say 5 years with the underlying tech improving each year as they release new products. OEM's would only have to toss their flavor of bios, start screens etc on and slap it into a chassis they know they won't have to retool for every year.

Taking a page from Apple, a reference board could have LPDDR5x ram on package or on the motherboard right up against the CPU for example and really let their products shine.
 

maddie

Diamond Member
Jul 18, 2010
4,738
4,667
136
OEM's can be a tough nut to crack. I've thought for a while that AMD should put together a reference laptop motherboard, and guarantee dimensions, cables and connections for say 5 years with the underlying tech improving each year as they release new products. OEM's would only have to toss their flavor of bios, start screens etc on and slap it into a chassis they know they won't have to retool for every year.

Taking a page from Apple, a reference board could have LPDDR5x ram on package or on the motherboard right up against the CPU for example and really let their products shine.
Well they have the money now to start thinking bigger dreams. One can hope.
 

jpiniero

Lifer
Oct 1, 2010
14,585
5,209
136

Ooh interesting, Barcello is still keeping the 5xxxU branding. There's a 5825U and a 5675U. OEMs aren't going to like it that they aren't mixing the gen branding with Rembrandt unless it's coming a lot sooner than Rembrandt is.
 

leoneazzurro

Senior member
Jul 26, 2016
924
1,451
136

Ooh interesting, Barcello is still keeping the 5xxxU branding. There's a 5825U and a 5675U. OEMs aren't going to like it that they aren't mixing the gen branding with Rembrandt unless it's coming a lot sooner than Rembrandt is.

Branding Lucienne as 5xxx series was seen as a fiasco by most users, and that was mainly due to OEMs, Worth nothing that MT performance of a 5700U was indeed between a 5800U and a 5600U. and feature set was almost on par. Btw, in this case the differences between Rembrandt and Barcelo are quite significant on the feature set, because Rembrandt will have RDNA2 graphics, CVML block, and DDR5 being the most distinctive differences. So it would have been quite difficult for AMD to market these as 6XXX series even if OEM wanted to.
 

ahimsa42

Senior member
Jul 16, 2016
225
194
116

Ooh interesting, Barcello is still keeping the 5xxxU branding. There's a 5825U and a 5675U. OEMs aren't going to like it that they aren't mixing the gen branding with Rembrandt unless it's coming a lot sooner than Rembrandt is.
not sure who would be interested in these anyway with Rembrandt likely so close to being released. i currently have a 4500U laptop & have to decide if i should upgrade to Rembrandt or wait for zen4 Phoenix in late 2022/early 2023.
 

jpiniero

Lifer
Oct 1, 2010
14,585
5,209
136
not sure who would be interested in these anyway with Rembrandt likely so close to being released. i currently have a 4500U laptop & have to decide if i should upgrade to Rembrandt or wait for zen4 Phoenix in late 2022/early 2023.

DDR5 pricing might make Barcello appealing to OEMs. Are they making any core changes? Outside of IGP gaming the performance difference might only be clocks.
 

leoneazzurro

Senior member
Jul 26, 2016
924
1,451
136
not sure who would be interested in these anyway with Rembrandt likely so close to being released. i currently have a 4500U laptop & have to decide if i should upgrade to Rembrandt or wait for zen4 Phoenix in late 2022/early 2023.

Those who will be interested in this are i.e. people having old equipment and looking for cheap light laptops with good CPU and good enough GPU, i.e. most of corporate users (non-executives).
 

zir_blazer

Golden Member
Jun 6, 2013
1,164
406
136
AMD could consider selling motherboard OEMs radeon mobile chips that would be integrated directly onto the motherboard.

I wish they would get more creative and do something like that.
They already did. I'm positively sure to have seen an industrial Motherboard that had a Embedded Radeon Exxx soldered in, but I'm not hitting google matches.


