Framework Desktop Power Mode Tuning For Better Performance Or Power Efficiency

Written by Michael Larabel in Computers on 27 August 2025 at 11:00 AM EDT. Page 6 of 7. 4 Comments.
Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Prompt Processing 1024. Performance was the fastest.
Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Prompt Processing 1024. Performance was the fastest.
Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Prompt Processing 1024. Performance was the fastest.

The power saver mode did enjoy a huge lead in delivering the best performance-per-Watt for CPU-based inferencing with Llama.cpp for prompt processing.

Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Text Generation 128. Performance was the fastest.
Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Text Generation 128. Performance was the fastest.
Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Text Generation 128. Performance was the fastest.
Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Prompt Processing 1024. Performance was the fastest.
Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Prompt Processing 1024. Performance was the fastest.
Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Prompt Processing 1024. Performance was the fastest.

The Framework Desktop with AMD Strix Halo could lead to some nice power efficient AI inferencing at the edge.

NAMD benchmark with settings of Input: STMV with 1,066,628 Atoms. Performance was the fastest.
NAMD benchmark with settings of Input: STMV with 1,066,628 Atoms. Performance was the fastest.
NAMD benchmark with settings of Input: STMV with 1,066,628 Atoms. Performance was the fastest.

For the NAMD molecular biology software the balanced default yielded the best performance-per-Watt.

BRL-CAD benchmark with settings of VGR Performance Metric. Performance was the fastest.
BRL-CAD benchmark with settings of VGR Performance Metric. Performance was the fastest.
BRL-CAD benchmark with settings of VGR Performance Metric. Performance was the fastest.

Depending upon whether you strive for optimal energy efficiency or the peak performance out of your hardware, the power modes on the Framework Desktop make it easy to adapt.

Related Articles
HP Z4 G6i: A Linux-Friendly Workstation Powered By The Intel Xeon 600 Series
Framework Laptop 16 With GeForce RTX 5070 12GB, One-Piece Keyboard & Haptic Touchpad
Framework Laptop 13 Pro: Aiming To Be One Of The Best Upgradeable Linux Laptops
Snapdragon X Elite Laptop Experience Still Lackluster On Ubuntu 26.04 LTS
BOSGAME VTA-439: A Great, Linux-Friendly Mini PC Powered By AMD Ryzen AI 9 HX 470
Razer Certifying Their First Laptop For Linux: Razer Blade 18 RZ09-0582