Question AMD Rembrandt/Zen 3+ APU Speculation and Discussion

[izaic3]

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):

• Zen 3 "enhanced"
• DDR5 and LPDDR5 support
• PCIE4
• 6nm
• CVML
• Navi 2/RDNA/RDNA 2
• 12 CU
• Socket FP7/AM5
• Possible USB4
• 2021(unlikely) -2022 (more likely) release date

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.

 

[LightningZ71]

But, that's the big advantage of Rembrandt: it isn't restricted to 2MB of vram. It'll take as muh as needed, provided that the OS hasn't committed it all. On a 16GB laptop, it'll easily have 4+GB available.

As for the expense of making a chip specificaly for MX550, it's likely doing what some of the previous MX chips did and use previous gen chips that are EOL. Case in point, these are probably just the same chips from the 1650/ti, just with only half the VRAM channels not activated, likely come inactive internal bits, but with higher clocks.

 

[jpiniero]

But, that's the big advantage of Rembrandt: it isn't restricted to 2MB of vram. It'll take as muh as needed, provided that the OS hasn't committed it all. On a 16GB laptop, it'll easily have 4+GB available.

As for the expense of making a chip specificaly for MX550, it's likely doing what some of the previous MX chips did and use previous gen chips that are EOL. Case in point, these are probably just the same chips from the 1650/ti, just with only half the VRAM channels not activated, likely come inactive internal bits, but with higher clocks.

Videocardz thinks it's Ampere. Like it's a 3050L with it's bus cut to 64-bit.

 

[LightningZ71]

I wouldn't doubt it. I wonder if it will be able to do dlss? That could be a game changer there...

 

[jpiniero]

I wouldn't doubt it. I wonder if it will be able to do dlss? That could be a game changer there...

Oh yeah. I think the question is more will they disable Ray Tracing. Which I still haven't gotten an answer as to whether Rembrandt has RT cores.

Granted 2 GB might be a tad little for RT, but hey.

 

[Shivansps]

But, that's the big advantage of Rembrandt: it isn't restricted to 2MB of vram. It'll take as muh as needed, provided that the OS hasn't committed it all. On a 16GB laptop, it'll easily have 4+GB available.

As for the expense of making a chip specificaly for MX550, it's likely doing what some of the previous MX chips did and use previous gen chips that are EOL. Case in point, these are probably just the same chips from the 1650/ti, just with only half the VRAM channels not activated, likely come inactive internal bits, but with higher clocks.

Is AMD allowing more than 2GB dynamic VRAM in the drivers yet? Because last time i checked that was the limit, you can set 50% of the ram as fixed vram in the bios but some games seems to have problems to use both dynamic and fixed and the same time. they end up using one or the other, using both is not as common as you may think.

 

[Shivansps]

Oh yeah. I think the question is more will they disable Ray Tracing. Which I still haven't gotten an answer as to whether Rembrandt has RT cores.

Granted 2 GB might be a tad little for RT, but hey.

Here the thing, the RT cores are built into RDNA2 arch, if the APU werent going to have RT core it wouldnt be a RDNA 2 APU per se.

 

[uzzi38]

DG1 has 4GB of memory, means there is a good chance DG2-128 mobile will get 4GB as well, desktop DG2-128 seems to get 6GB. 2GB VRAM is a joke nowadays.

DG1 used LPDDR4X, which is pretty light on power consumption. That's the real issue when talking 30W TDPs at the end of the day - last thing you really want is 1/4 your power budget being allocated to memory+IMC power. Oh, and also SoC power too, you can run the internal fabric at extremely low clocks on top with such a small amount of memory bandwidth.

It's possible, but I have my doubts if I'm honest. Depends on if they could fit a memory subsystem that can support twice the memory bandwidth od DG1 whilst retaining throughput on the GPU core well enough at that same 30W TDP I guess.

 

[uzzi38]

Videocardz thinks it's Ampere. Like it's a 3050L with it's bus cut to 64-bit.

Yeah it's 100% GA107.

 

[LightningZ71]

Is AMD allowing more than 2GB dynamic VRAM in the drivers yet? Because last time i checked that was the limit, you can set 50% of the ram as fixed vram in the bios but some games seems to have problems to use both dynamic and fixed and the same time. they end up using one or the other, using both is not as common as you may think.

