Power Modes & Performance
Warning
This article is a work in progress!
Intro
This guide is supposed to show how the performance of the APU changes on various power levels and to help you pick the optimal power mode for your Strix Halo system.
All tests were run at fixed fans speed of 80% to make sure that cooling level is constant, CPU temp never exceeded 75 °C. For GPU becnhmarks please note that the GPU was in a passthrough mode to the Windows VM and only 12 CPU cores were passed as well, so there is at least 5% of a performance drop. In general, look at the percentage changes.
Everything single-threaded is omitted here, since a single core could never use the full power limit even if it's 55W.
CPU - PassMark
Details:
- PassMark PerformanceTest for Linux, version 11.0.1002
- Average of 5 medium tests for CPU, to lower the influence of the initial power boost (
./pt_linux_x64 -i 5 -d 2)
| Metric | 55W | 85W | 120W |
|---|---|---|---|
| CPU Mark | 49635 | 58042+16.9% | 62805+8.2% |
| Integer Math (MOps/s) | 193409 | 222270+14.9% | 233483+5.0% |
| Floating Point Math (MOps/s) | 107679 | 127114+18.0% | 138255+8.8% |
| Find Prime Numbers (MPrimes/s) | 263 | 286+8.7% | 302+5.6% |
| Random String Sorting (kStrings/s) | 71735 | 88253+23.0% | 98842+12.0% |
| Data Encryption (MB/s) | 33582 | 40310+20.0% | 44246+9.8% |
| Data Compression (KB/s) | 539857 | 683287+26.6% | 777048+13.7% |
| Physics (Frames/s) | 4802 | 5477+14.0% | 5806+6.0% |
| Extended Instructions (MMatrices/s) | 38464 | 49404+28.4% | 58761+18.9% |
CPU - Geekbench
Notice
Geekbench does NOT test the system under full load, a lot of the tests don't even utilize all of the available cores, consider these results to be a reflection of an average daily lite usage with no extreme load
Details:
- Geekbench for Linux, version 6.4.0
- 1 test on every power mode with a 3 minute delay
| Metric | 55W | 85W | 120W |
|---|---|---|---|
| Multi-Core Score | 17328 | 17544+1.2% | 17516-0.2% |
| File Compression | 11184 | 12063+7.9% | 12197+1.1% |
| Navigation | 13578 | 13050-3.9% | 13322+2.1% |
| HTML5 Browser | 16030 | 17040+6.3% | 15610-8.4% |
| PDF Renderer | 25044 | 24893-0.6% | 26726+7.4% |
| Photo Library | 25989 | 27260+4.9% | 27099-0.6% |
| Clang | 42312 | 44101+4.2% | 44315+0.5% |
| Text Processing | 3723 | 3714-0.2% | 3667-1.3% |
| Asset Compression | 30848 | 31534+2.2% | 32421+2.8% |
| Object Detection | 13032 | 13075+0.3% | 12545-4.1% |
| Background Blur | 12406 | 12308-0.8% | 12207-0.8% |
| Horizon Detection | 20156 | 19947-1.0% | 19792-0.8% |
| Object Remover | 18049 | 18382+1.8% | 18796+2.3% |
| HDR | 13929 | 13775-1.1% | 13791+0.1% |
| Photo Filter | 14335 | 13066-8.9% | 13849+6.0% |
| Ray Tracer | 38300 | 40547+5.9% | 38962-3.9% |
| Structure from Motion | 22188 | 22210+0.1% | 21352-3.9% |
GPU - Various Benchmarks
Details:
- UNIGINE Superposition was run with the DirectX 1080p Extreme preset.
- 3DMARK Steel Nomad Light was run in Vulkan.
| Benchmark | 55W | 85W | 120W |
|---|---|---|---|
| PassMark 2D Mark | 621 | 706+13.7% | 758+7.4% |
| PassMark DirectX 10 | 59 | 65+10.2% | 78+20.0% |
| PassMark DirectX 11 | 112 | 134+19.6% | 148+10.4% |
| PassMark DirectX 12 | 63 | 67+6.3% | 80+19.4% |
| PassMark GPU Compute | 6229 | 6777+8.8% | 7311+7.9% |
| UNIGINE Superposition | 5247 | 6103+16.3% | 6522+6.9% |
| 3DMARK Steel Nomad Light | 8494 | 9883+16.4% | 10608+7.3% |
| 3DMARK Time Spy | 8891 | 10230+15.1% | 10710+4.7% |
GPU - Gaming
WIP
Conslusions
For CPU we have a classic situation of diminishing returns, where 85W mode seems to be the sweet spot. Going higher doesn't make much sense unless you want to squeeze the absolute maximum out of your system, and going lower is only suitable if you don't use your system resources fully (especially when you don't utilize all of the cores) or don't care about performance loss.
For GPU the situation is a bit different, seems that it can utilise power more effectively, but more testing is needed, I'm on it.