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]

Given the structure of the CCX, if there was a cache like that for the iGPU, it would look like a cross between the SLCs of ARM chips or the M1, and Crystal Well of the Broadwell days. Having the iGPU reaching into the L3 for the CCX would disrupt how the CCX functions too much. However, having a large L3 for the processors DOES take pressure off of the memory subsystem, allowing the iGPU to have more access to it in more cases.

 

[Joe NYC]

Given the structure of the CCX, if there was a cache like that for the iGPU, it would look like a cross between the SLCs of ARM chips or the M1, and Crystal Well of the Broadwell days. Having the iGPU reaching into the L3 for the CCX would disrupt how the CCX functions too much. However, having a large L3 for the processors DOES take pressure off of the memory subsystem, allowing the iGPU to have more access to it in more cases.

Yeah, that's what I thought. Even tough it is a single piece of silicon, iGPU does not have access into the CPU L3.

It probably does not make a lot of sense to invest resources into making the L3 sharable on the same die, because future APUs may be chiplet based, ot monolithic.

 

[Spicy]

A bit confused by your post. Renior and Cezanne desktop APUs launched well after mobile variants. Cezanne desktop just launched. We aren’t getting an APU release prior to Q3 of next year, and tgat part won’t be Rembrandt. Laptop Rembrandt launches very soon.

Renoir desktop APUs (4000G/GE) were never released for consumers. Only for OEM (Dell, Asus, HP, ...).
Old rumors announce first AM5 motherboards for Q2 2022. But they will be for OEM only, probably.
If the launch of 6000U/6000H coincides with the launch of 5000U/H (CES), and if the launch 6000G Pro versions coincides with the launch of 5000G Pro (Q2), we could expect a public lauch of 6700G at the same time as 5700G, or... never, if AMD decides to do so.

Why did you first say that the upcoming Rembrandt will be released for OEM only and consumer AM5 motherboard will never be released?
You are the first person to have planted this doubt in my mind.

BTW, any guestimates of Rembrandt vs. Xbox / PlayStation?

PS5 is 8 Zen 2 cores and 36 CU RDNA 1.5 (custom design between RDNA1 and 2)
XBox is 8 Zen 2 cores and 56 CU RDNA2.
(Not to mention memory bandwidth and SSD bandwidth)
I think RMB is not comparable.

 

[Spicy]

How does it work with Cezanne? Isn't L3 dedicated only to the CPU, while GPU needs to go directly to memory?

Do you think it will change in Rembrandt? Some sort of sharing of L3? or 2 separate caches, CPU L3 + Infinity Cache?

No idea. I have no expertise in this area. Only my readings.
I think I remember that RMB will have 2MB of L2 cache. Two groups of 6 CU, with 1MB each.
But no Infinity Cache, unfortunately

 

[uzzi38]

How does it work with Cezanne? Isn't L3 dedicated only to the CPU, while GPU needs to go directly to memory?

Do you think it will change in Rembrandt? Some sort of sharing of L3? or 2 separate caches, CPU L3 + Infinity Cache?

Yes that is how it works on Cezanne, no RMB will not change anything in that respect afaik.

 

[ahimsa42]

perhaps the reported lack of infinity cache is another clue that rembrandt will be a single generation, OEM mobile only platform as a transition to AM5?

 

[leoneazzurro]

perhaps the reported lack of infinity cache is another clue that rembrandt will be a single generation, OEM mobile only platform as a transition to AM5?

AM5 is desktop platform, mobile package for Rembrandt is called FP7 according to rumors and generally these platforms are depending mainly on the system memory used (Cezanne/Renoir were FP6 ) and I/O.

 

[ahimsa42]

my bad- i meant DDR5 not AM5. FP7 is supposed to be only for rembrandt so phoenix will be on FP8.

 

[ahimsa42]

first rembrandt sighting? very interesting aside from the 3080 ti.

