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.

 

[IntelUser2000]

I will get them from Anandtech then

No, you need to find a test that takes different SKUs from the same layout/arrangement that has different cache sizes and the clock speed fixed.

5700G is a different chip from 5800X.

 

[Abwx]

Oh it does, look at this. That's nearly 5% IPC boost right there

View attachment 55473

5700G has 4.6GHz single core while the 5800X is at 4.7GHz, but still, according to Computerbase who have reliable numbers the latter perform 6% better in CB R23 ST, the difference is 7% for R20 and 8% for R15.

Page 42 - Question - AMD Rembrandt/Zen 3+ APU Speculation and Discussion

Page 42 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[nicalandia]

5700G has 4.6GHz single core while the 5800X is at 4.7GHz,

The one I posted was for the Non-X Desktop 5800(non apu)

 

[Hitman928]

You think that AMD labs did use a very poor cooler on a third rate 5800U to downplay its scores.?.
So bad that it didnt even manage to dissipate the power in ST mode..?.

Edit :

Likely that they are comparing to the numbers they published when the 5800U was released.

I don't understand what you are talking about. I'm not saying the 5800u couldn't sustain it's single core max boost, but having used a lot of laptops in my time, I also understand that absolute sustained max boosts aren't a guarantee. Has nothing to do with the thermals but rather the power management algorithms and little background tasks that pop up that momentarily drop the clock.

 

[Abwx]

The one I posted was for the Non-X Desktop 5800(non apu)

That change nothing, there s something like 4-5% difference between the two chips once frequencies are normalized.

For RMB we have an official comparison from AMD that point to 4% IPC uplift in ST, dunno why there s some skepticism about the thing.

 

[nicalandia]

No, you need to find a test that takes different SKUs from the same layout/arrangement that has different cache sizes and the clock speed fixed.

5700G is a different chip from 5800X.

5700G is Monolithic Zen3 with half cache than the desktop 5000 series, I believe I posted the non-X 5800 CPU with is for OEMs with boosts of 4.6

 

[IntelUser2000]

5700G is Monolithic Zen3 with half cache than the desktop 5000 series, I believe I posted the non-X 5800 CPU with is for OEMs with boosts of 4.6

The non-X 5800 is still based on the same die.

That's pretty much the fundamentals of benchmarking - keep everything as equal as possible.

Especially when traditional knowledge says Cinebench doesn't really care about caches.

 

[nicalandia]

That change nothing, there s something like 4-5% difference between the two chips once frequencies are normalized.

For RMB we have an official comparison from AMD that point to 4% IPC uplift in ST, dunno why there s some skepticism about the thing.

I agree, I mean it's Zen3+, it's like what Zen+ was to OG Zen, refined and overall higher performance due to IPC and better process

 

[IntelUser2000]

I agree, I mean it's Zen3+, it's like what Zen+ was to OG Zen, refined and overall higher performance due to IPC and better process

It's probably 3-5% faster like Intel Ticks and Zen+.

 

[Hans de Vries]

There's your 15W numbers with footnotes, now stop complaining

View attachment 55457

View attachment 55458

Thanks Uzzi38

The battery live apparently improves under the general use case even when the 6800U has a 28W max TDP when compared to the 15W TDP 5800U. That might be true for general web browsing and watching videos.

 

[Hans de Vries]

The interesting take-away is that: Even when operating with a higher 28W max TDP, the 6800U has a longer battery live as the 15W TDP 5800U in a general use case. With all those new power management features, adaptive power management framework and new sleep states it requires less watts over time in the general use case.

 

[eek2121]

Those aren't footnotes for the presentation though, just the statements on that product webpage , none of which mention Intel. So why would there be a system config for it?

The 25w vs 28w is scummy - just trying to hit that nice 2x figure for the wow factor.

Nothing scummy about it. AMD increased TDPs this generation to more closely match Intel. Intel’s U chips were 12-28W, AMD decided to do 15-28W.

You read too much into stuff. Rembrant is faster than Cezanne.

 

[flatwhite]

Not much 1080p gaming performance improvement in AAA titles for Alder Lake compared to 11th Tiger Lake

 

[tamz_msc]

So there s enough cores on a 4C CPU, and since perfs are vastly different this cant be due to CPUs that perform the same given that they are not TDP limited to 15W, only remaining explanation is the GPU and BW, MX450 has a 64b bus with GDDR6 but RMB has a 128b bus.

We do not know what configuration the MX450 test system was. There is no way it loses to an Iris Xe 96 EU like it does on some games in that chart, unless it's the crippled 12W version, or if the game is handicapped by 2GB VRAM.

EDIT: Looks like they've tested the Dell Inspiron 13 5310, which has the 25 W MX450 but throttles severely under gaming, behaving more like a 12 W MX450.

AMD Unveils New Ryzen Mobile Processors Uniting “Zen 3+” core with AMD RDNA 2 Graphics in Powerhouse Design

Browse AMD’s company-wide and financial press releases.

ir.amd.com

https://www.notebookcheck.net/Dell-Inspiron-13-5310-review-More-graphics-more-battery-life.558503.0.html

The above "The Witcher 3 Ultra - Graphics stress for the MX450" shows how the GPU clock speeds drops from 1800 in the first minutes to 1200 MHz, because the system tries to reduce the heat development from almost 70 ºC (158 ºF) down to significantly below 60 ºC (140 ºF). The throttling also happens in other games, as "Strange Brigade" or "X Plane 11.11" also show.

