Discussion PES | Assessing Power and Performance Efficiency of x86 CPU architectures

[BorisTheBlade82]

Dear Community,

so this is my first thread here as a long-time lurker - but I felt the desire to share a small hobby-project of mine from the last couple of months with you...

Performance Efficiency Suite - What is it about?
Most Reviewers solely focus on what they consider to be the most important aspect of modern CPUs - the absolute performance. But this is only one side of the equation. Today Power Efficiency is at least as important - or to be more precise: The amount of energy (Wattseconds or Joules) a CPU needs in order to accomplish a given workload. Sadly most Reviewers shy away from the extra mile it needs to assess this aspect. This suite measures the Total Package Power of a CPU while running the Cinebench R23 benchmarks first in single-threaded mode (1 run), then running in multi-threaded mode (for 10 minutes + whatever it takes to finish the last run). The results will be rendered in the provided Results.xlsx Excel file. To combine Efficiency and Performance there is also a score provided called Performance Efficiency Score (how amazingly inspired I am ).

In the meantime I was able to aggregate more than 80 samples from members of the 3DC & CB communities (see below).

How-To

1. Unzip the latest release to wherever you want EXCEPT on your local OneDrive folder.
2. Open Settings.txt and insert your local Cinebench23 Directory.
3. Run PES Start - it will ask for Administrator rights as these are needed for measuring Package Power
4. Wait until the Powershell finishes.
5. Open the Excel file...
6. Allow external connections (to the generated CSV-files with the data)
7. Go to Data -> Refresh all
8. Enjoy and share your results - just take a screenshot of what the Excel renders.
9. If you want to do multiple measurements with different settings just copy the Excel file (inside the root-folder) before running and refreshing the data.

Some explanations about the Suite

Spoiler

• This Suite has been made possible by Michael Möller and his amazing free and open-source Open Hardware Monitor and his .NET Library OpenHardwareMonitorLib.dll - Thanks a lot!!!
  Homepage: https://openhardwaremonitor.org/
  GitHub: https://github.com/openhardwaremonitor
• The results for the Package Power look pretty accurate compared to the sparse data the internet provides. Seems, that the vendors are much more honest with those sensors than they are with temperature etc.
• The suite basically consists some powershell scripts and an Excel file for presentation purposes
  • RunAsAdminWrapper.ps1
    This is needed to have a convenient relative path shortcut in the root folder and request admin-rights at the same time
  • Main.ps1
    • After setting up some stuff it basically starts the Cinebench R23 one at a time. It then checks for the "Cinebench.exe" process being active.
    • While this is true it queries the Package Power Sensor data with a lower bound of 10ms (in order to keep CPU-load of the script at bay).
    • After each run the aquired data gets pushed to CSV files located in the LogCsv subfolder.
  • Results.xslx
    • The Excel file basically just does some import, calculations and a hopefully nice presentation of the data.
    • Histogram
      The bold line shows a running average of the last 100 data-points which should be sufficiently accurate. The pale line shows each single data-point.
    • Calculation of Total Package Consumption
      To get that number we need the integral. That is why we first calculate the timeframe between two data-points and then multiply the measured value.
    • Everything else in that Excel is hopefully more or less self-explaining

Online Resources

• PES GitHub Repo
• 3DCenter.org Forum Thread (German)
• ComputerBase.de Forum Thread (German)

Disclaimer
I am by no means a Powershell professional or a professional Reviewer. I was just sick of the lack of information and wanted to propose a low-effort solution. Any input for further improvement is highly welcomed. Please feel free to use/extend/rip-off this solution as you wish. But please share your findings to the world.

 

[BorisTheBlade82]

Here is the error
View attachment 81609

Well, did you put the whole PES directory into your OneDrive? This was something I already identified early on: Due to some Office365 internal shenanigans, file access works differently (broken) when the folder is in OneDrive as opposed to somewhere else on your storage.

 

[Markfw]

Well, did you put the whole PES directory into your OneDrive? This was something I already identified early on: Due to some Office365 internal shenanigans, file access works differently (broken) when the folder is in OneDrive as opposed to somewhere else on your storage.

on the hard drive. The first time I have EVER used OneDrive was to put the files there for you.

 

[BorisTheBlade82]

So I just updated the rankings and matrices in the opening posts in order to make @Markfw 's leadership official
BTW I also removed the Geekbench 5 rankings, as due to the nature of the non-fixed workload it is rather pointless (just needed a couple more samples to be sure - thanks to everyone contributing).
Have fun!

 

[igor_kavinski]

Thanks, @BorisTheBlade82 !!!

LIbreO works great now!

Sharing the results of our new server:

 

[igor_kavinski]

Dell Latitude 3520 with i5-1135G7

It had four power settings in BIOS so tested all of them. Windows performance plan was set to High Performance but I think the BIOS power mode took precedence.

