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.

 

[Saylick]

The results align with the other reviews I think, in that RMB is great at 35W TDPs or less but does not scale all that much beyond 35W, at which Alderlake-M starts to shine. Given that the chip was targeted at slim, premium notebooks, it's hard to fault AMD given that RMB appears to achieve their goals nicely. In other words, RMB isn't an outright winner against ADL-M across all TDPs (and it wasn't designed to be), but it is very competitive and a good all-around SoC for the markets it was designed to compete in. If anything, the review showed that AMD really needs a higher IPC core design to go toe-to-toe against Golden Cove. The Zen 3+ cores are pretty much maxed out in ST performance and Alderlake-M is able to pull ahead in MT because it has more physical cores on die such that it can achieve higher throughput without needing to go further up the freq-voltage curve, kind of like how Renoir took the MT lead from Intel a few generations ago because it was able to pack 8 full cores while Intel was still on 4.

I wish AT did a comparison on battery life, but I understand such a comparison would be difficult since battery life apples-to-apples comparisons typically require that everything about the laptop outside of the SoC is the same to isolate all the variables.

 

[LightningZ71]

So.. Are you suitably shocked yet?

I for one concede I was too pessimistic in predicting 1050ti for desktop TDPs

I won't quite say "Shocked" but VERY pleasantly surprised at what I've been seeing in benchmarks so far. Being able to keep up with the 1650 Max-Q is a big deal for sure in mobile. I'm wondering more about how the desktop version will perform now...

 

[krumme]

Picasso was a well balanced product, and this Rembrant seems to be the same. Albeit at a higher price segment.
At realistic tdp levels.
It actually makes sense to use this for casual gaming in a slim 15-25w case. Imo its a fantastic product from a performance standpoint. Lets see how its priced. I wouldnt hold my hopes up as it have no compettition. But at least, like Picasso, its on a cheaper node, so it was made to be lean and cheap. No funky 5nm here.

The AT review is very interesting from a technical perspective and solid work here. But the idea in 2022 to test how a mobile 65w -cpu load only- performs seems far out to me. Its desktop. Serious...
Dont know, perhaps people get portable desktops today instead of regular desktops?

 

[jpiniero]

The AT review is very interesting from a technical perspective and solid work here. But the idea in 2022 to test how a mobile 65w -cpu load only- performs seems far out to me. Its desktop. Serious...
Dont know, perhaps people get portable desktops today instead of regular desktops?

You'd be surprised. There's plenty of gaming laptops sold with triple digit+ TDP GPUs.

 

[krumme]

You'd be surprised. There's plenty of gaming laptops sold with triple digit+ TDP GPUs.

Yeaa saw some of them and they are cheap vs making your own rig today. When Rembrant kind of closes the 1050 ti and below segment, people will start to wonder if they really need ultra or high settings or will make do with low or mid. At 144Hz 1440/1080 a midrange mobile gpu will be plenty. No need for a highend 800 usd desktop gpu even at more normal prices. Then why not have a heavy laptop instead of desktop. More social.

Obviously Adler is stronger above the 35w for the cpu part so probably Amd will bring some zen4 for that segment. Makes sense. Use 12nm for lowest end, then Rembrant for 9-35w and the majority of the market, and some binned zen 4 for the top.

 

[moinmoin]

Obviously Adler is stronger above the 35w for the cpu part so probably Amd will bring some zen4 for that segment. Makes sense. Use 12nm for lowest end, then Rembrant for 9-35w and the majority of the market, and some binned zen 4 for the top.

I don't even think Zen 4 will need to be binned once idle power usage of the IOD finally is being improved which should happen in time for Zen 4. Then desktop grade chips will go on a rampage for anything above 35W (which I still consider stupid in mobile products, but whatever), which I expect Raphael-H to target.

 

[ahimsa42]

first 6800U real world gaming benchmarks which i have seen

 

[Panino Manino]

His GPU is locked at 1900MHz? Why?

 

[DAPUNISHER]

His GPU is locked at 1900MHz? Why?

Considering the vid is providing bench results for our edification, the reason is bad, and they should feel bad -

Please bear in mind that I underclocked the iGPU to maintain low temps

 

[Zepp]

first 6800U real world gaming benchmarks which i have seen

nice

ETA prime https://www.youtube.com/c/ETAPRIME/videos
and AMD APU Gaming https://www.youtube.com/channel/UCJeQuSEYPNnYD75KA4K0fRA/videos channels have Rembrandt performance videos up as well

 

[moinmoin]

ETA prime https://www.youtube.com/c/ETAPRIME/videos
and AMD APU Gaming https://www.youtube.com/channel/UCJeQuSEYPNnYD75KA4K0fRA/videos channels have Rembrandt performance videos up as well

Only H/HS parts though.

 

[Abwx]

RMB CPU review at Computerbase :

AMD Ryzen 9 6900HS im Test

AMD Ryzen 6000 in Form des Ryzen 9 6900HS ist unterhalb von 45 Watt am stärksten. Intel Alder Lake hat dort bei der Effizienz das Nachsehen.

www.computerbase.de

A previous review was centered on the iGPU :

AMD Radeon 680M: Die Ryzen-6000-iGPU im Test

AMD Ryzen 6000 setzt erstmals auf eine iGPU mit RDNA 2. Der Leistungssprung, den die Radeon 680M auf dieser Basis im Test macht, ist groß.

www.computerbase.de

For the CPU efficency they use blender since it last long enough, contrary to CB, for the turbos being neutered.

