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.

 

[eek2121]

When OEMs feel like it? I'll point out that we are talking about mobile only.

The desktop RMB likely will be launched together with the AM5 platform, which I expect to happen closer to the launch of Zen 4 (whenever that is).

The chips that are in production are mobile variants. There is a TINY chance we could see a desktop release, but I am expecting, barring leaked information to the contrary, to not see a desktop variant until at least 12 months from the release of Zen 3 APUs. The only factor that may move up this deadline is a desire to get off 7nm to improve production capacity. Even in that case, I doubt we'll see anything before the middle of next year.

I guess we'll need to make a difference between mobile and desktop APUs from here on out. The Ryzen APUs so far exist to be competitive in the mobile market, i.e. with competitive integrated graphics performance. I don't expect that to happen with Raphael. There graphics performance is not the issue, instead having graphics at all has been the competitive advantage Intel desktop chips enjoyed over Ryzen CPUs so far that Raphael appears to put an end to.

AMD likely will differentiate them for us. A guess: Raphael will likely be Ryzen 6600 and up, Rembrandt will likely be the models below that (on the desktop at least). Of course, this is wild (somewhat educated) guessing at this point. I don't usually make guesses about stuff I know little about, so it will be interesting to see how it plays out.

A top end "Zen 3+" APU will likely be slower than a Zen 4 chip with GPU. "Zen 3+" implies a smaller die area and slightly higher clocks. I suspect even the fastest "Zen 3+" APU will be slower than a 6 core AMD Zen 4 product. It also fits in with market segmentation. 6nm is far cheaper than 5nm. Even if the 6nm die is monolithic.

 

[Insert_Nickname]

The chips that are in production are mobile variants.

It's the same die used for mobile and desktop. Those dies that for some reason don't meet the criteria for mobile end up as desktop APUs. It's all about the binning.

 

[eek2121]

It's the same die used for mobile and desktop. Those dies that for some reason don't meet the criteria for mobile end up as desktop APUs. It's all about the binning.

I was referring to the final product. I am aware they are the same die.

 

[Insert_Nickname]

I was referring to the final product. I am aware they are the same die.

Oh, you meant packaged for mobile. My mistake.

 

[eek2121]

Oh, you meant packaged for mobile. My mistake.

No, I should have been more clear. I apologize. Mobile makes me lazy. 😀

 

[ahimsa42]

Van Gogh seems to offer excellent performance in the new steam deck even with zen 2. so will Rembrandt be similar or less/greater?

Steam Deck Handheld Console Benchmarked - 60 FPS With Decent Image Quality In Several AAA Games, 30 FPS in Cyberpunk 2077 at Native Resolution

The first performance benchmarks of the Steam Deck handheld console in several AAA and eSports games have been published

wccftech.com

 

[GreenReaper]

The new Steam Deck is using 16GB LPDDR5 linked to a 20 CU RDNA 2 graphics core. It is not clear that we will get 20 CUs in a regular APU. Also, the screen is only 1280x800 60Hz (although you can link others) - so the performance may be slightly misleading. Most testers also weren't trying to run in ultra settings, which can be both GPU and CPU-intensive.

Overall, I would expect Rembrandt to be as fast or faster, especially if in a bigger device without such strict thermal or power constraints. But graphics-wise, the power will depend on the CU configuration; and the further you push the resolution, the slower it will go. Also if Rembrandt does not come with Infinity Cache equivalent, that might crimp performance compared to the uplift seen in dedicated cards.

 

[Kepler_L2]

Van Gogh seems to offer excellent performance in the new steam deck even with zen 2. so will Rembrandt be similar or less/greater?

Steam Deck Handheld Console Benchmarked - 60 FPS With Decent Image Quality In Several AAA Games, 30 FPS in Cyberpunk 2077 at Native Resolution

The first performance benchmarks of the Steam Deck handheld console in several AAA and eSports games have been published

wccftech.com

Rembrandt is significantly faster, 50% more CUs, higher clocks, higher power limits and better IMC.

 

[moinmoin]

Van Gogh seems to offer excellent performance in the new steam deck even with zen 2. so will Rembrandt be similar or less/greater?

Van Gogh is the new low end APU for low power systems. That you are able to get a full system including this chip as portable for $399 should tell you everything. Every new AMD APU after this should improve on it (except for AV1 support for whatever crazy reason...).

