Discussion Intel current and future Lakes & Rapids thread

[IntelUser2000]

The first one is the 30% figure being referenced I presume? In which case note that the claim is that the processors can use up to 30% less power, not the system. Page 8 of this AMD presentation comparing the 4000 series mobile to 5000 series illustrates the difference, and in that case how battery life can increase more than the power reduction in the processor - https://www.amd.com/system/files/documents/battery-life-whitepaper.pdf

For 5000 they did that yes, but for 6000 it's different: https://i.itworldcanada.com/wp-content/uploads/2022/01/amd-ces-2022-ryzen-6000-mobile-8.jpg

What I'm currently more interested in is how ADL will compare to TGL in battery life. Substantive claims on that aspect were suspiciously absent in Intel's announcements. There are obviously going to be improvements compared to TGL, but to what extent will have to wait for reviews I guess.

The only way for Intel to be significantly better is to go on-die PCH like with AMD. Yes, AMD had that since pre-Ryzen but they were behind in power management.

Meteorlake with Foveros is the first generation that'll make it possible, as it becomes an effective on-die, or will get close to it. Of course possible doesn't guarantee it as Lakefield demonstrated.

 

[mikk]

Just the opposite - 15W Cezanne U is very rare.

NBC samples they have of the 5800U:

25/30
25/30
22/27
25/30

This is false, check out the tests. You cannot use the power numbers from their 5800U overview, they are incorrect.

25/30= 7872
25/30= 8068
25/30= 8154
22/27= 9539


You can see from this it's false, the lowest power budget device gets the highest Cinebench R23 score. The Aero 13 which is 25/30 in the overview only uses 15W in the Witcher 3 stresstest in their test. The fastest in this overview, a HP ProBook x360, uses 27W for some seconds and sustained 22W they say, means this power budget number in the overview seems to be correct unlike the others. On ultrabookreview there is a ZenBook 13 review with a 5800U, it scores 8777 and uses 19W of sustained power.

To summary: ~15W= 8000 points, ~19W= 8800 points, ~22W= 9500 points.

Apparently there is no 28W device on notebookcheck. A real 28W device should score higher than this in Cinebench R23.

Dude the scores are from CPUMonkey. I don't know where they are getting that from and I might have been bit hasty on the claims but Alderlake isn't the magic bullet for mobile.

CPUMonkey is a fake page. They are guessing the scores of upcoming CPUs.

 

[IntelUser2000]

This is false, check out the tests. You cannot use the power numbers from their 5800U overview, they are incorrect.

It's not "false" or "incorrect". They are based on numbers from real reviews. The problem is that they average out the numbers, which is what makes exact comparisons very difficult.

Alderlake's big gain over the predecessor is in an area where the market couldn't have cared less, otherwise Renoir and Cezanne would have took significant marketshare. Actually maybe this is a bit of AMD's fault too, but still.

Trust me, I embrace the hybrid methodology and love the E cores but like I said before, Intel's silence regarding Alderlake-U/P is deafening. These guys love to tout their horn, why the silence? Intel's CES presentation was a snoozefest. Not one mention really?

Because probably, maybe, umm, it loses in MT significantly despite the massive gains, the battery life is nothing special and probably behind, and of course the iGPU is slower too. The latter two I bet you the market cares much more. If they have an AMD convertible under 2.5lbs and if I was looking to buy one now, I'd be seriously tempted.

Intel had a massive battery life lead with Haswell, extended it with Broadwell, regressed a bit with Skylake but brought new features for more fine tuning, and Kabylake made it really good.

Icelake and Tigerlake regresses on that department. Man, they have so much more work to do! This part will not be fixed until Meteorlake at the earliest.

 

[mikk]

It's not "false" or "incorrect". They are based on numbers from real reviews. The problem is that they average out the numbers, which is what makes exact comparisons very difficult.

You are wrong, check out the real test. Power budget is lower in 3 of the 4 devices and the Cinebench scores are too low anyways. Only the HP ProBook x360 is more or less accurate and runs much faster.

