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.

 

[Timur Born]

It's noteworthy that Geekbench seems quite memory bandwidth sensitive. So all those people overclocking their memory for higher MTs will get better scores even when their CPU package power doesn't increase.

We also cannot be entirely sure that everyone posting results has their CPU DC loadline properly set. MSI Z790 boards seem to use 80 / 0.800 mOhm even when their stock LLC seems to be 0.110 mOhm (based on what a user posted). My own Gigabyte board uses 0.900 mOhm for both stock DC LL and LLC (Intel's suggested 0.110 mOhm not even offered), but since I changed the LLC I had to measure matching DC LL values by trial & error. Had I not done the last step then my power measurements posted here would be wrong.

 

[Timur Born]

I wonder why my measurement needed 24 runs of CB23 to finish while others used considerably less?

 

[BorisTheBlade82]

I wonder why my measurement needed 24 runs of CB23 to finish while others used considerably less?

Hehe, the MT run is designed so that the loop needs to last at least 10 minutes. If a run has finished and 10 minutes haven't been reached, then the next one gets started.

So basically: The more runs, the faster your machine 😉

 

[mmaenpaa]

LENOVO TP T14s G3 i7-1260P 16GB/512
W11P

AC:


DC:

 

[Timur Born]

I posted corrected measurements, but then it turned out that the specific combination of 8x P-core P95 Small + 4x E-core P95 Small (3rd cluster) isn't stable yet. So I come back later for new results as I had to change my former undervoltage to accommodate for the super crazy spiky transients of some Folding@home work units (220 W max measures, but still triggering a 253 W power limit).

 

[Timur Born]

CPU: 13900K !undervolted! + 1x 5.9 GHz single-core OC + 253W power limit | Motherboard: Gigabyte Z790 Aero G | GPU: Intel iGPU | RAM: DDR5 Patriot 6000-36 at 5600 CR1T | Hard Drive: Adata SX8200 Pro

View attachment 77035

View attachment 77036

Here are corrected measurements for my 13900K + 5600 CR1T. While the last results were stable for stuff like Prime95 at the power limit and everything else they did not proof stable for Folding@home (FAH).

Many lower voltage settings can produce single WHEA within multiple hours of testing for specific work units, at least in combination with other background load (like videos, opening/closing applications). FAH is very special in that some work units may measure a maximum CPU Package power of only 220 W, but still trigger the power limit at 253 W, which are some crazy transients (regardless of LLC settings). I suspect that many overclocks and undervolts out there would fail FAH in that regard.

Turbo Ratio Limits - AVX2, Resolved: 59x (1c), 58x (2c), 56x (3-4c), 55x (5-8c)
Turbo Ratio Limits (E-cores): 43x (1-16c)
Maximum Per-core Ratio Limits (Current): 55, 55, 59, 58, 55, 55, 56, 56, 43, 43, 43, 43, 43, 43, 43, 43, 44, 44, 44, 44, 44, 44, 44, 44



Geekbench is sensitive to memory bandwidth, so my 5600 MT are at a disadvantage.

 

[Timur Born]

I will try to re-enable "Voltage reduction initiated TVB" to reduce voltages at lower temperatures. This should increase efficiency of CB23 single-core and maybe GB5.

Unfortunately I cannot use this to reduce CB23 all-core power, because it operates in the same temperature window as Folding@home (for which higher voltages are needed).

 

[Timur Born]

After enabling "Voltage reduction initiated TVB", increasing voltages by 20 mV to match the old offset at 79°C: Anything below 79°C would use less voltage than before and anything above 79°C would use more voltage. Running the test at 100% fan-speed (1500-1600 rpm) I get better CB23 single-core efficiency (expected), same GB5 and worse CB23 multi-core efficiency (also expected).

The reason for the worse CB23 multi-core efficiency is that CB23 mostly runs above 79°C, so voltages increase vs. my non TVB settings. And the reason why I used 79°C as the reference point is that this is what Folding@home runs at and I need to keep voltage the same on that one. I may try to lower it a bit more, but generally TVB might be the better choice, because almost all daily stuff runs at lower temperatures (including games).

 

[mmaenpaa]

LENOVO P15V G3 R7-6850H/15.6FHD/32GB/1TB/RTX-A2000/11P10P/3P
(nice laptop, even during CBR23 MT test fan profile is not annoyingly loud, and ST loads are practically silent)
AC:


DC:

 

[BorisTheBlade82]

LENOVO P15V G3 R7-6850H/15.6FHD/32GB/1TB/RTX-A2000/11P10P/3P
(nice laptop, even during CBR23 MT test fan profile is not annoyingly loud, and ST loads are practically silent)
AC:
View attachment 80502
View attachment 80503
DC:
View attachment 80504
View attachment 80505

Nice, on battery it seems to turn into a U SKU. Not that bad of a decision by the OEM. Surely there is a way to influence this as a user, isn't it?

 

[moinmoin]

Nice, on battery it seems to turn into a U SKU. Not that bad of a decision by the OEM. Surely there is a way to influence this as a user, isn't it?

At least RyzenAdj allows that, just use the --power-saving and --max-performance switches to force the other mode respectively. (And since there's a Windows build of that you Windows users can give it a try as well.)

