Date: Sat, 14 Apr 2012 17:37:43 +0200 From: magnum <john.magnum@...hmail.com> To: "john-dev@...ts.openwall.com" <john-dev@...ts.openwall.com> Subject: Re: Xsha512 OpenCL Code Great! If you can come up with some kind of automatic adjustment to device it would be even better. IMHO we should make all OpenCL formats run fairly well on CPU. As seen on this list some run systems with 48 or more cores... magnum On 14 apr 2012, at 17:13, myrice <qqlddg@...il.com> wrote: > Hi, all > > I just implement XSHA512 on OpenCL. It is already in myrice-JtR of my github(https://github.com/qqldd/myrice-JtR). > Here are benchmark on GTX580: > > Benchmarking: Mac OS X 10.7+ salted SHA-512 [OpenCL]... DONE > Many salts: 52937K c/s real, 49569K c/s virtual > Only one salt: 24197K c/s real, 22469K c/s virtual > > And my G9600MGS: > Benchmarking: Mac OS X 10.7+ salted SHA-512 [OpenCL]... DONE > Many salts: 1395K c/s real, 1381K c/s virtual > Only one salt: 1330K c/s real, 1317K c/s virtual > > In G9600MGS, it performs better than cuda code. Amazing! > > At this moment, I haven't implemented cmp_all on GPU. Hopefully add it in next few days. > > Besides, though there is find_best_workgroup in opencl_xsha512_fmt.c, I don't use it. Since the max_group_size is too large(512 for G9600MGS, 1024 for GTX580). There is always a CL_OUT_OF_RESOURCES error. I fix it(local_work_size) to 128 on 9600 and 512 on GTX580. > > Please notice that if you want to change local_work_size(I use THREADS in top of opencl_xsha512_fmt.c and xsha512_kernel.cl), you should both change the defination in both opencl_xsha512_fmt.c and xsha512_kernel.cl. I will solve it by pass a parameter to the OpenCL code in next commit. > > Thanks! > Dongdong Li > Content of type "text/html" skipped
Powered by blists - more mailing lists
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.