However there are some real 25W devices, HP Pro Book 455 G8 for example. 9661 points in Cinebench R23.

 

[Khato]

For 5000 they did that yes, but for 6000 it's different: https://i.itworldcanada.com/wp-content/uploads/2022/01/amd-ces-2022-ryzen-6000-mobile-8.jpg

The only way for Intel to be significantly better is to go on-die PCH like with AMD. Yes, AMD had that since pre-Ryzen but they were behind in power management.

Meteorlake with Foveros is the first generation that'll make it possible, as it becomes an effective on-die, or will get close to it. Of course possible doesn't guarantee it as Lakefield demonstrated.

Thanks for the link - not certain why my brief searching for battery life claims on the mobile 6000 series didn't yield it. Regardless, their choice of tasks there is interesting in that they show far more gains on the two with constant network traffic compared to web browsing. If you take their increase from 21.5 hours of local video playback on the mobile 5000 series to 24 hours on the mobile 6000 series as a baseline (10% system power reduction) then an additional 5% on web browsing and 20-30% on network streaming applications implies a nice efficiency increase in connectivity. I would say WiFi, but the footnotes state "WLAN enabled and disconnected" which is a bit odd for laptop testing.

Agreed that Intel removing the PCIE power overhead on communication to the PCH will definitely help. Having it on package minimizes the penalty, but just makes it comparable to AMD's chiplet approach on desktop.

 

[insertcarehere]

From AMD's own footnotes:

Engineering projections are not a guarantee of final performance. Performance projection by AMD engineering staff based on expected battery life for idle, video playback, modern standby, and esports gaming vs. Ryzen 5000 series mobile processors with the same TDP. Power reduction projections based on pre-silicon engineering analysis . Estimated battery life increases assume identically HW & SW configured systems. Specific projections are based on reference design platforms and are subject to change when final products are released in market. RMB-3

Take those battery life claims with some large grains of salt, they're not based on actual chipsets and devices.

 

[uzzi38]

Thanks for the link - not certain why my brief searching for battery life claims on the mobile 6000 series didn't yield it. Regardless, their choice of tasks there is interesting in that they show far more gains on the two with constant network traffic compared to web browsing. If you take their increase from 21.5 hours of local video playback on the mobile 5000 series to 24 hours on the mobile 6000 series as a baseline (10% system power reduction) then an additional 5% on web browsing and 20-30% on network streaming applications implies a nice efficiency increase in connectivity. I would say WiFi, but the footnotes state "WLAN enabled and disconnected" which is a bit odd for laptop testing.

Agreed that Intel removing the PCIE power overhead on communication to the PCH will definitely help. Having it on package minimizes the penalty, but just makes it comparable to AMD's chiplet approach on desktop.

This is what you're looking for.

 

[IntelUser2000]

If you take their increase from 21.5 hours of local video playback on the mobile 5000 series to 24 hours on the mobile 6000 series as a baseline (10% system power reduction) then an additional 5% on web browsing and 20-30% on network streaming applications implies a nice efficiency increase in connectivity. I would say WiFi, but the footnotes state "WLAN enabled and disconnected" which is a bit odd for laptop testing.

Majority of laptops use the Intel WiFi anyways. Switching out the default Qualcomm WiFi for the Intel AX200 is what allowed my previous Haswell laptop to go from 2W CPU package power to 1W. And I got the maximum 866Mbps speed downstairs and 2 rooms away from where the router is.

Agreed that Intel removing the PCIE power overhead on communication to the PCH will definitely help. Having it on package minimizes the penalty, but just makes it comparable to AMD's chiplet approach on desktop.

The power is not the problem, because we're talking 200mW at max. The problem is latency.

Actually, you are right too, power is the problem, but the chipset not being able to power down is the problem.

