Date: Mon, 31 Aug 2015 10:26:43 +0300 From: Solar Designer <solar@...nwall.com> To: john-dev@...ts.openwall.com Subject: Re: PHC: Argon2 on GPU On Sun, Aug 30, 2015 at 11:56:32PM +0200, Agnieszka Bielec wrote: > [a@...er run]$ LWS=64 GWS=2348 ./john --test --format=argon2d-opencl > --v=4 --dev=5 > Benchmarking: argon2d-opencl [Blake2 OpenCL]... > memory per hash : 1.50 MB > Device 5: GeForce GTX TITAN > Options used: -I ./kernels -cl-mad-enable -cl-nv-verbose -D__GPU__ > -DDEVICE_INFO=65554 -DDEV_VER_MAJOR=352 -DDEV_VER_MINOR=21 > -D_OPENCL_COMPILER -DBINARY_SIZE=256 -DSALT_SIZE=64 > -DPLAINTEXT_LENGTH=125 > Local worksize (LWS) 64, global worksize (GWS) 2304 > using different password for benchmarking > DONE > Speed for cost 1 (t) of 1, cost 2 (m) of 1536, cost 3 (l) of 1 > Many salts: 9124 c/s real, 9035 c/s virtual > Only one salt: 9124 c/s real, 9124 c/s virtual Your specified GWS gets rounded to 2304 (as you can see above), and besides you can do slightly better: [solar@...er run]$ LWS=48 GWS=2304 ./john -test -format=argon2d-opencl -v=4 -dev=5 Benchmarking: argon2d-opencl [Blake2 OpenCL]... memory per hash : 1.50 MB Device 5: GeForce GTX TITAN Options used: -I ./kernels -cl-mad-enable -cl-nv-verbose -D__GPU__ -DDEVICE_INFO=65554 -DDEV_VER_MAJOR=352 -DDEV_VER_MINOR=21 -D_OPENCL_COMPILER -DBINARY_SIZE=256 -DSALT_SIZE=64 -DPLAINTEXT_LENGTH=125 Local worksize (LWS) 48, global worksize (GWS) 2304 using different password for benchmarking DONE Speed for cost 1 (t) of 1, cost 2 (m) of 1536, cost 3 (l) of 1 Many salts: 9680 c/s real, 9600 c/s virtual Only one salt: 9600 c/s real, 9600 c/s virtual > I just discovered that argon2d which uses coalescing is faster on both > TITAN, it's slower with coalescing on my 960m so I didn't tested this > on super. I should. I was trying many things for some period of time > but these changes in results are from that we tested argon more > carefully. to turn on coalescing in argon2d it's only needed to change > 4 lines but I will make #define USE_COALESCING 0/1 very soon It's unexpected that coalescing for Argon2d is of so much help. Please confirm these speeds by actual cracking, with a wordlist containing no duplicate lines. > I need to retest Lyra and yescrypt using GWS not of power 2 Looks like we need a script to test many LWS/GWS combinations. I am running a simple script (testing 128 combinations) against Argon2d on TITAN X now (and we'll probably need to test many more once we have improved the code some further, since the current code is obviously still far from optimized and thus might not be worth a lot of tuning and benchmarking effort). Alexander
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.