SRBase has been a vital project for mathematical research, and its commitment to supporting NVIDIA, AMD, and Intel GPUs has been a great example of heterogeneous computing.

However, in the modern 2026 landscape, the continued lack of Apple Silicon iGPU support is a massive missed opportunity for high-efficiency distributed research.

For years, the standard reason for not supporting the M-series GPU was the lack of native FP64 or the deprecation of OpenCL on macOS.

But we now have definitive proof that these are no longer roadblocks.

Projects like Einstein@Home and PrimeGrid have successfully deployed Apple Silicon iGPU applications that handle high-precision scientific math with incredible performance.

On my base M4 Mac Mini, I am currently running 10 concurrent tasks on the iGPU for other projects with 100% stability.

Apple Silicon’s unified memory architecture is actually a superior solution for many math-heavy tasks, as it eliminates the PCIe bandwidth bottleneck that plagues traditional discrete cards.

By not offering an arm64-apple-darwin GPU plan, SRBase is leaving thousands of the world's most power-efficient compute nodes untapped.

If we can support Intel Arc, we can support Apple Silicon.

It is time to modernize the codebase and allow Mac users to contribute their iGPU power to the search for Sierpinski/Riesel bases.

When can we expect to see an experimental test unit for the M-series?

Thank you for the candid response, rebirther. I appreciate the clarification on which sub-projects are viable for GPU acceleration.

I am very interested in testing the trial factoring (TF) applications on the M4 iGPU. Even if the Sierpinski/Riesel base searches remain CPU-bound, Trial Factoring is a perfect candidate for the high-bandwidth unified memory architecture of Apple Silicon.

I would be happy to attempt running the standalone mfakto app as a proof of concept. If the performance and stability are there, perhaps we can look into what it would take to build that "mac wrapper" to integrate it officially into SRBase.

Do you have a preferred set of test parameters for a standalone mfakto run on ARM64, or should I just report back with raw completion times on a specific workunit?