Modern CPUs go from the lowest C state to active state hundreds of times a second. If the transition is too slow, it'll simply not power down, or forced to go down to much higher C2/C3 state rather than C6 and below.

Icelake is awesome in idle battery life: https://www.notebookcheck.net/Hot-Ice-Dell-XPS-13-7390-2-in-1-Core-i3-Laptop-Review.434484.0.html#toc-8

35 hours and 39 minutes screen-on idle! But no one cares about idle battery life. Yes, normally the idle power matters, but in Icelake's case it can't power down fast enough to take advantage of the low idle power anyway.

You need on-die PCH to make it transition fast enough. Of course simple port to on-die(or Foveros) is not enough. You have to do whatever tech "magic" is needed to take advantage of it.

 

[szrpx]

Alder Lake’s Power Efficiency – A Complicated Picture

Reviews across the internet show Alder Lake getting very competitive performance with very high power consumption. For example, Anandtech measured 272 W of package power during a POV-Ray run. Our o…

chipsandcheese.com

Great analysis by Chips&Cheese

 

[Hulk]

Alder Lake’s Power Efficiency – A Complicated Picture

Reviews across the internet show Alder Lake getting very competitive performance with very high power consumption. For example, Anandtech measured 272 W of package power during a POV-Ray run. Our o…

chipsandcheese.com

Great analysis by Chips&Cheese

That is really interesting. It looks to me like ADL at optimum efficiency is about 4.2GHz for Golden Cove and 3.2GHz for Gracemont. When I get the time I'm going to run some benches at those frequencies and compare the reduction in power to the reduction in performance.

Also, I've stated this before when it comes to CPU's you have Efficiency, Compute (throughput), and Cost. Pick two.
Intel picked Compute and Cost with ADL. If you think about it a lack of efficiency isn't paid for by Intel's bottom line but instead the purchaser's bottom line! Us! Brilliant strategy actually.

 

[moinmoin]

Alder Lake’s Power Efficiency – A Complicated Picture

Reviews across the internet show Alder Lake getting very competitive performance with very high power consumption. For example, Anandtech measured 272 W of package power during a POV-Ray run. Our o…

chipsandcheese.com

Great analysis by Chips&Cheese

By the look of it Intel should have put an integer-only Gracemont core into each Golden Cove core and use that for integer workload at below 3.2 GHz and Golden Cove for everything else.

 

[coercitiv]

That is really interesting. It looks to me like ADL at optimum efficiency is about 4.2GHz for Golden Cove and 3.2GHz for Gracemont. When I get the time I'm going to run some benches at those frequencies and compare the reduction in power to the reduction in performance.

You can also use this as reference, it's a test of how ADL behaves around the V/f curve inflection points for both core types. Keep in mind that running 4.2Ghz on P will over-volt the E cores by 50-60mV @ 3.2Ghz.

Here's the 12700K running 3.6Ghz P-core / 3Ghz E-core / 3Ghz bus, scores just under 18K in CB23 while staying under 75W. The CPU is not undervolted but the motherboard AC DC Loadline parameters are set to favorable settings (on Auto my board overvolts the CPU like crazy). RAM is overclocked but I can't be bothered to undo that as well, not for CB testing anyway.

View attachment 55712

 

[eek2121]

Alder Lake’s Power Efficiency – A Complicated Picture

Reviews across the internet show Alder Lake getting very competitive performance with very high power consumption. For example, Anandtech measured 272 W of package power during a POV-Ray run. Our o…

chipsandcheese.com

Great analysis by Chips&Cheese

My only gripe is with the AMD system comparison. They used Windows on the AMD side, which makes the comparison meaningless.

 

[nicalandia]

Double post

 

[uzzi38]

HUB have got the best ADL-H results I've seen so far.

The power scaling results are also significantly better than those we've seen so far. These are definitely much more positive results for ADL-H overall. I'm surprised by the level of difference, but knowing it's there is neat.