In actuality, this depends entirely on the motherboard manufacturer and their Bios/Uefi team. While the automatic setting is supposed to handle allocations dynamically, and can apportion seemingly up to 1/4 of system ram if left alone on a properly configured system. If AUTO fails to work right, and you have a good bios, you can usually find a setting called "Uma buffer size" which can statically assign a certain amount to VRAM (also seems to be limited to 1/4 total ram). It's usually under Advanced or NB configuration. There are a few videos online of people with 64GB ram setting 16GB and having it available in programs like Photoshop.

I've personally seen 4GB myself on a 4700u.

 

[Insert_Nickname]

While the automatic setting is supposed to handle allocations dynamically, and can apportion seemingly up to 1/4 of system ram if left alone on a properly configured system.

Thing is that statically allocated memory can only be used for graphics. If you manually select only 1 or 2GB you have more memory available for the OS. My Athlon is quite happy with an allocation of just 96MB. Still has 4GB dynamic available.

For gaming you of course want more memory, so YMMV. There isn't a perfect cover-all setting.

 

[Shivansps]

In actuality, this depends entirely on the motherboard manufacturer and their Bios/Uefi team. While the automatic setting is supposed to handle allocations dynamically, and can apportion seemingly up to 1/4 of system ram if left alone on a properly configured system. If AUTO fails to work right, and you have a good bios, you can usually find a setting called "Uma buffer size" which can statically assign a certain amount to VRAM (also seems to be limited to 1/4 total ram). It's usually under Advanced or NB configuration. There are a few videos online of people with 64GB ram setting 16GB and having it available in programs like Photoshop.

I've personally seen 4GB myself on a 4700u.

The "auto" setting is the one that depends on the motherboard vendor and the only thing it does is to set the fixed VRAM.
For what i saw on consumer boards it is up to 2GB with 4G decoding off and up to 50% of ram with 4G decoding on.

But that changes nothing of the dynamic vram, that one is set by the driver. Ill need to check again because maybe they changed it in the last few months, but the dynamic vram was limited to just 2GB last time i checked. If i have time ill check again today.

Normally what do you want is to set a very low amount of fixed VRAM... 64-128MB is just fine. and let the games run on the dynamic vram. But 2GB is just not enoght these days, even for a IGP. Remember that W10-11 uses a non-trivial amount of vram too.

Also remember that fixed and dynamic are two diferent memory partitions, for example, an ETH miner cant start on 4GB Fixed + 2GB Dynamic even trought the miner can see the 6GB avalible, it cant allocate more than 4GB of vram in one go.
That something i noticed some games tend to want to run only on fixed or only on dynamic.

What is very funny about all this is that both implementation of Navi 24 ( RX 6500XT and RX 6400 ) seem to come with just 4GB VRAM... the RX 6400 specially is equal to the RMB IGP except for the dedicated vram and the 16mb ic.
We may come to some cases were the RMB igp is just faster because it can have more vram.

 

[Tuna-Fish]

I guess, but it's just not economical to produce such low end dies at this point. At the very least not on leading edge silicon, maybe we could see another gen on N6 or older nodes still. I just have my doubts.

There is market for very low-power GPUs for applications like multi-monitor setups, for workstations that need a lot of dislays or for shopfronts and the like. There will always be a very low-end GPU on the market from both vendors that target that space.

Of course, those GPUs do not need to be refreshed very often. They might last many years in the lineup.

 

[Shivansps]

Update:
They finally fixed dynamic VRAM... there is no need for special settings on bios, i would recomend to set dedicated vram to the lowest as possible on a 16GB system.

A did some quick tests, win11 task manager seems to think there is up to 8GB vram avalible, PhoenixMiner can see exactly 6GB what is enoght to start up the ETH miner (lets remember this was no possible just a few months ago when the 5700G launched whiout going to bios and set 8GB dedicated vram).

And yeah, Windows is right, Shadow of the Tomb Raider with ultra textures can pass the 7GB VRAM mark.

Im not sure if this is limited to Windows 11 or it also applies to Windows 10, but the fact is they finally got dynamic vram right and they really needed to get this done ahead of RMB launch.

So it is official now that a Navi 24 GPU has less memory avalible than a RMB APU. This will be interesting.

 

[DisEnchantment]

Rembrandt should be very interesting besides the core itself.