 

[Markfw]

RMB CPU review at Computerbase :

AMD Ryzen 9 6900HS im Test

AMD Ryzen 6000 in Form des Ryzen 9 6900HS ist unterhalb von 45 Watt am stärksten. Intel Alder Lake hat dort bei der Effizienz das Nachsehen.

www.computerbase.de

A previous review was centered on the iGPU :

AMD Radeon 680M: Die Ryzen-6000-iGPU im Test

AMD Ryzen 6000 setzt erstmals auf eine iGPU mit RDNA 2. Der Leistungssprung, den die Radeon 680M auf dieser Basis im Test macht, ist groß.

www.computerbase.de

For the CPU efficency they use blender since it last long enough, contrary to CB, for the turbos being neutered.

I don't even know what language to use for google translate. Bottom line, how good is it ?

 

[Hitman928]

I don't even know what language to use for google translate. Bottom line, how good is it ?

It's German. Here's a translated link: https://www-computerbase-de.transla...l=auto&_x_tr_tl=en&_x_tr_hl=en&_x_tr_pto=wapp

Edit: A very brief read over seems to draw the same conclusions we already know, above 45W, ADL-H scales better and so has better performance. At <= 45W, Rembrandt catches and then surpasses ADL-H performance the further you go down in power. The iGPU in Rembrandt is significantly better than Xe in ADL.

Edit2: Looks like in long Blender run, RMB slightly edges ADL-H, even at 70W. This is at least partly due to Blender not favoring Intel as much as Cinebench as well as the long blender run pretty much negating the very high turbo powers ADL-H is allowed to reach.

 

[Abwx]

I don't even know what language to use for google translate. Bottom line, how good is it ?

As good as a 8C/16T favoured by a half node shrink can be, very efficient up to 70W although the sweet spot is in the 25-35W range.

If anything this review show that the efficency numbers published by other sites are BS when it comes to ADL, to sumarize it s benched at the higher turbo TDP and then stated as being using long term TDP.

This is possible thanks to Cinebench runs being short enough to allow running at max power while the duration of Blender more or less level the excess power gains.

 

[moinmoin]

This is possible thanks to Cinebench runs being short enough to allow running at max power while the duration of Blender more or less level the excess power gains.

I'm not following all those reviews, is there any which translates all those boosting shenanigans into simple Joules per workload numbers?

 

[Abwx]

I'm not following all those reviews, is there any which translates all those boosting shenanigans into simple Joules per workload numbers?

For all CPU tested in Blender BMW scene TDP is stabilsed after 30s at most if we except a single Intel SKU, so the long duration bench wich last at least 15mn is quite accurate for efficency estimations.

Removing those turbo tricks we can see that RMB@35W is 2% faster than a 12700H@45W and 12% faster at same 45W power.

At 70W for both RMB is 5% faster than a 12900HK, so all reviews are overestimating the latter s perf/Watt since few benches are that long.

 

[The Hardcard]

AMD Ryzen 9 6900HS Review

The new Ryzen 6000 mobile CPUs upgrade AMD's mobile offering with Zen 3+ cores, RDNA2 graphics, and DDR5 memory. Today we're checking out the Ryzen 9 6900HS,...

www.techspot.com

Techspot/ Hardware Unboxed review

 

[igor_kavinski]

AMD Ryzen 9 6900HS Review

The new Ryzen 6000 mobile CPUs upgrade AMD's mobile offering with Zen 3+ cores, RDNA2 graphics, and DDR5 memory. Today we're checking out the Ryzen 9 6900HS,...

www.techspot.com

Techspot/ Hardware Unboxed review

Not that impressive in this review. Losing quite a few benchmarks.

 

[Hitman928]

Not that impressive in this review. Losing quite a few benchmarks.

It's a 35W 6900HS versus a 45W 12700H in most of those benches. The 12700H also uses up to ~120W during boost versus I believe ~75W for the 6900HS. So it's not exactly a surprise that the 6900HS is behind in performance in a lot of cases. There's no battery life or perf/w test in the review so ADL-H comes off looking golden, but if you compared battery life or efficiency, it would take the shine off of it quite a bit. Techspot/HWU's ADL-H samples also seem to outperform almost all the other reviews I've seen for some reason. Not sure if they just got some really good samples or if their TDP restricted settings aren't getting rid of the boost power states like they think they are.

 

[Abwx]

Not that impressive in this review. Losing quite a few benchmarks.

If most benches are short enough such that one CPU can achieve most of the score at 115W then no wonders that it can win quite a lot of benches against one that has a 70W boost.

Beside in the longer benches where this boost is partly leveled one chip is at 45W for the rest of the loading while the other is at 35W.

Actually only Computerbase managed to sort out a real comparison at several power levels.

 

[ahimsa42]

Not that impressive in this review. Losing quite a few benchmarks.

personally i don't care at all about synthetic benchmarks. the in game tests have been exactly what AMD promised if not even better and there will also likey be future improvements due to driver optimization. once the 6800U machines become available, if one wants to game on a laptop with no igpu there really are no other options to consider in 2022.

 

[Panino Manino]

I'm suprised buy the lack of battery benchmarks in these early reviews.
Longer battery was one of the selling points and should show the differeence from the competition the most.

 

[uzzi38]

I'm suprised buy the lack of battery benchmarks in these early reviews.
Longer battery was one of the selling points and should show the differeence from the competition the most.

UltrabookReview did a nice breakdown at least of power consumption in varying workloads

It's only vs the last gen G14, but still, at least someone's done a proper breakdown.

 

[scineram]

The leaked Surface spec sheet shows +2h for the RMB units over ADL.