In this update we stripped down more processing from the ethash algorithm and moved more buffers into GPU only memory. This should give performance improvements to non-M1 Macs, however very little will be noticed on Apple Silicon due to its use of unified memory.
Also in this update, the low power levels have been tweaked down to perform to the same degree as the initial OpenCL version to address the M1 Mac overwhelming issue previously reported.
Please let me know if this indeed does help the stability of M1 Macs, or if not in the meantime you can still download the OpenCL version from the link in the About section.
To learn more about the differences between Private and Shared Memory or lack there of in the M1 world, see Metal Best Practices Guide.