Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 2 May 2012 15:50:06 -0300
From: Claudio André <claudioandre.br@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: cryptsha512-opencl

>cryptsha512-opencl currently in magnum-jumbo still fails tests on both
>GPUs in the bull machine.
It works on NVIDIA now.

OpenCL platform 0: NVIDIA CUDA, 1 device(s).
Using device 0: GeForce GTX 570
...
Calculating best keys per crypt, this will take a while Optimal keys per
crypt 7168
(to avoid this test on next run, put "cryptsha512_KPC = 7168" in john.conf,
section [Options:OpenCL])
Local work size (LWS) 32, Keys per crypt (KPC) 7168
Benchmarking: crypt SHA-512 (rounds=5000) [OpenCL]...DONE
Raw: 3018 c/s real, 3018 c/s virtual

I liked this style of KPC seeking. I recommend.
----
kpc:    512 1051 c/s       5259446 rounds/s   0.487 sec per crypt_all()+
kpc:   1024 2089 c/s      10446542 rounds/s   0.490 sec per crypt_all()+
kpc:   1536 1583 c/s       7915400 rounds/s   0.970 sec per crypt_all()
kpc:   2048 2102 c/s      10514150 rounds/s   0.974 sec per crypt_all()
kpc:   2560 2611 c/s      13059767 rounds/s   0.980 sec per crypt_all()+
kpc:   3072 2106 c/s      10532374 rounds/s   1.458 sec per crypt_all()
kpc:   3584 2448 c/s      12243880 rounds/s   1.464 sec per crypt_all()
kpc:   4096 2785 c/s      13929308 rounds/s   1.470 sec per crypt_all()+
kpc:   4608 2367 c/s      11836217 rounds/s   1.947 sec per crypt_all()
kpc:   5120 2622 c/s      13111256 rounds/s   1.953 sec per crypt_all()
kpc:   5632 2865 c/s      14326604 rounds/s   1.966 sec per crypt_all()+
kpc:   6144 2525 c/s      12629180 rounds/s   2.432 sec per crypt_all()
kpc:   6656 2727 c/s      13638341 rounds/s   2.440 sec per crypt_all()
kpc:   7168 2917 c/s      14586892 rounds/s   2.457 sec per crypt_all()+
...

2012/4/29 Solar Designer <solar@...nwall.com>

> Claudio -
>
> cryptsha512-opencl currently in magnum-jumbo still fails tests on both
> GPUs in the bull machine.  Can you please reproduce this and try to fix
> it, using your remote access to this machine?
>
> Lukas' CUDA code:
>
> user@...l:~/john/magnum-jumbo/src$ ../run/john -te -fo=cryptsha512-cuda
> Benchmarking: cryptsha512-cuda [SHA512-based CRYPT]... DONE
> Raw:    3853 c/s real, 3853 c/s virtual
>
> OpenCL code on the same card:
>
> user@...l:~/john/magnum-jumbo/src$ ../run/john -te -fo=cryptsha512-opencl
> OpenCL platform 0: NVIDIA CUDA, 1 device(s).
> Using device 0: GeForce GTX 570
> Segmentation fault (core dumped)
>
> OpenCL code on HD 7970:
>
> user@...l:~/john/magnum-jumbo/src$ ../run/john -te -fo=cryptsha512-opencl
> -pla=1OpenCL platform 1: AMD Accelerated Parallel Processing, 2 device(s).
> Using device 0: Tahiti
> Compilation log: LOOP UNROLL: pragma unroll (line 210)
>    Unrolled as requested!
> LOOP UNROLL: pragma unroll (line 140)
>    Unrolled as requested!
> [...]
> LOOP UNROLL: pragma unroll (line 333)
>    Unrolled as requested!
>
> Local work size (LWS) 64, Keys per crypt (KPC) 8192
> Benchmarking: crypt SHA-512 (rounds=5000) [OpenCL]... FAILED
> (get_hash[0](0))
>
> OpenCL code on CPU (works):
>
> user@...l:~/john/magnum-jumbo/src$ ../run/john -te -fo=cryptsha512-opencl
> -pla=1 -dev=1
> OpenCL platform 1: AMD Accelerated Parallel Processing, 2 device(s).
> Using device 1: AMD FX(tm)-8120 Eight-Core Processor
> Compilation log: LOOP UNROLL: pragma unroll (line 140)
>    Unrolled as requested!
> LOOP UNROLL: pragma unroll (line 130)
>    Unrolled as requested!
> [...]
> LOOP UNROLL: pragma unroll (line 333)
>    Unrolled as requested!
>
> Local work size (LWS) 64, Keys per crypt (KPC) 8192
> Benchmarking: crypt SHA-512 (rounds=5000) [OpenCL]... DONE
> Raw:    1432 c/s real, 179 c/s virtual
>
> (BTW, I reverted the CPU to stock clocks for now.)
>
> Thanks,
>
> Alexander
>

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.