AMD Portal Entry Confirms Historical Testing of Ryzen "Rembrandt" Zen 3+/RDNA2 APUs on AM5 Platform

[T0@st]

Over the past week or two, next-gen processor fanatics have been combing through AMD's Technical Information Portal—many truth seekers are trying to determine details of new product lineups; likely in anticipation of official unveilings (at Computex 2025). Super sleuths are still trying to find Team Red's mythical Ryzen Threadripper PRO "9995WX" processor; an alleged 96-core "Shimada Peak" range-topping beastie. Going against forward progress grains, VideoCardz has uncovered much older technologies—harking back to the days of Zen 3+ and RDNA 2. Revisiting late summer 2023, Team Red-related leaks suggested the potential arrival of (6 nm) Ryzen "Rembrandt" 6000G series desktop APUs.

We now know that AMD skipped a couple of AM5-bound steps, and eventually released its Ryzen "Phoenix" 8000G (Zen 4 and RDNA 3) family early last year. The discovery of an "AMD AM5 Functional Test Vehicle APUs (Family 19h Models 40h-4Fh) are based on the Zen 3+ 6 nm core with Radeon Navi Graphics" entry within Team Red's technical info database, has reopened online discussions about forgotten/abandoned intellectual properties. VideoCardz believes that the "19h" tag denotes "Rembrandt" family origins—additionally, their curiosity was piqued by the possibility of 6000G prototypes units escaping out into the wild. Looking ahead, the AM5 platform is expected to welcome (next-gen) Ryzen 9000G "Gorgon Point" desktop (Zen 5 + RDNA 3.5) processors.



View at TechPowerUp Main Site | Source

 

[ymdhis]

That would somewhat explain why there was such a huge gap between the 5600G and 8600G. Hopefully it's something that won't be repeated, these chips are extremely useful, they have lower idle power due to the monolithic die and the iGPU is absolutely brutal. I'm hoping I can upgrade to a 9600G this year, with possibly another upgrade while still on AM5. The iGPU is over twice as powerful than my 5600G. Far better option than getting a videocard nowadays, especially since I need to upgrade for more memory and m.2 slots anyway.

 

[AusWolf]

If this comes out cheap, I might finally bring myself to upgrade my aging (but still rocking) 4th gen i7 HTPC.

 

[kondamin]

What a waste

 

[Chrispy_]

It was probably canned due to AM5 launching at a time of very high DDR5 costs and motherboards being all over $200 for the first year or so.

APU's aren't cheap themselves, but presumably the bulk of their sales are to system integrators building machines where the IGP can offset GPU costs. I can definitely see how high AM5 platform costs negated the advantage for the majoity of their existing customers, which explains the cancellation (or simply development halt) of a desktop AM5 variant.

 

[Spartoz]

AMD doing this again... I really don't get why they launch "new" CPUs with an older architecture. I mean, they have a modern, better architecture available, the R&D is already done and everything, so why??
Imagine if intel, just launched Ice lake CPUs again on a modern socket, that would make no sense... Still, AMD is doing this every year. It's like the "Mendocino" APUs, why not use Zen 4 instead of Zen 2 ? They even used RDNA2 which was never seen before on Zen 2, so they had to completely design a new chip for those APUs... Makes no sense to me

 

[SL2]

AMD doing this again...

What do you mean? Does anything in the article imply that they'd actually launch it, now? What do you think "historical testing" means? Or, "forgotten/abandoned intellectual properties"

It's like the "Mendocino" APUs, why not use Zen 4 instead of Zen 2 ?

There were no Zen4 APU's available in 2022.

Cheaper nodes for budget CPU's?

They even used RDNA2 which was never seen before on Zen 2, so they had to completely design a new chip for those APUs... Makes no sense to me

You forgot about van Gogh, how convenient. Yes, different node.

Steam Deck - Wikipedia

en.wikipedia.org

Also, we have no idea of WHEN they designed it, or for what purpose originally. It's entirely possile that it was done at an earlier stage than the launch suggests.

 

[Wirko]

It was probably canned due to AM5 launching at a time of very high DDR5 costs and motherboards being all over $200 for the first year or so.

APU's aren't cheap themselves, but presumably the bulk of their sales are to system integrators building machines where the IGP can offset GPU costs. I can definitely see how high AM5 platform costs negated the advantage for the majoity of their existing customers, which explains the cancellation (or simply development halt) of a desktop AM5 variant.

