Open dmarcuse opened 8 years ago
Not completely fixed as of 452ca4a, but mitigated.
It seems to me that the miner kernel isn't clearing it's solution cache after finding a valid solution. When it rolls over to the next block, it for some reason submits all the solutions it was previously working on. This varies greatly, but seems to get worse the longer an easy block is worked on. Example: This block was worked on for about 5 minutes by my GPU at work level 98711. It spams a bunch of rejected solutions before eventually continuing on as normal.
I don't know much about how it's coded (nor do I really understand enough to read it), but maybe memory needs to be cleaned up when a valid solution is found before continuing to the next block. I'm not even sure this is relevant to the original issue here, and if it's not I'll start a different thread.
Seeing as this bug has persisted from the original version of the Java portion of the code, I'd imagine it's a bug with the OpenCL mining kernel.