So in short, while Dell uses a powerful version of the MX450 with 25 watts, it is not being used to its full advantage. Even the slowest N18S-LP version in the HP Envy 13-ba1475ng gets a better performance in real games in some respects ("The Witcher 3," "X-Plane 11.11").

So in effect the comparison was made against an MX450 that's closer to the 12 W variant than the full 25 W variant. No wonder the graphs are off.

 

[TESKATLIPOKA]

There's your 15W numbers with footnotes, now stop complaining

View attachment 55457

View attachment 55458

Thanks, uzzi38. So at 15W TDP, It is ~17% faster in MT than R7 5800U.
This should be more than enough to make R7 6800U faster than Alder Lake U within 15W TDP, although ST could be a little faster on ADL.
Rembrandt looks good, especially 6800U, 6900HS and 6980HS.

 

[TESKATLIPOKA]

How many TFLOPs is RMB iGP at peak then? 3.6ish?

Yes, 2FLOPs * 12CU * 64SP * 2.4GHz = 3686 GFlops
For comparison, Xe 96EU in ADL-H peaks at only 2227 GFlops.

 

[TESKATLIPOKA]

Here is the official slide about Cinebench R23 with footnotes. It is R7 5800U 15W vs R7 6800U 28W.

I don't know why the CB score is so low, when even R7 5800U set at 25W TDP manages ~25% higher score compared to 15W. Let's not forget about the ~17% performance increase in 9 Multi-threaded tests at 15W TDP that uzzi38 posted here.

Edit: Yes, I will wait for reviews, It's not like we have other options. For example, Notebookcheck.net mentions PL1 and PL2 in their reviews.

 

[leoneazzurro]

Mostly because CB has become less a CPU test in general and more a FP unit test, which is quite probably unchanged between Zen3 and Zen3+. The ST test gains are more indicative as with almost certainity they are not influenced BY TDP and thus are more related to the ST boost clock difference but nevertheless there is a 10% gain with a 6% boost clock difference, meaning there is probably anyway a small IPC uplift. While in other workloads you can see different gains respect to CB because these could take advantage of the other innovations (i.e. RAM bandwidth). In CB MT you are almost surely TDP limited, but comparisons are tricky as the "nominal" TDP may be different from the actual power draw of the chip/system . So we need to wait for actual reviews. My impression is that on CPU side the IPC is there but quite limited, while general power efficiency was substantially improved even after adding a ton of other features (imptoved iGPU, USB4, PCIE 4.0, and so on), which in many cases go in the opposite direction of power reduction.

 

[Gideon]

Not much 1080p gaming performance improvement in AAA titles for Alder Lake compared to 11th Tiger Lake

I remember @mikk claiming it will double and AMD needs to double their own just to match it

 

[Shivansps]

The RX6500XT numbers with 16MB IC and 16 CUs

Supports the idea of RMB IGP with good rams being right at GTX1650 levels. If not TDP limited. And remember, RMB IGP can use up to 8GB VRAM.

 

[ryanjagtap]

Does anyone think that with the igp and dgpu in AMD advantage laptops being RDNA2, AMD will do something along the lines of Radeon Dual Graphics which it used to have or some other features similar to Intel Deep Link?

 

[Abwx]

Here is the official slide about Cinebench R23 with footnotes. It is R7 5800U 15W vs R7 6800U 28W.

I don't know why the CB score is so low, when even R7 5800U set at 25W TDP manages ~25% higher score compared to 15W. Let's not forget about the ~17% performance increase in 9 Multi-threaded tests at 15W TDP that uzzi38 posted here.

That mean that the 5800U ran the bench at about 20W average, wich is in line with 15W nominal and 25W cTDP.

The RX6500XT numbers with 16MB IC and 16 CUs

Supports the idea of RMB IGP with good rams being right at GTX1650 levels. If not TDP limited. And remember, RMB IGP can use up to 8GB VRAM.

There s the IF cache and 144GB/s bandwith, as to being TDP limited, why should it be the case with a 107W card.?.

Radeon RX 6500 XT: AMD bringt RDNA 2 mit TSMC N6 für 199 $ in den Handel

AMD hat auf der CES die Radeon RX 6500 XT und damit ein neues Einstiegsmodell des RDNA-2-Portfolios angekündigt. 199 $ lautet der Preis.

www.computerbase.de

 

[LightningZ71]

6400 numbers should be a lot closer, but, even that has IF cache.

 

[JoeRambo]

With all those new power management features, adaptive power management framework and new sleep states it requires less watts over time in the general use case.

It's "up to 15%" in web browsing, that is not that big of advance, but still nice optimization. The video cases saving more depend on hardware acceleration for encoding and decoding and benefit from new RDNA2 features.

 

[ahimsa42]

seems to be a lot of controversy over rembrandt but i am very pleased at what was revealed yesterday. as long as the pricing is in line & there is good availability (along with igpu only options of course!) i don't see how this is not a major win for both AMD & consumers. i am looking forward to the in game benchmarks coming out in the next few months & will certainly be waiting for a memorial day/labor day sale to snag a 15" 6800U to replace my current 4500U (which i hope to sell for $250-300 to help offset the cost of the new laptop).