That's if desktop Zen3+ on AM5 ever was part of the plan - and it's totally possible that it never was. If AMD just needed a test "vehicle" for internal use, they could design a new substrate, glue the chip on it, tweak the BIOS a bit and here it was, a cheap, quick and dirty solution. It would have worked, perhaps with many bugs and shortcomings but who cared.

Edit: I probably can't write complex conditional sentences in past tense correctly but whatever, it's conditional and it was in the past.

 

[Chrispy_]

That's if desktop Zen3+ on AM5 ever was part of the plan - and it's totally possible that it never was. If AMD just needed a test "vehicle" for internal use, they could design a new substrate, glue the chip on it, tweak the BIOS a bit and here it was, a cheap, quick and dirty solution. It would have worked, perhaps with many bugs and shortcomings but who cared.

Edit: I probably can't write complex conditional sentences in past tense correctly but whatever, it's conditional and it was in the past.

I know what you mean.
Internal prototyping is easy for their MCD designs since the switch from a DDR4 to a DDR5 memory controller is just a change of IO die.
That's why I assume it was a platform cost issue, and not one of technical difficulty.

 

[ymdhis]

AMD doing this again... I really don't get why they launch "new" CPUs with an older architecture. I mean, they have a modern, better architecture available, the R&D is already done and everything, so why??
Imagine if intel, just launched Ice lake CPUs again on a modern socket, that would make no sense... Still, AMD is doing this every year. It's like the "Mendocino" APUs, why not use Zen 4 instead of Zen 2 ? They even used RDNA2 which was never seen before on Zen 2, so they had to completely design a new chip for those APUs... Makes no sense to me

They didn't have Zen4 APU at the time, they only just finished doing the chiplet versions. The monolithic chips always follow later, which makes sense, since the chiplets are easier to respin if something goes wrong, plus they need extra work to reintegrate the uncore into the monolithic die. So to have any APU available whatsoever for the platform, they would've had to port an older Zen 3+ core to AM5 so it could be launched side by side with the Zen 4 chiplet based cpu.

I have no clue about Mendocino though. I can only guess that it was either a test run, or a partner asked them for it, or they didn't have anything in that power/perf slot so they had to release a stopgap (I didn't check whichever one of these is true).

 

[Apocalypsee]

If this comes out cheap, I might finally bring myself to upgrade my aging (but still rocking) 4th gen i7 HTPC.

I'll buy this for my main PC if it's good. Only thing is I would miss CUDA support tho.

 

[AusWolf]

I'll buy this for my main PC if it's good. Only thing is I would miss CUDA support tho.

You still have CUDA if you pair it with an Nvidia GPU. The only thing is the PCI-e lanes. That's where AMD tends to cheap out with their APUs.

On the other hand, you'd have ROCm.

 

[Apocalypsee]

You still have CUDA if you pair it with an Nvidia GPU. The only thing is the PCI-e lanes. That's where AMD tends to cheap out with their APUs.

On the other hand, you'd have ROCm.

Only earlier APU have their PCIe lanes cut off, my 5700G got 20 lanes so full x16 for graphics but only gen 3.

Hoping ROCm supports this new APU, I might migrate to Linux too

 

[AusWolf]

Only earlier APU have their PCIe lanes cut off, my 5700G got 20 lanes so full x16 for graphics but only gen 3.

They have them cut off to the GPU. The 8500G has 4 lanes for it, the 8700G has 8.

Hoping ROCm supports this new APU, I might migrate to Linux too

I made the leap in October. Haven't looked back since.

 

[Dr. Dro]

If this comes out cheap, I might finally bring myself to upgrade my aging (but still rocking) 4th gen i7 HTPC.

Looks like this was only a test processor and will never be released, nor does it have a reason to be released in 2025... Zen 3 is now obsolete as is Navi 1.

 

[AusWolf]

Looks like this was only a test processor and will never be released, nor does it have a reason to be released in 2025... Zen 3 is now obsolete as is Navi 1.

True. Besides, the 8500G is actually dirt cheap (although its iGPU isn't the best).

I've just looked, the complete lack of m-ITX A620 boards is a much bigger concern.