 

[moinmoin]

The power scaling results are also significantly better than those we've seen so far. These are definitely much more positive results for ADL-H overall. I'm surprised by the level of difference, but knowing it's there is neat.

Personally I think none of these power scaling results are anything to write home about. 11800H and 12700H manage the best power scaling in the group, achieving ~1.5 times the performance with 2.7 times the power. While better than the others that's still quite the loss in efficiency, making them the one-eyed among the blinds. Seems like 35W (or lower) remains the efficiency sweet spot for now. So for me the U series remains more interesting.

 

[Dayman1225]

Bunch of leaks of Sapphire Rapids 48c x2 system:

Spoiler: Benchmarks for SPR

https://twitter.com/i/web/status/1492476537785835530

https://twitter.com/i/web/status/1492469096482422785

https://twitter.com/yuuki_ans/status/1492459664260272134?s=21

https://twitter.com/yuuki_ans/status/1492453197830524934?s=21

https://twitter.com/yuuki_ans/status/1492483806678290434?s=21

https://twitter.com/yuuki_ans/status/1492450390717071362?s=21

https://twitter.com/yuuki_ans/status/1492450163083780096?s=21

https://twitter.com/yuuki_ans/status/1492430961753030657?s=21


https://twitter.com/yuuki_ans/status/1492370937613717505?s=21

 

[nicalandia]

SPR is in going to land when Genoa is laying waste and its not going to fair any different than the rest of Xeon

 

[igor_kavinski]

Only 270W? Come on, Intel! You can do 500W.

 

[DrMrLordX]

SPR is in going to land when Genoa is laying waste and its not going to fair any different than the rest of Xeon

That may be true, but those scores are from an ES that apparently isn't performing up to expectations (for a final product). Surely Intel has some QS available by now?

 

[nicalandia]

That may be true, but those scores are from an ES that apparently isn't performing up to expectations (for a final product). Surely Intel has some QS available by now?

Not performing as expected? Boosting to 3.3 Ghz the CB R23 ST numbers match for Golden Cove IPC. These will not boost to 5 Ghz and neither Epyc cpus. MT CB R23 numbers match with Golden Cove IPC performance plus the known scalability of CB R23




If AMD ever releases ThreadRipper Chagall, I don't see how a 7990WX would not be able to beat Sapphire Rapids, The current Zen2 based 3990WX is already matching the CBR20 MT scores posted and in the clear lead when OC

 

[DrMrLordX]

Not performing as expected?

Read the replies after the tweets, that's what they were saying. Also the turbo max score you're citing was likely achieved during the ST tests that Sapphire Rapids actually won. Actual clocks for the chips were (according to the Twitter commentators) somewhere in the ballpark of 2.1-2.2 GHz during the MT portion of the workload.

Honestly, just wait for the final product to get benched by someone like ServeTheHome or whoever actually gets around to it. It may take awhile for anyone to get release silicon and be able to bench it without violating an NDA or just making Intel hot under the collar.

 

[nicalandia]

Read the replies after the tweets, that's what they were saying. Also the turbo max score you're citing was likely achieved during the ST tests that Sapphire Rapids actually won. Actual clocks for the chips were (according to the Twitter commentators) somewhere in the ballpark of 2.1-2.2 GHz during the MT portion of the workload.

Honestly, just wait for the final product to get benched by someone like ServeTheHome or whoever actually gets around to it. It may take awhile for anyone to get release silicon and be able to bench it without violating an NDA or just making Intel hot under the collar.

Will any of that change the bleak outcome of SPR landing so late to the party?

I know it's just CBR20 and different segments but Zen2 based TR is already matching and beating the MT performance

 

[eek2121]

Apparently Intel is charging extra for AVX512 and AMX. Zen 4 will likely have support for free. What is Intel thinking?

 

[gdansk]

Apparently Intel is charging extra for AVX512 and AMX. Zen 4 will likely have support for free. What is Intel thinking?

They still have some of the mindset of a captive market. And no doubt there are many companies that will do nothing to fix their delusions...