I tested gigawatts implementation.
14.33 collisionspermin instead of 180 collisionspermin stock client, 32 threads.
The point is that gigawatts algorithm could yield a GPU implementation which could be a 100x faster version of a slower algorithm, with a net gain.
But I will try to change the memory requirements and see what happens on the CPU if we bring it down to using very little memory.
What were you using for CHAIN_LENGTH?
I've been running a test with CHAIN_LENGTH = 2 on a 1 CPU/1GB test box overnight and got this:
http://i.imgur.com/wknTeLb.pngIf I try to run the default implementation, it crashes because it requires too much memory. However, on a 2 CPU/2 GB test box running only one thread, I get around 2.2 hpm. However, I need to rerun these tests because the hashperminute calculation
was updated with a recent commit.
I've also updated the source code:
http://pastebin.com/VTCMFQXgIt now uses the thread interruption from the recent commit and makes debug messages optional.
To compile: Pull the source from the main github branch, then take my code and overwrite momentum.cpp with it. Compile as normal.