ASUS's Next-Gen ROG Zephyrus Duo 16 GX650 Laptop Rocks AMD Ryzen 9 6900HX 'Rembrandt' CPU & NVIDIA GeForce RTX 3080 Ti GPU

ASUS's flagship ROG Zephyrus Duo 16 GX650 laptop featuring the AMD Ryzen 9 6900HX CPU & NVIDIA GeForce RTX 3080 Ti GPU has leaked out.

wccftech.com

 

[eek2121]

Renoir desktop APUs (4000G/GE) were never released for consumers. Only for OEM (Dell, Asus, HP, ...).
Old rumors announce first AM5 motherboards for Q2 2022. But they will be for OEM only, probably.
If the launch of 6000U/6000H coincides with the launch of 5000U/H (CES), and if the launch 6000G Pro versions coincides with the launch of 5000G Pro (Q2), we could expect a public lauch of 6700G at the same time as 5700G, or... never, if AMD decides to do so.

Why did you first say that the upcoming Rembrandt will be released for OEM only and consumer AM5 motherboard will never be released?
You are the first person to have planted this doubt in my mind.

AM5 launches with Zen 4.

 

[Spicy]

AM5 launches with Zen 4.

And what about APU type 1 / Model 40h-4Fh?
Do you have any news about this? Canceled?

AMD Zen4 Ryzen CPUs confirmed to offer integrated graphics - VideoCardz.com

AMD to offer integrated on-chip graphics for next-gen Ryzen series The Next-generation of AMD Ryzen CPUs is now confirmed to offer integrated GPUs. Upcoming Ryzen CPU series based on Zen4 architecture will all offer integrated GPU by design. While this does not mean that literally all CPUs will...

videocardz.com

 

[uzzi38]

And what about APU type 1 / Model 40h-4Fh?
Do you have any news about this? Canceled?

AMD Zen4 Ryzen CPUs confirmed to offer integrated graphics - VideoCardz.com

AMD to offer integrated on-chip graphics for next-gen Ryzen series The Next-generation of AMD Ryzen CPUs is now confirmed to offer integrated GPUs. Upcoming Ryzen CPU series based on Zen4 architecture will all offer integrated GPU by design. While this does not mean that literally all CPUs will...

videocardz.com

Rembrandt, Raphael, Phoenix in that order.

 

[Spicy]

Yes, I'm agree. So AM5 launches with Zen3+, not Zen4. The question is: DIY or OEM?

VideoCardz said family number 19h is Zen 4, but it's wrong. It's Zen 3.

List of AMD CPU microarchitectures - Wikipedia

en.wikipedia.org

In fact, family 17h covers Zen, Zen+ and Zen 2. So, family 19h should cover Zen 3 and Zen 4.
Zen 2 is an optimization of Zen (with a node shrink), while Zen 3 is a new design. And Zen 4 will be an optimization of Zen 3 with a node shrink (with same familly number) while Zen 5 will be a new design (with a node shrink too). Tick-Tock.

 

[eek2121]

And what about APU type 1 / Model 40h-4Fh?
Do you have any news about this? Canceled?

AMD Zen4 Ryzen CPUs confirmed to offer integrated graphics - VideoCardz.com

AMD to offer integrated on-chip graphics for next-gen Ryzen series The Next-generation of AMD Ryzen CPUs is now confirmed to offer integrated GPUs. Upcoming Ryzen CPU series based on Zen4 architecture will all offer integrated GPU by design. While this does not mean that literally all CPUs will...

videocardz.com

Yes, I'm agree. So AM5 launches with Zen3+, not Zen4. The question is: DIY or OEM?

VideoCardz said family number 19h is Zen 4, but it's wrong. It's Zen 3.

List of AMD CPU microarchitectures - Wikipedia

en.wikipedia.org

In fact, family 17h covers Zen, Zen+ and Zen 2. So, family 19h should cover Zen 3 and Zen 4.
Zen 2 is an optimization of Zen (with a node shrink), while Zen 3 is a new design. And Zen 4 will be an optimization of Zen 3 with a node shrink (with same familly number) while Zen 5 will be a new design (with a node shrink too). Tick-Tock.

Rembrandt desktop launches AFTER Zen 4. Rembrandt mobile (on FP7) is the chip launching at CES.

