Date: Wed, 19 Aug 2015 04:10:18 +0200 From: Agnieszka Bielec <bielecagnieszka8@...il.com> To: john-dev@...ts.openwall.com Subject: Re: PHC: Argon2 on GPU 2015-08-19 2:46 GMT+02:00 Solar Designer <solar@...nwall.com>: > On Tue, Aug 18, 2015 at 01:35:47AM +0200, Agnieszka Bielec wrote: >> 2015-08-16 23:51 GMT+02:00 Solar Designer <solar@...nwall.com>: >> > What speeds are you getting on well's CPU for Argon2d at these settings? >> > With memory (de)allocation out of the loop, like we had for the Lyra2 >> > and yescrypt benchmarks. >> > >> > Also, please set m=1536, so we'd have exactly 1.5 MiB. >> >> I made some tests although there is still a bug on super's AMD. > > You mean the auto-tuning weirdness? yes, auto-tuning weirdness I discovered that slow-down is after I de-allocate and allocate again this big buffer. I blocked allocate_clobj and deallocate_clobj when certain part of the code was reached and I was playing around with allocation/deallocation in crypt_all() >> a@...l:~/f/run$ ./john --test --format=argon2d >> Will run 8 OpenMP threads >> Benchmarking: argon2d [Blake2 AVX]... (8xOMP) >> memory per hash : 1.50 MB >> 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: 7760 c/s real, 971 c/s virtual >> Only one salt: 7808 c/s real, 976 c/s virtual > > I assume it's just a temporary glitch that "Many salts" appeared > slightly slower than "Only one salt" here. They should be almost the > same, or "Many salts" very slightly better. In fact, once you're done > debugging these formats, you'll need to set BENCHMARK_LENGTH to -1 to > suppress these separate benchmark (there will be just one then: Raw). I tested bcrypt and scrypt and the same situation none@...e ~/Desktop/r/run $ ./john --test --format=bcrypt Will run 8 OpenMP threads Benchmarking: bcrypt ("$2a$05", 32 iterations) [Blowfish 32/64 X2]... (8xOMP) using different password for benchmarking DONE Speed for cost 1 (iteration count) of 32 Many salts: 6096 c/s real, 768 c/s virtual Only one salt: 6225 c/s real, 782 c/s virtual > >> a@...l:~/f/run$ ./john --test --format=argon2d --cost=1:1,1536:1536,5:5 >> Will run 8 OpenMP threads >> Benchmarking: argon2d [Blake2 AVX]... (8xOMP) >> memory per hash : 1.50 MB >> using different password for benchmarking >> DONE >> Speed for cost 1 (t) of 1, cost 2 (m) of 1536, cost 3 (l) of 5 >> Many salts: 7288 c/s real, 911 c/s virtual >> Only one salt: 7224 c/s real, 904 c/s virtual > > What's that "cost 3 (l)"? lanes
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.