Turbo



Optimized



Cool



Quiet



Quiet mode really WAS quiet.

 

[mmaenpaa]

AMD Ryzen 7 7700, AM5, 3.8 GHz, 8-Core,Boxed (100-100000592BOX)
Asrock B650M Pro RS (BIOS 1.28)
KINGSTON 32GB 6000MT/s DDR5 CL36 DIMM (KF560C36BBEAK2-32)
All settings at default (except memory EXPO 6000MHz), W11 Pro (latest chipset & OS updates as of today)

 

[mmaenpaa]

Same setup as above but different memory profiles. It would seem that at least with these workloads it could be better to leave memory at JEDEC profile (which requires less voltage).

EXPO 5600 @1.25V (EXPO 6000 uses 1.35V)


AUTO (JEDEC) 4800 @1.1V

 

[moinmoin]

it could be better to leave memory at JEDEC profile (which requires less voltage)

Yeah, using JEDEC frequencies usually should consum less power than OC frequencies. I'd be surprised if efficiency wise there's any memory OC that's recommendable.

 

[JoeRambo]

I'd be surprised if efficiency wise there's any memory OC that's recommendable.

Frankly CB23 does not care about memory subsystem at all, so this power efficiency delta is pretty much the max. Max power for zero gains.
For the rest of workloads, when talking about "sane" setup, faster memory will have very similar power efficiency. Some extremes like compression will probably come out way more efficient with memory OC, due to CPU "race to idle".

There is also a question of how much extra power is burned by memory OC: memory size x voltage is a factor, but also CPU side of things. Esp on AMD SoC voltages burn wattage real quick. While on Intel i have always found minor increases ( or reporting was off ).

 

[mmaenpaa]

LENOVO L15 G4 R7-7730U/15.6FHD/16GB/512SSD/W11P
(one 16GB SODIMM so single channel memroy)

AC:


DC:

 

[mmaenpaa]

DELL PM3581 I7-13700H/15.6FHD/16GB/512SSD/RTXA1000/WWR/11P
For some reason Excel gives divide / 0 error on the results for this laptop (JAKO=DIVIDE).
I believe this the first Intel 13th gen laptop I have tested.

EDIT:
@BorisTheBlade82 (I meant to ping Boris, not Igor )
I can send you the results via email / some other method if you want them?
EDIT2:
I believe openhardwaremonitor.dll can not read power readings (last update was 12/2020)

 

[BorisTheBlade82]

DELL PM3581 I7-13700H/15.6FHD/16GB/512SSD/RTXA1000/WWR/11P
For some reason Excel gives divide / 0 error on the results for this laptop (JAKO=DIVIDE).
I believe this the first Intel 13th gen laptop I have tested.

EDIT:
@BorisTheBlade82 (I meant to ping Boris, not Igor )
I can send you the results via email / some other method if you want them?
EDIT2:
I believe openhardwaremonitor.dll can not read power readings (last update was 12/2020)
View attachment 84965
View attachment 84964

Well, mostly I just need to make OHM aware of a new CPU in the source code. If you could supply me a CPU-Z screenshot, I could give it a try (after my vacation in Japan in around 3 weeks 😉)

 

[ikjadoon]

i7-12700K @ stock w/ PL1 = PL2 = 4096W + 32GB DDR5-6400 @ 32-39-39-102, 2T. Cooled with an Arctic Liquid Freezer 280 in push-pull, but an aggressively quiet fan curve (fans at 40% until the CPU hits 95C, heh).

I'm coming to terms that Intel CPUs are the Samsung Exynos of the desktop world 😂 This is a fantastic tool. I also have a MacBook Air M1 and will see if I can add any further tests (if helpful) once I grab asitop to pull in Powermetrics data.

I think something interfered in the CBR23 1T test, so I'll re-run it again later today.

I only had GB6 installed, so here's the official legacy downloads page for Geekbench 2, 3, 4, and now 5.

 

[ikjadoon]

i7-12700K @ stock w/ PL1 = PL2 = 4096W + 32GB DDR5-6400 @ 32-39-39-102, 2T. Cooled with an Arctic Liquid Freezer 280 in push-pull, but an aggressively quiet fan curve (fans at 40% until the CPU hits 95C, heh).

I think something interfered in the CBR23 1T test, so I'll re-run it again later today.

I shut down background programs, but still had a few small spikes on the 1T test. Which is weird: I imagined the point of a hybrid CPU + "Thread Director" meant the other 7 P-cores and 4 E-cores would allow this one P-core to focus near-exclusively on the CBR23 1T test.

I guess not. Or maybe whatever caused the spikes was already using 19 threads and needed to interrupt this already-saturated P-core thread, lol.

