On Sun, Nov 14, 2010 at 01:53:42AM +0100, Georg Hofstetter wrote: > > Hi Sylvain, > > > > > But is it fast enough ? > > > > I had a shot at writing a bitslincing impl for NVidia using opencl for > > my laptop card. (9600M GT). But it turned out that a bitslicing 128 > > bits impl using SSE x86_64 would be just as fast ... > > > > Not sure. I tried different setups. > I used my defect HD 4850 which is causing to crash the computer on high > loads and used that for testing A5Ati on win32/x64. Most of the results > with that 4850 were found faster than the A5Cpu. > But there were also searches that took longer using A5Ati than A5Cpu. > So I am not sure if comparing just a few numbers cracked with Kraken is > generally a good idea.
the lookup process is not a good benchmark for GPU processing speed. Either the GPU was idle most of the time or it did compute many more chains that filled up the queue at the bottleneck which is the harddisk in your setup and those extra computations went unnoticed, because they got thrown away after the result was found. Table generation would be a good benchmark. The HD3870 should still outperform your CPU by several 100% though, so it may be a good idea to re-add support for it. _______________________________________________ A51 mailing list A51@lists.reflextor.com http://lists.lists.reflextor.com/cgi-bin/mailman/listinfo/a51
