It's a good question, but I think that the problem is that the overhead of switching between threads is much greater than the number of cycles wasted by waiting

It's a good question, but I think that the problem is that the overhead of switching between threads is much greater than the number of cycles wasted by waiting for the memory request. I think the fastest context switches on Linux are around 1 µs (1/1000000 of a second, or about 4,000 cycles)[1], whereas a request from RAM is around 25 ns (1/40 of a µs, or about 100 cycles)[2]. It's possible my numbers are off by a bit, but I think this means that on a standard CPU, you really need to parallelize the search more explicitly rather than letting the OS handle it for you. GPU's on the other hand do take a similar "brute force" approach to context switching, and clearly have good results on some similar problems. [1] https://eli.thegreenplace.net/2018/measuring-context-switching-and-memory-overheads-for-linux-threads/ https://eli.thegreenplace.net/2018/measuring-context-switchi... [2] https://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/7 https://www.anandtech.com/show/9483/intel-skylake-review-670...