Have no other explanation why suddenly ~40W package power could jump to 50W or 60W for a few seconds. But at least it's smoother than the last one.

 

[ikjadoon]

Just some other data to comparing SPECint2017 1T power consumption, as another reference to Cinebench here:

AMD Zen4 @ 5.275 GHz: avg 8.7W for 10.2 pts
Intel Raptor Lake @ 5.2 GHz: avg 24.1W for 10.4 pts

Zen4 is consuming ~1/3 the power of Intel for 98% of RPL's performance. Perf-per-watt landslide.

 

[JoeRambo]

Zen4 is consuming ~1/3 the power of Intel for 98% of RPL's performance. Perf-per-watt landslide.

You know what is not fun? That marketing morons decided to throw this landslide victory away and in it's place was put the chip we know as 7950X.
It probably averages some 20W and scores 11 points or so. All thanks to marketing geniuses trying to beat Intel while feeding 1.35V+ to a chip that needs ~1V for 5Ghz.

 

[Saylick]

You know what is not fun? That marketing morons decided to throw this landslide victory away and in it's place was put the chip we know as 7950X.
It probably averages some 20W and scores 11 points or so. All thanks to marketing geniuses trying to beat Intel while feeding 1.35V+ to a chip that needs ~1V for 5Ghz.

Ehh, the EPYC guys know what a gem they have on their hands.

 

[Markfw]

You know what is not fun? That marketing morons decided to throw this landslide victory away and in it's place was put the chip we know as 7950X.
It probably averages some 20W and scores 11 points or so. All thanks to marketing geniuses trying to beat Intel while feeding 1.35V+ to a chip that needs ~1V for 5Ghz.

All my 6 7950x's are set to 1v and 80c max temp. It works just fine and beats any Raptorlake at the things I do, so what do I care ? Oh, and uses less power

 

[Abwx]

You know what is not fun? That marketing morons decided to throw this landslide victory away and in it's place was put the chip we know as 7950X.
It probably averages some 20W and scores 11 points or so. All thanks to marketing geniuses trying to beat Intel while feeding 1.35V+ to a chip that needs ~1V for 5GHz.

That work only for INT apparently since he got such a perf/watt with SPECint, that s more or less corelated by Computerbase H264 test using Handbrake with the 7950X consuming 189W and still outperforming the 13900K by 11%, wich mean that it would match it at about 138W or even less.

Obviously AMD made a big push for INT code efficency since that s what matter in servers, that being said 1V may be enough at 5GHz with a single core running INT code, but with all 16 cores loaded internal voltages drops are huge and surely that close to 1.3V must be provided to the socket for the thing being perfectly stable and competitive in FP with loads like CB and other Blender.

 

[H433x0n]

Just some other data to comparing SPECint2017 1T power consumption, as another reference to Cinebench here:

AMD Zen4 @ 5.275 GHz: avg 8.7W for 10.2 pts
Intel Raptor Lake @ 5.2 GHz: avg 24.1W for 10.4 pts

Zen4 is consuming ~1/3 the power of Intel for 98% of RPL's performance. Perf-per-watt landslide.

They’re using CPU package power for RPL and CPU core power (measuring only CCD) for 7950X. So yeah, no wonder it’s drastically different.

A 7950X wouldn’t even boot your PC at 9W. They should’ve used Phoenix as a real apples/apples for ST power usage.

 

[Abwx]

They’re using CPU package power for RPL and CPU core power (measuring only CCD) for 7950X. So yeah, no wonder it’s drastically different.

They use CPU package power for both, stop inventing what suits your beliefs...

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Leistungsaufnahme und Effizienz

Intel Raptor Lake im Test: Leistungsaufnahme und Effizienz / Leistungsaufnahme in Anwendungen ab Werk / Der Verbrauch steigt

www.computerbase.de

See.?..

CPU package power in Handbrake encoding, that s clearly stated on the graph, 188.9W for the 7950X and 248.9W for the 13900K.

And the scores in Handbrake :

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Benchmarks in Apps

Intel Raptor Lake im Test: Benchmarks in Apps / So wurde getestet / Leistung in Single-Core-Lasten (klassisch)

www.computerbase.de

7950X is 13% faster actually than the 13900K, i said 11%, at least you were usefull to help correct the thing, it s even worse for the 13900K than what i stated in the first place, the 7950X@142W is still 9% faster, it would match it at 115W.

 

[H433x0n]

They use CPU package power for both, stop inventing what suits your beliefs...

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Leistungsaufnahme und Effizienz

Intel Raptor Lake im Test: Leistungsaufnahme und Effizienz / Leistungsaufnahme in Anwendungen ab Werk / Der Verbrauch steigt

www.computerbase.de

See.?..

