john-dev - gpg and gpg-opencl benchmarks

Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date: Sun, 5 Apr 2015 12:44:04 +0300
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: gpg and gpg-opencl benchmarks

Frank, magnum -

The speeds reported by our gpg and gpg-opencl benchmarks look too high
to me:

[solar@...er run]$ export GOMP_CPU_AFFINITY=0-31
[solar@...er run]$ ./john -te -form=gpg-opencl -dev=1
Will run 32 OpenMP threads
Device 1: Tahiti [AMD Radeon HD 7900 Series]
Local worksize (LWS) 64, global worksize (GWS) 262144
Benchmarking: gpg-opencl, OpenPGP / GnuPG Secret Key [SHA1 OpenCL]... (32xOMP)DONE
Speed for cost 1 (iteration count) of 65536, cost 2 (hash algorithm [1:MD5 2:SHA1 3:RIPEMD160 8:SHA256 9:SHA384 10:SHA512 11:SHA224]) of 2, cost 3 (cipher algorithm [1:IDEA 2:3DES 3:CAST5 4:Blowfish 7:AES128 8:AES192 9:AES256]) of 3
Raw:    619237 c/s real, 46098 c/s virtual

619237*65536/10^9 = 40.6 billion SHA-1/second

This is way too much for one GPU in a 7990.

Is this really "for cost 1 (iteration count) of 65536"?  And if it is,
why is it so fast?

The --format=gpg exhibits the same problem, only with speeds lower
according to the CPU/GPU performance difference.  (So it reports speeds
that look too high for the CPUs.)

Alexander