OEM's can be a tough nut to crack. I've thought for a while that AMD should put together a reference laptop motherboard, and guarantee dimensions, cables and connections for say 5 years with the underlying tech improving each year as they release new products. OEM's would only have to toss their flavor of bios, start screens etc on and slap it into a chassis they know they won't have to retool for every year.

Taking a page from Apple, a reference board could have LPDDR5x ram on package or on the motherboard right up against the CPU for example and really let their products shine.
This was talked about nonstop since the days of the first APUs when Notebook vendors loved to wreck them with Single Channel RAM, and the conclusion was that AMD should do something like Intel Centrino (A set of guidelines that vendors have to comply to get the right to use a quality logo, like Microsoft also does for some of their Windows compatible logos), but that AMD didn't had the muscle to force vendors to adopt that.
Also, first tier vendors doesn't seem to like to use reference designs. I think that a AMD reference design may be put to use in second tier vendors like System76 that uses Clevo build Notebooks. I'm personally tired of seeing powerful APUs next to a two generations old low end GPU, as those are precisely the thing of stuff that the APU was intended to replace, but I hear some people claiming that Notebook vendors do that cause their systems are more marketeable that way (Having a GeForce means gaming, right?).
 

moinmoin

Diamond Member
Jun 1, 2017
4,944
7,656
136
Imagine that, say, RDNA3 has a pure compute (GCD), no external IO, that is modular and reusable. Then, it would not have to take 12-18 months to integrate it into the APU, to spin different APU configurations...

That's probably where AMD would like to be in the future...
What you are talking about already exists as something modular usually referred to as IPs. ATi did that with graphics before AMD bought it, then the approach was applied more and more to CPUs as well. The Construction and Cat cores were the first CPU families to use it, though imo AMD only made excellent use of the flexibility it affords starting with Zen by the way of Infinity Fabric.

Unfortunately those tables detailing all the IPs with their version numbers are not often being shared. An example would be Renoir's iGPU consisting of GCN 5.1, DCN 2.1, VCN 2.0 etc.

What still takes time is adapting all IPs to a given node and organizing them in an efficient layout, then going through all the time consuming and costly silicon mask design and verification steps.
 

ahimsa42

Senior member
Jul 16, 2016
225
194
116

ahimsa42

Senior member
Jul 16, 2016
225
194
116
from the things i have heard recently, rembrandt may be the sweet spot for mobile in 2022 as far as price per performance. it would also allow one to wait a few years and skip zen 4 phoenix to upgrade in 2025 with zen 5.
 

Joe NYC

Golden Member
Jun 26, 2021
1,935
2,272
106
from the things i have heard recently, rembrandt may be the sweet spot for mobile in 2022 as far as price per performance. it would also allow one to wait a few years and skip zen 4 phoenix to upgrade in 2025 with zen 5.

When people hear "upgrade" they likely understand swap CPU. But Rembrandt mobos may not support PCIe5
 

leoneazzurro

Senior member
Jul 26, 2016
924
1,451
136
Well, PCIe 5 will be practically useless on desktop for quite a while. There are basically no peripheals out there which are compatible (cost?), and IIRC the only way to connect something else other than a GPU needs a riser from a X8 slot.. quite practical indeed. And even the GPUs so far show small differences between 4.0 and 3.0 (Doom Eternal apart), so I wonder if we'll see differences at all in a reasonable timeframe between 4.0 and 5.0. That said, it is unlikely someone buying a Rembrandt will pair it with a discrete GPU, if not as afterthought, as less cache on die will likely impact gaming performance against a probably cheaper Zen3.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
PCIe 5.0 will be a bigger impact on DirectX direct storage loading where it directly pulls textures from the SSD into video card memory and in any situations where there is texture spill from VRAM. As we saw in the 5500/5500XT, the 4GB cards took a big hit ove the 8GB cards when they were used in a slower PCIE slot configuration. The same will likely be true at higher levels with higher qualit textures, etc. It will be enhanced by DDR5's greater bandwidth.
 
  • Like
Reactions: Rigg