CPU package power in Handbrake encoding, that s clearly stated on the graph, 188.9W for the 7950X and 248.9W for the 13900K.

And the scores in Handbrake :

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Benchmarks in Apps

Intel Raptor Lake im Test: Benchmarks in Apps / So wurde getestet / Leistung in Single-Core-Lasten (klassisch)

www.computerbase.de

7950X is 13% faster actually than the 13900K, i said 11%, at least you were usefull to help correct the thing, it s even worse for the 13900K than what i stated in the first place, the 7950X@142W is still 9% faster, it would match it at 115W.

So you’re telling me that a 7950X pulls 9-13W maxing out a single core workload? The full package power is higher than that just sitting idle in windows.

Look at the source you linked. The 7950X is pulling 21W at idle and 53W under a ST workload (32W delta). You’re saying in Specint 1T the entire CPU pulls 13W?

 

[JoeRambo]

Look at the source you linked. The 7950X is pulling 21W at idle and 53W under a ST workload (32W delta). You’re saying in Specint 1T the entire CPU pulls 13W?

I think they were measuring delta between idle and load with SPEC ? Not quite sure and anyway 5.2Ghz is not the point to select to do the testing. Without full disclosure there are way too many gotchas that can happen with both CPUs:

For example on Intel:

1) There is huge variance between individual CPU and their builtin V/F curves, talking about 0.1V level of difference on unlucky vs lucky sample
2) Said V/F curve has points at specific frequencies ( like 4.9, 5.1 max turbo ) and the voltage for your selected frequency is interpolated. Esp fun combined with (1) where chip needs ridiculous voltage for turbo, but you test at 100mhz lower speed and chip is still fed ridiculous voltage.
3) Uncore speeds are impacted by voltage selected for static core clock, it might not be enough
4) In reverse of (3), CPU core voltage might be impacted by uncore speeds being default. Say CPU core asks for 1.2V VID, but uncore at 4.XGHz defaults is asking 1.3V => that's what core will be fed as max of ( P cores / e cores / Uncore ) is selected and fed everywhere.
( 3, 4 are BIOS setting dependant)


Not an easy task to isolate core power usage at speeds above ~4.X ghz, i'd not trust random guy with SPEC on internet to do a good enough job.

The way i'd do it is ~4.5Ghz, each chip with voltage manually found to be yCruncher stable, Uncore also fixed 4.5Ghz for Intel and then we're talking about estimation that makes sense.

Look at the source you linked. The 7950X is pulling 21W at idle and 53W under a ST workload (32W delta).

That's about right, ~30w for ST ( again, not core power, but delta between idle and running heavyish load on 1 thread, that lights up caches, fabric, IMC) with stock marketing moron settings and clocks. Both vendors have horrible efficiency with low loads, except AMD also has real bad idle power consumption.
7800X3D is the chip to buy if you don't want to tune anything and just enjoy it and Intel has some great chips and deals if you don't mind tinkering and manually undervolting and underclocking.
For running large MT loads all the time, the way @Markfw or I do ( taking manual control of clocks and voltages and temps ) is the path to very efficient compute Intel can't touch.

Can we now get back to PES testing?


P.S. I miss the forum greats like TheStilt, @Idontcare and others that were beyond just linking some source on internet and capable of running experiments on their own and presenting findings in a way that roasted all resident faction warriors properly.

 

[Abwx]

So you’re telling me that a 7950X pulls 9-13W maxing out a single core workload? The full package power is higher than that just sitting idle in windows.

Look at the source you linked. The 7950X is pulling 21W at idle and 53W under a ST workload (32W delta). You’re saying in Specint 1T the entire CPU pulls 13W?

Im talking of Handbrake here, but still, the number related by the member in SPECint is credible.

The 21W at idle is with overclocked RAM, at stock setting the idle power is 16.3W, look better, so the cores themselves pull 189 - 16 = 173W.

That s barely 11W per core, frequency is in the vicinity of 5GHz since it s an all core loading, so 13W for a single core at 5.2GHz in SPECint is in line with Computerbase numbers in Handbrake.

In Cinebench ST the core pull 36W, but that s at 5.7GHz and we all know that the V/F curve is extremely steep in the last drops of frequency, we can see that from 142W to 205W the frequency, and hence score, increase only by 7% in their CB test.

 

[BorisTheBlade82]

Hey @TESKATLIPOKA ,

as you are the very first person I know of owning a Phoenix based Notebook, maybe you might want to do me a little favour:

Would you like to run Performance Efficiency Suite? Everything you need to know can be found in the Opening Post.

If you don't have MS Excel, you can just send me the CSV files that get created in the sub folder.

If your Notebook provides different Power Presets, it would be nice if you could give all of them a try - just as your time permits.

Thanks and Merry Christmas!