EDIT: Also, if you are in doubt, AMD has already stated as much.

 

[Joe NYC]

Rembrandt desktop launches AFTER Zen 4. Rembrandt mobile (on FP7) is the chip launching at CES.

EDIT: Also, if you are in doubt, AMD has already stated as much.

I have not seen anything to that effect.

 

[soresu]

Zen 2 is an optimization of Zen (with a node shrink), while Zen 3 is a new design.

New design or not, to call Zen 2 a mere node shrink optimisation of Zen is clearly incorrect.

It doubles the µArch FP/SIMD resources, giving a greater perf improvement for SIMD heavy workloads (of which there are many consumer and beyond) than Zen 3 does over Zen 2 overall, while still giving a hefty 15% overall IPC improvement over Zen.

Though the 15% overall IPC might represent unexploited low hanging fruit in the Zen 1 design, the doubling of FP/SIMD resources is certainly not an optimisation - and probably viable only because of the 14/12nm -> 7nm node change.

 

[ahimsa42]

I have not seen anything to that effect.

i thought rembrandt is supposed to be mobile only?

 

[Joe NYC]

i thought rembrandt is supposed to be mobile only?

It will be primarily mobile, like Cezanne, but Cezanne made it to AM4 as well. So I would not be surprised if Rembrandt eventually makes it to AM5.

 

[Spicy]

New design or not, to call Zen 2 a mere node shrink optimisation of Zen is clearly incorrect.

It doubles the µArch FP/SIMD resources, giving a greater perf improvement for SIMD heavy workloads (of which there are many consumer and beyond) than Zen 3 does over Zen 2 overall, while still giving a hefty 15% overall IPC improvement over Zen.

Though the 15% overall IPC might represent unexploited low hanging fruit in the Zen 1 design, the doubling of FP/SIMD resources is certainly not an optimisation - and probably viable only because of the 14/12nm -> 7nm node change.

I'm agree, it is strange to say that Zen 2 is only an "evolution" of Zen. But it is not my opinion, it is the AMD's point of view, about their Zen µArchitectured famlily.

Here is an an old interview (end of 2019) of Forrest Norrod (AMD's senior vice president and general manager):

When asked about what kind of performance gain Milan's CPU core microarchitecture, which is known as Zen 3, will deliver relative to the Zen 2 microarchitecture that Rome relies on in terms of instructions processed per CPU clock cycle (IPC), Norrod observed that -- unlike Zen 2, which was more of an evolution of the Zen microarchitecture that powers first-gen Epyc CPUs -- Zen 3 will be based on a completely new architecture.
Norrod did qualify his remarks by pointing out that Zen 2 delivered a bigger IPC gain than what's normal for an evolutionary upgrade -- AMD has said it's about 15% on average -- since it implemented some ideas that AMD originally had for Zen but had to leave on the cutting board. However, he also asserted that Zen 3 will deliver performance gains "right in line with what you would expect from an entirely new architecture."
[...]
He also indicated that AMD's server CPU launches are set to rely on the "tick-tock" cadence that was once the hallmark of Intel CPU launches, with the launch of a CPU platform that relies on a new manufacturing process node but the same microarchitecture as the last platform (the "tick") followed by a platform that relies on a new microarchitecture but the same manufacturing process node (the "tock").

In this context, Rome (Zen 2) represents a tick, thanks to its use of a 7nm process that's much more advanced than the 14nm process used by Naples (Zen), while Milan (Zen 3) represents a tock, since it will feature a new microarchitecture but rely on a 7nm process. And presumably, AMD's fourth-gen Epyc platform (Zen 4) -- it's codenamed Genoa, due in 2021 and expected by many to rely on TSMC's next-gen, 5nm, process node -- will represent another tick.

They share the same family number (17h). This seems clear.

 

[tomatosummit]

It will be primarily mobile, like Cezanne, but Cezanne made it to AM4 as well. So I would not be surprised if Rembrandt eventually makes it to AM5.