 

[BorisTheBlade82]

At least RyzenAdj allows that, just use the --power-saving and --max-performance switches to force the other mode respectively. (And since there's a Windows build of that you Windows users can give it a try as well.)

Yes, I already used that with my Renoir in the past. As per my knowledge Lenovo has their Vantage application, which should bring some presets as well and might be more attractive to casuals than RyzenAdj. To me it was more about Lenovo's decision regarding the default behaviour.

 

[mmaenpaa]

HP EB 655 G9 R7 5825U 15.6i 16GB 512GB
AC:


DC:

 

[Timur Born]

**Efficiency per single core (1 P core = 3 E cores in on-die space) on undervolted 13900K**

CB23 (Floating-Point), single-core:
1x58 P 2299 33.3 W
1x45 P 1783 16.0 W (=22.5% less performance at less than half the power usage)
1x45 E 1258 17.7 W (best cluster)
1x43 E 1213 17.4 W (normal cluster)

7-Zip (Integer), single-core:
1x58 P 10196 28.3 W
1x45 P 8296 13.3 W (=18.5% less performance at less than half the power usage)
1x45 E 7617 16.9 W (best cluster)
1x43 E 7342 16.5 W (normal cluster)

 

[Timur Born]

CB23 (Floating-Point), single-core:
1x43 E 1213 17.4 W
1x33 E 913 4.3 W (=24.7% less performance at a quarter the power usage)

7-Zip (Integer), single-core:
1x43 E 7342 16.5 W
1x33 E 5756 4.3 W (=21.6% less performance at close to a quarter the power usage)

BUT: the CPU Package Power reading at this low load may be be rather incorrect. The Corsair PSU measures a difference of only 6 W instead of 13 W between both CB23 runs.

 

[Markfw]

OK, I am trying to run this on a EPYC Genoa 9554. When I clcik on the ICON and run as administrator, some window comes up so fast I can not see, and then nothing. CB 23 is installed and Geekbench 5. I don't have excel loaded yet. Could that be the problem ?

 

[BorisTheBlade82]

OK, I am trying to run this on a EPYC Genoa 9554. When I clcik on the ICON and run as administrator, some window comes up so fast I can not see, and then nothing. CB 23 is installed and Geekbench 5. I don't have excel loaded yet. Could that be the problem ?

Well no. You need to edit the Settings.txt first, where you need to provide your CB & GB folders.
Excitedly awaiting the best MT result ever 😉

Oh, and afterwards just attach the CSV files generated in the sub-folder here, so I can provide you the charts - the Libre Office sheet sadly is still a bit buggy (at least on my machine).

 

[Markfw]

Well no. You need to edit the Settings.txt first, where you need to provide your CB & GB folders.
Excitedly awaiting the best MT result ever 😉

Oh, and afterwards just attach the CSV files generated in the sub-folder here, so I can provide you the charts - the Libre Office sheet sadly is still a bit buggy (at least on my machine).

I did that....(editing the settings.txt file)

Edit, but the reason I am doing this ? Dual 9554's hold the world record for CB23. Even a single should be awesome !

 

[Markfw]

Well no. You need to edit the Settings.txt first, where you need to provide your CB & GB folders.
Excitedly awaiting the best MT result ever 😉

Oh, and afterwards just attach the CSV files generated in the sub-folder here, so I can provide you the charts - the Libre Office sheet sadly is still a bit buggy (at least on my machine).

The problem was I put quotes around the Geekbench directory name. Once I removed those it worked. But when try to "refresh all" I get errors, even if I run excel as administrator.

Here are the 3 csv files that I think it uses.

Edit: it won't let me attach them.

 

[Markfw]

If I run CB23 by itsself, I get 60,054

 

[BorisTheBlade82]

The problem was I put quotes around the Geekbench directory name. Once I removed those it worked. But when try to "refresh all" I get errors, even if I run excel as administrator.

Here are the 3 csv files that I think it uses.

Edit: it won't let me attach them.

Just give it one of the allowed extensions, hopefully it will not check metadata 😉
But usually it just works with Excel.

 

[Markfw]

Just give it one of the allowed extensions, hopefully it will not check metadata 😉
But usually it just works with Excel.

 

[Timur Born]

CB23 (Floating-Point), multi-core, CPU / PSU Out / PSU In:

55x/43x 40551 253 W 348 W 377 W
45x/33x 32257 121 W 184 W 207 W (=20.5% less performance at close to half the power usage)

Undervolted 13900K at 8x 4.5 GHz HT P cores + 16x 3.3 GHz E cores:

 

[BorisTheBlade82]

@Markfw
For me it is absolutely unclear, why the Excel results renderer did not work for you. The CSV files are absolutely fine.
Anyway! Here are the results for your AMD EPYC Genoa 9554 64c/128t - as was to be expected, the CB23 ST and GB5 results are nothing groundbreakting because of the rather high idle offset, while your system achieved the single highest CB23 MT PES-score of all time.
Admittedly, I would have expected a bigger margin to the first undervolted Consumer SKUs - pretty sure, your 9654 would make another quite substantial leap.



I will do my best to put this and also all the other results posted into the rankings and matrices ASAP - gratz to the new Leader of the Pack

 

[Markfw]

Here is the error