 

[Dr. Dro]

Problem with Phoenix isn't even the iGPU, it's the extremely limited amount of PCIe lanes available, makes the processor embarrassingly useless. I spoke strongly about it when W1zz reviewed it, people were upset with me, but it's the truth. I could understand 4 lanes on something like the 4700S desktop kit (reused PS5 SoC with the graphics disabled), since it's just a salvaged chip, but... not on a socket AM5 chip.

 

[Aerpoweron]

i really would like to see some Zen 4 or 5 stuff on AM4. APUs will be difficult since they need a completely new die. But some Zen 4 or 5 CPU chiplets with an Zen 2 / 3 IO-die would be possible.

Would be fun to see an X3d chip with Zen5 on AM4
Gaming performance of the latest stuff on an 8 year old board.

 

[AusWolf]

Problem with Phoenix isn't even the iGPU, it's the extremely limited amount of PCIe lanes available, makes the processor embarrassingly useless. I spoke strongly about it when W1zz reviewed it, people were upset with me, but it's the truth. I could understand 4 lanes on something like the 4700S desktop kit (reused PS5 SoC with the graphics disabled), since it's just a salvaged chip, but... not on a socket AM5 chip.

I can sort of understand 8 lanes on the 8700G which is a decent all-around chip, but not 4 lanes on the 8500G which has a crap iGPU, so you'd probably end up using a dGPU with it anyway, which oops, isn't such a good idea because, well, 4 lanes only.

Edit: What I really don't understand about the 8500G is why not give it Zen 4c cores only (even if just 4), and use the extra die space for an actually useful iGPU? That way, having only 4 lanes for a dGPU wouldn't be a problem, because you wouldn't need one to begin with.

i really would like to see some Zen 4 or 5 stuff on AM4. APUs will be difficult since they need a completely new die. But some Zen 4 or 5 CPU chiplets with an Zen 2 / 3 IO-die would be possible.

Would be fun to see an X3d chip with Zen5 on AM4
Gaming performance of the latest stuff on an 8 year old board.

Nobody would switch to AM5 then, so not gonna happen.

 

[Dr. Dro]

Market segmentation, I guess.

 

[AusWolf]

Market segmentation, I guess.

Yeah, but it makes the 8500G pointless, imo. If I wanted 6 cores with a useless iGPU, I'd just get a 7600 instead.

 

[Aerpoweron]

Edit: What I really don't understand about the 8500G is why not give it Zen 4c cores only (even if just 4), and use the extra die space for an actually useful iGPU? That way, having only 4 lanes for a dGPU wouldn't be a problem, because you wouldn't need one to begin with.

The 8500G is just a mobile chip put on AM5. The sad thing is, there is no other way to get the Zen4c cores on AM5.
For the targeted mobile market, the 8500G chip is a good balance.
I would estimate the Radeon 740 iGPU is around the 5600G iGPU performance. I have to look at my 3d mark benchmarks.

 

[Wirko]

Looks like this was only a test processor and will never be released, nor does it have a reason to be released in 2025... Zen 3 is now obsolete as is Navi 1.

Of course. This news is actually about an archaeologist digging a bone fragment from deep under the ground, then analysing the DNA (muh) and trying to infer what the animal looked like.

I know what you mean.
Internal prototyping is easy for their MCD designs since the switch from a DDR4 to a DDR5 memory controller is just a change of IO die.
That's why I assume it was a platform cost issue, and not one of technical difficulty.

Maybe silicon availability issue as well. AMD can't get wafers in limitless quantities, so they have to allocate them wisely between different products. What other products used 6 nm in 2023?

 

[AusWolf]

The 8500G is just a mobile chip put on AM5. The sad thing is, there is no other way to get the Zen4c cores on AM5.
For the targeted mobile market, the 8500G chip is a good balance.
I would estimate the Radeon 740 iGPU is around the 5600G iGPU performance. I have to look at my 3d mark benchmarks.

Personally, I don't care if it's Zen 4 or Zen 4c. It's pretty much the same architecture anyway. What I'd want from a chip like this is a decent iGPU, or enough PCI-e lanes (preferably the former), but it has neither.

 

[blinnbanir]

True. Besides, the 8500G is actually dirt cheap (although its iGPU isn't the best).

I've just looked, the complete lack of m-ITX A620 boards is a much bigger concern.

A620 ITX | Newegg.ca

Search Newegg.ca for A620 ITX. Get fast shipping and top-rated customer service.

www.newegg.ca