Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 19 Sep 2011 17:02:45 +0400
From: Solar Designer <>
Subject: Re: OpenCL performance numbers on Radeon HD 5870

Paul - thanks!

All - this started as off-list discussion between Paul, Lukas, and me -
initially focusing on poor performance of the phpass OpenCL code that
Paul was observing.  I asked Paul to bring it to the list.

On Mon, Sep 19, 2011 at 12:43:47PM +0000, Relkin, Paul wrote:
> ./john -test -format=phpassopencl
> Benchmarking: PHPASSOPENCL [PORTABLE-MD5]...
> OpenCL Device: Cypress
> Raw:    35415 c/s real, 131764 c/s virtual

This is poor performance.  Lukas is going to look into it once he has a
similar GPU himself (should be soon).  Meanwhile, the CUDA and OpenCL
patches are providing much better performance on NVidia cards: 200k c/s
to 300k c/s real at a phpass benchmark similar to the above on GTX 460.

> ./john -test -format=cryptmd5opencl
> Benchmarking: CRYPTMD5OPENCL [MD5-based CRYPT]...
> OpenCL Device: Cypress
> Error while building program [-11]
>  /tmp/ catastrophic error: cannot open source file
>           "../src/cryptmd5opencl_fmt.h"
>   #include "../src/cryptmd5opencl_fmt.h"
> ./john -test -format=cryptsha256opencl
> Benchmarking: CRYPTSHA256OPENCL [SHA256-based CRYPT]...
> OpenCL Device: Cypress
> Error while creating program [-30]

Lukas is to look into these failures as well.  Obviously, this did not
happen for the systems that he developed and tested the code on.

Thanks again,


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.