 

[Bigos]

The new Steam Deck is using 16GB LPDDR5 linked to a 20 CU RDNA 2 graphics core. It is not clear that we will get 20 CUs in a regular APU.

Steam deck is using Van Gogh, which has 8 CUs.

 

[ahimsa42]

so perhaps a 6800U will have around RX560 performance?

 

[rainy]

Rembrandt is significantly faster, 50% more CUs, higher clocks, higher power limits and better IMC.

Not to mention Zen 3+ cores which increase even more gap between Zen 3 vs Zen 2.

 

[TESKATLIPOKA]

Rembrandt would have been better for Steam deck than Van Gogh.

 

[moinmoin]

Rembrandt would have been better for Steam deck than Van Gogh.

Depends on whether one considers shorter battery life and higher price better.

 

[TESKATLIPOKA]

I only talked about performance.
They can limit TDP to be comparable to Van Gogh.
The price would end up higher, true.

 

[ahimsa42]

has there been any word of infinity cache being implemented on rembrandt cpu's?

 

[Kepler_L2]

has there been any word of infinity cache being implemented on rembrandt cpu's?

It doesn't have any.

 

[Bigos]

has there been any word of infinity cache being implemented on rembrandt cpu's?

Yes, there was a word. No infinity cache (L3 cache).

Page 219 - Question - Speculation: RDNA2 + CDNA Architectures thread

Page 219 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Still not 100% confirmed before release, obviously.

 

[Joe NYC]

has there been any word of infinity cache being implemented on rembrandt cpu's?

That would totally make sense, but I have not heard anything.

It seems like AMD seems to want to fall short of making these APUs into killer product, that would fill the gap between iGPU and dGPU performance... Maybe it would be challenging for AMD to price such a product appropriately in the market place...

 

[Shivansps]

It seems like AMD seems to want to fall short of making these APUs into killer product, that would fill the gap between iGPU and dGPU performance... Maybe it would be challenging for AMD to price such a product appropriately in the market place...

Oh they are going to regret that if the DG2 ends up to be good at entry level. Either way, i think it is because having IF cache what is the GPU L3, along with another CPU L3 is having two independient L3 systems make no sence for them, it should be one big shared L3 or a shared L4. I think they are going to figure that out first.

Anyway, the roumored Navi 24 shouldnt be THAT much faster than the >>>fully enabled RMB<<<.

 

[LightningZ71]

For the purpose of gaming, as long as AMD's APUs have GPU performance that's at least relatively close to Intel's Xe product, then it's not really going to make a big difference to the end user and their purchasing decisions. I haven't seen many games that are noticeably different in their performance between the 96eu Xe and the 8CU vega in properly implemented laptops. Save for a few titles that seem to really like one architecture over another, it's rarely more than a few FPS difference at the same settings.

We'll see what the next round looks like between them. A good RDNA2 implementation should be a noticeable step up from VEGA8, at least based on improved memory usage efficiency. It's likely we'll see Intel move to 128eu of an improved design for their "APU's" How that shakes out is anyone's guess, but, in my opinion, both products will be able to produce good looking visuals at 1080p with playable frame rates on most reasonably coded titles. Both of the current products have been doing that at 720p/900p upscaled for current games for a while now.

 

[Joe NYC]

Oh they are going to regret that if the DG2 ends up to be good at entry level. Either way, i think it is because having IF cache what is the GPU L3, along with another CPU L3 is having two independient L3 systems make no sence for them, it should be one big shared L3 or a shared L4. I think they are going to figure that out first.

Anyway, the roumored Navi 24 shouldnt be THAT much faster than the >>>fully enabled RMB<<<.

AMD has innovative solution for improved performance on standalone CPUs and GPUs with caches, AMD has been good in combining CPU and GPU, now AMD just needs to put all the pieces together.

I think most likely, Rembrandt just missed the train, in putting the puzzle pieces together, having a stacked L3 (combined or separate).

 

[uzzi38]

I think most likely, Rembrandt just missed the train, in putting the puzzle pieces together, having a stacked L3 (combined or separate).

Except Raphael don't have one either

 

[moinmoin]

Except Raphael don't have one either

Wouldn't Raphael's iGPU be weaker anyway?

 

[Joe NYC]

Except Raphael don't have one either

That's a bummer about the whole situation, since AMD has all the pieces of the puzzle, but still not able to put them together.

How many years has it been since ATI acquisition? 15?