Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

The term for describing is memory hard functions. RandomX[0] is one such example where GPU parallelism does not net them a large advantage over CPUs.

[0]: https://github.com/tevador/RandomX



Thinking about it, this could be the way forwards. Memory offers several natural bottlenecks:

1. Memory size - Memory is somewhat costly (even now), with most entry laptops being stuck in the range of 8GB.

2. Access bandwidth - Getting a CPU to communicate with the RAM takes some time, improvements are incremental and are fundamentally limited.

3. Thread access - Threads compete for bandwidth, as long as cache hits are low.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: