January 24, 2026
CPU vs. ROCm vs. RustiCl
CPU: AMD Ryzen 9 745HX 64GB
GPU: AMD RX 9060XT 8GB
Drop your log files here
Supports multiple dt_bench.txt files at
once.
The benchmark compares three different calculation methods on the same hardware basis. The result clearly shows how important the right driver is for OpenCL acceleration in darktable.
The results in detail:
1st place: ROCm (AMD proprietary/official)
- Time: 2.763 seconds
- Analysis: Using the official ROCm interface gets the most out of the graphics card. In this scenario, it is approximately 2.5 times faster than the open-source driver RustiCL and almost 4 times faster than pure CPU calculation.
2nd place: RustiCL (open source / Mesa)
- Time: 7.089 seconds
- Analysis: RustiCL (the modern OpenCL implementation in the Mesa driver) works and is faster than the CPU, but loses ground significantly to ROCm. It appears that RustiCL is not yet able to utilize the hardware resources of this specific card as efficiently as the specialized ROCm stack.
3rd place: CPU only (Ryzen 9 7945HX)
- Time: 10.624 seconds
- Analysis: Although the Ryzen 9 7945HX is a very powerful high-end laptop CPU, it is no match for GPU acceleration. This illustrates that a dedicated graphics card is almost indispensable for image processing in Darktable.