The last rumour was a q1 launch for rembrandt laptop and q2 for rembrandt desktop.
People are speculating that it's an oem only launch for am5, ryzen 6000g and a low end chipset, possibly without a southbridge. Then q3 or q4 being the main retail consumer launch with zen4 cpus.

AMD has recently been very quiet on these kind of oem launches so I'd be suprised if we don't get a proper leak or information until well after the q2 launch. Even bristol ridge on am4 from years back was scant more than whispers.

 

[Spicy]

AMD has recently been very quiet on these kind of oem launches so I'd be suprised if we don't get a proper leak or information until well after the q2 launch. Even bristol ridge on am4 from years back was scant more than whispers.

OEM 5000G were launched in April 2021. DIY 5600G and 5700G were launched in August

 

[Spicy]

Form the stolen Gigabyte confidential documents, we can learn a little about what Zen 4 will be. From my point of view (I am not an expert), it looks like a slight evolution of Zen 3 µarchitecture, but probably enough to have important performance gains (rumors say +20%? +30%?).
Conversely, at the end of this interview, AMD says their next amazing architecture will be Zen 5.
Tick and Tock.

Zen+ was an oddity. My opinion is that this generation might have never existed, if Intel had been more aggressive at the time. Ryzen 2000 gave them an extra year to refine Zen 2.

 

[moinmoin]

Zen+ was an oddity. My opinion is that this generation might have never existed, if Intel had been more aggressive at the time. Ryzen 2000 gave them an extra year to refine Zen 2.

This was also while for a long time GloFo still seemed to be in play wrt 7nm.

 

[tomatosummit]

Form the stolen Gigabyte confidential documents, we can learn a little about what Zen 4 will be. From my point of view (I am not an expert), it looks like a slight evolution of Zen 3 µarchitecture, but probably enough to have important performance gains (rumors say +20%? +30%?).
Conversely, at the end of this interview, AMD says their next amazing architecture will be Zen 5.
Tick and Tock.

Zen+ was an oddity. My opinion is that this generation might have never existed, if Intel had been more aggressive at the time. Ryzen 2000 gave them an extra year to refine Zen 2.

It's no secret that zen archs are "new" design followed by improved design each increment. The last anand interview with mike clark stated that they feel refining each design multiple times has insignificant gains after the first iteration and it seems to serve amd better to to have a real ground up refresh every other iteration. Assuming the 18month cadence is a real thing then it's a new design every three years.

Zen+ was purely a refresh on an improved manufacturing process. A kabylake to skylake if you will.
300mhz, much better boosting algorithms and improved power delivery along with better motherboards and possibly agesa were where the performance came from. The supposed 1-3% ipc improvement I would argue is within the margin of error or a bug that was fixed in the previous b2 silicon that went into TR and epyc.

But rembrant is pretty much going to be zen3+ as it's going to be zen3 on N6 so hopefully we can see those improvements again and how they'll fare against the other zen iterations around at the same time, particularly against zen2 in van gogh as I think that's also N6?

 

[soresu]

I'm agree, it is strange to say that Zen 2 is only an "evolution" of Zen. But it is not my opinion, it is the AMD's point of view, about their Zen µArchitectured famlily.

Here is an an old interview (end of 2019) of Forrest Norrod (AMD's senior vice president and general manager):



They share the same family number (17h). This seems clear.

True, though family number is not necessarily an accurate indication of difference - even CDNA2 still retains a number similar to Vega (GFX9) despite the large amount of changes made to differentiate it from the OG Vega design.

Also Norrod is not the best authority on the subject of µArch details, or how to classify them - he is a sales/marketing specialist, rather than R&D like Mike Papermaster, Mike Clark, or Suzanne Plummer.

Given his speciality and the timeframe being after Zen 2 was released I think it isn't surprising that he would downplay Zen 2 in favor of playing up the mystery of the coming Zen 3.

I think the product family number has more to do with compiler targets, although what exactly needs to change for a compiler to need a different target I don't know - off the top of my head I don't know what instruction set changes happened between Zen 1/+ and 2.

I think cache infrastructure plays some part in compiler targets too - anyone knowledgable please feel free to disabuse me of that 😅.