Thread: Faster GPU-ECM with CGBN View Single Post
 2021-08-25, 15:42 #4 chris2be8     Sep 2009 2,333 Posts I'm trying to build it on Linux, but have not had much luck. I copied the process I used to build the previous version: Code: Download gmp-ecm-gpu_integration.zip unzip gmp-ecm-gpu_integration.zip cd gmp-ecm-gpu_integration autoreconf -si ./configure --enable-gpu=30 # The previous version needed --enable-gpu=sm30. The following messages look relevant: -snip- checking that CUDA Toolkit version and runtime version are the same... (9.1/9.1) yes -snip- configure: with_cgbn: , , " -snip- make # This fails with messages about CGBN being missing. What version of CUDA Toolkit and runtime is needed to support CGBN? And where is cgbn.h on your system? Also what makes of GPU support it? Mine may be too old. Chris PS I use ECM set to do 512 bit arithmetic for numbers below 2^506. This is about 3 times faster than the 1024 bit version. So you may be *only* getting a factor of 3 over all ranges. But that's still very nice. Last fiddled with by chris2be8 on 2021-08-25 at 15:43