• New HSP
• New Q5 and C6 DSP Cores
• New Memory Controller
• New GFX
• New DCN
• New VCN
• New Power Management Framework
• New ISP
• Most likely New AOP taken from VGH --> this is new as suggested by @KOMACHI_ENSAKA is an A55 present in VGH. Very likely related to the new power management framework mentioned by Hallock
• Process Update

Apparently this is not all
Rembrandt also has a new Xtensa LX6 Chip doing remote management duties in a framework called AIM-T (AMD Integrated Management technology), similar to Intel remote management engine. I guess this will go to all the Pro SKUs. I guess they just license in bulk a bunch of Cadence Tensilica IPs
New Z state deep sleep likely handled by the AOP (Valve also mentioned a feature like this on Steam Deck, instant wake up from days of suspend state)
Smartshift 2.0

Another thing to note also, lots of Rembrandt ACP firmware and support added to SOF framework.
Nice work on this, at least OEMs can work with Tensilica SDK directly to customize the Audio DSP FW with out of box support in SOF, QCM Hexagon SDK may still be superior but SOF is much more desired due to more industry support and being open source. (MTK is also using SOF along with Intel)

3 weeks to go, but zero leaks so far. AMD is tightening the screws looks like.

 

[eek2121]

3 weeks to go, but zero leaks so far. AMD is tightening the screws looks like.

The lack of leaks blows my mind. Then again, Intel has a pretty tight lid on their GPU as well. I understand why companies wouldn't want leaks to happen, but they make things boring. 😂

 

[deasd]

some guys said what would be going to released at CES is AM5 APU... based on 6nm Rembrandt, and V-cache(Zen3S)

https://twitter.com/x/status/1470299382784606209

 

[izaic3]

Another benchmark. Appears to be another ES, so grain of salt of course.

 

[ahimsa42]

along with seeing the actual benchmarks at CES and hopefully finding out the price range, i wonder if AMD will reveal if RMB will be mobile only or if there will also be a desktop G version? it seems that many people assume that there will be one by default but it's possible with zen4 coming so soon they may skip this version ala renoir.

 

[jpiniero]

there will also be a desktop G version?

Yes but it might be OEM only. Also it's coming later.

 

[eek2121]

some guys said what would be going to released at CES is AM5 APU... based on 6nm Rembrandt, and V-cache(Zen3S)

https://twitter.com/x/status/1470299382784606209

Unless AMD flat out lied regarding AM5, I doubt it. More likely, Rembrandt desktop is coming earlier than we think along with Zen 4. There is a 4GHz base/4.8 ghz boost 8c16t "G" ES chip using desktop DDR5 floating around, so clearly AMD is closer to final silicon than most realize.

 

[leoneazzurro]

Final packaging (of AM5 Rembrandt) because final silicon for RMB is the same for the mobile and desktop version, and mobile is clearly targetted for a CES launch...

 

[uzzi38]

some guys said what would be going to released at CES is AM5 APU... based on 6nm Rembrandt, and V-cache(Zen3S)

https://twitter.com/x/status/1470299382784606209

I do not believe the desktop version is any time soon, because AM5 isn't.

 

[ahimsa42]

only 4% incease-but of course this is very likey an enginering sample which will be further optimized so it may translate closer to 10-15% over a 5800U. not bad considering the greatly increased RDNA2 igpu & the fact that it is the same platform as the 5000 series cpu's. hopefully RMB will also include power managment improvements for increased battery life over the 5800U too.

AMD Ryzen 6000U 'Rembrandt' APU Shows Up With DDR5 Memory Support In ASUS's Next-Gen Vivobook, Faster Than Ryzen 7 5800U

ASUS's next-gen laptops with AMD Ryzen 6000 Rembrandt APUs have once again leaked out and this time over at the BAPCo benchmark repository.

wccftech.com

 

[leoneazzurro]

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.

 

[deasd]

I do not believe the desktop version is any time soon, because AM5 isn't.

yeah I think so... from the words on the image it seems there would be a paper launch/confirmation of AM5 related product at CES, 'long way from sale' could be this.
But 'AM5 APU sample earlier than Zen3' is very supposed to be AM5 Rembrandt sample though, what's unsure is we don't know the production progress of AM5 mobo.

Perhaps... it's Lisa Su's 'Plan B' which to accelerate AM5 platform production when presume low-mid end of AlderLake is a threat?