Question Alder Lake - Official Thread

Page 78 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Makaveli

Diamond Member
Feb 8, 2002
4,715
1,049
136
It certainly make sense, or the unexpected wonders of CPU upgrades. :mask:


Yup its like there isn't thousands of people out there on AM4 that are still using pre zen 3 cpu's. As time goes on the numbers will change but we are not there yet.
 
  • Like
Reactions: Elfear

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
AMD only stands a chance in battery life with RMB, on pure performance Intel holds the lead.

I think they'll take MT CPU, iGPU, and battery life lead on the 15W sector, and 28W sector up to maybe 4+8. AMD currently has a massive MT leadership on the subportable parts, something Intel will need to overcome first.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,708
3,554
136
I think they'll take MT CPU, iGPU, and battery life lead on the 15W sector, and 28W sector up to maybe 4+8. AMD currently has a massive MT leadership on the subportable parts, something Intel will need to overcome first.
My comment had more to do with CPU performance in the 45 W -H segment.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
My comment had more to do with CPU performance in the 45 W -H segment.

Yup. I thought I would add to it. They aren't so bad on the -H. Perhaps Alderlake will make it actually competitive on perf/watt, not just performance.

The on-package PCH should in theory help with battery life over Tigerlake-H. Again in theory.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,572
146
Yup. I thought I would add to it. They aren't so bad on the -H. Perhaps Alderlake will make it actually competitive on perf/watt, not just performance.

The on-package PCH should in theory help with battery life over Tigerlake-H. Again in theory.
Well you can see the improvement in battery life in the review. It's not too far behind last year's 5900HX model of the same laptop, which is a decent improvement over Tiger Lake-H.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,572
146
My comment had more to do with CPU performance in the 45 W -H segment.
Good lol, because these performance numbers makes me more certain than ever that there'll be a crossover point between Rembrandt and 6+8 in that 28-45W region when it comes to MT performance.
 

insertcarehere

Senior member
Jan 17, 2013
639
607
136
I think they'll take MT CPU, iGPU, and battery life lead on the 15W sector, and 28W sector up to maybe 4+8. AMD currently has a massive MT leadership on the subportable parts, something Intel will need to overcome first.
That previous massive MT leadership is partially based on TGL-U being a 4C/8T part vs the 8C/16T of the -Hs, while AMD used the same 8C/16T dies from ultraportable right to DTR designs. That doesn't apply this round as both Intel and AMD will be using the same respective sets of parts (14C/20T or 8C/16Ts) in at least some of their T&Ls which would be used in gaming/DTR notebooks, so ADL-H's MT prowess should in theory translate well to ADL-P.
 
  • Like
Reactions: controlflow

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,572
146
Some context for that last post: scores are identical to 45W locked 5800H. 5900HX takes a 10% lead. I expect the 6900HX will have a slightly larger lead than that again over the 12700H, and should be similar to 12900HK

b02df2576be82abeab1223f72e649de4.jpg
 

tamz_msc

Diamond Member
Jan 5, 2017
3,708
3,554
136
Some context for that last post: scores are identical to 45W locked 5800H. 5900HX takes a 10% lead. I expect the 6900HX will have a slightly larger lead than that again over the 12700H, and should be similar to 12900HK

b02df2576be82abeab1223f72e649de4.jpg
Most Cezanne-H laptops have a short-duration turbo higher than the TDP, typically 65 W for the 5800H in your most popular models. Since this is R20, not R23 which runs for 10 minutes - how do we know that the power consumed during the run in this graph isn't 65 W?
 
  • Like
Reactions: controlflow

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,572
146
Most Cezanne-H laptops have a short-duration turbo higher than the TDP, typically 65 W for the 5800H in your most popular models. Since this is R20, not R23 which runs for 10 minutes - how do we know that the power consumed during the run in this graph isn't 65 W?

Because HUB lock PL1 and PL2 (+ AMD's boost settings) to 45W for the sake of their comparisons. It's the same reason why their TGL-H numbers look so low compared to to other reviewers - their reasoning is that they test the CPUs, not the laptops, so might as well compare performance at set power levels rather than at stock laptop settings. I absolutely love it, because we get neat power scaling charts like this as a result:

1643021539743.png
 
Last edited:
  • Like
Reactions: Elfear

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
Phoronix has some interesting results with its Linux testing OpenBenchmarking suite.


With latest Linux kernel and recent compiler it seems 12900K has no trouble beating 5950x in *most* of the tests. 10% extra performance.
I would not read into power efficiency results too much as Ryzen system had 96W average usage and Intel 97.4W.
Except of course Intel used internal GPU and AMD had RX6800. So Intel in its "Designed by Marketing morons" SKU delivers 10% extra performance for 20-25% extra system level power.


1643021464823.png

Real competitor for 5950X on Linux seems to be 12700, ~90% of performance for half the price at very similar power levels.
 

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
Most Cezanne-H laptops have a short-duration turbo higher than the TDP, typically 65 W for the 5800H in your most popular models. Since this is R20, not R23 which runs for 10 minutes - how do we know that the power consumed during the run in this graph isn't 65 W?

At 40W it does about 4500 pts.
There s a graph for CB power comsumption in the review below, idle is at 10W and loading at 58W, all measured at the main, this amount to 40W at most at the CPU level.


So power efficency of 6 + 8 ADL is comparable to Cezanne at this level, RMB should exceed 5000 pts at 45W and will be a tougher comparison.
 

insertcarehere

Senior member
Jan 17, 2013
639
607
136
At 40W it does about 4500 pts.
There s a graph for CB power comsumption in the review below, idle is at 10W and loading at 58W, all measured at the main, this amount to 40W at most at the CPU level.


So power efficency of 6 + 8 ADL is comparable to Cezanne at this level, RMB should exceed 5000 pts at 45W and will be a tougher comparison.

Except in the review they explicitly said:
Like Minisforum, Morefine also offers different settings in the BIOS (10-, 15-, 25-, 35-, 45-, and 54 watts) when it comes to the SoC's TDP. The AMD Ryzen 9 5900HX is used with 45 watts by default. However, the SoC can also be run with 54 watts. We performed all benchmarks with the increased TDP to enable a better comparison

And Prime 95 + CB R15 showed the whole device taking ~70w of power in CPU-heavy workloads, not 58w..
 
  • Like
Reactions: controlflow

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
Except in the review they explicitly said:


And Prime 95 + CB R15 showed the whole device taking ~70w of power in CPU-heavy workloads, not 58w..

The whole device is not the CPU, that s measurements at the main with idle power being 10W, so 70W amount to 46W at the CPU level.

Their measurements show that the chip doesnt get to 54W during Cinebench, power measured at the main is about 58W during this bench and amount to barely 40W CPU power.

You have a review at Computerbase of a Lenovo device that has configurable power with a 4400 score at 42W.

 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Phoronix has some interesting results

As always, care should be taken in interpreting any results from Phoronix, since they do not categorize based on ST or MT workloads. You need to know the applications/benches being used in the suite on a case-by-case basis to understand how each CPU fits the workload(s) that you plan to run under Linux.
 

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
Phoronix has some interesting results with its Linux testing OpenBenchmarking suite.


With latest Linux kernel and recent compiler it seems 12900K has no trouble beating 5950x in *most* of the tests. 10% extra performance.
I would not read into power efficiency results too much as Ryzen system had 96W average usage and Intel 97.4W.
Except of course Intel used internal GPU and AMD had RX6800. So Intel in its "Designed by Marketing morons" SKU delivers 10% extra performance for 20-25% extra system level power.


View attachment 56436

Real competitor for 5950X on Linux seems to be 12700, ~90% of performance for half the price at very similar power levels.

That s tests that rely on few threads, dunno what are they and how compiled they were..

On a lot of tests the 5700G perform about the same as a 5950X but once it s optimised for more than 8 cores the difference is staggering and the 12900K left in the dust.
 

insertcarehere

Senior member
Jan 17, 2013
639
607
136
The whole device is not the CPU, that s measurements at the main with idle power being 10W, so 70W amount to 46W at the CPU level.

Their measurements show that the chip doesnt get to 54W during Cinebench, power measured at the main is about 58W during this bench and amount to barely 40W CPU power.

Are we looking at the same review?

Screenshot_20220124-121537.png
They clearly measured the device maxing at 68-70w during a cpu-heavy workload, including Cinebench.
 

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
That s tests that rely on few threads, dunno what are they and how compiled they were..

On a lot of tests the 5700G perform about the same as a 5950X but once it s optimised for more than 8 cores the difference is staggering and the 12900K left in the dust.

ALL information about compile settings etc is in that benchmark and individual tests. If anything Phoronix is good at documenting this stuff.
The wide variety of tests ensures that everyone can look up what matters to them, but overall ADL looks to be incredibly strong on Linux.

Not everything can be "scaled" or even "optimized" for multiple cores. In fact sometimes in real world the opposite happens, due to scheduling complexities.

1643028954892.png


5700G with it's single CCD and 16MB of L3, but on-die IMC beating multi CCD desktop monster.
 
  • Like
Reactions: controlflow

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
A single threaded or so bench, so that s assuming that the system is loaded only with this app, wonder why multicore systems would be necessary.
You can check other benches, there s a boat load that are lowly threaded.

Yeah and over that boatload of mixed variety of workloads on Linux, 12900K is 10% faster than 5950x. While using on average maybe 20% more power, or 2W more if we consider that 5950X does not have IGP.
 
  • Like
Reactions: controlflow