|
Message-ID: <2cfcfe0a201e3324af3b1634a93fb11a@smtp.hushmail.com> Date: Fri, 27 Apr 2012 01:13:55 +0200 From: magnum <john.magnum@...hmail.com> To: john-dev@...ts.openwall.com Subject: Re: New RAR OpenCL kernel On 04/27/2012 12:40 AM, Solar Designer wrote: > On Thu, Apr 26, 2012 at 11:29:43PM +0200, magnum wrote: >> But with sane durations (like 3 secs @16K GWS) I now have ~4500 c/s on >> GTX570/580 as well as HD 7970. > > That's very nice. There ought to be much room for improvement on the > 7970, though. For GTX5x0 too, but even more for 7970. By the way, GTX680 is slower than GTX580 on the few things I have tested. So much for that flagship. At least for GPGPU, AMD is still the leader I guess. >> [...] I'm now trying to figure out how >> to make the host code "sort" the lengths in some effective way. I got a >> vague idea: What if I launch all applicable kernels at once (the host >> code may decide that we need kernels for eg. length 4, 5 and 6) after >> the host code sets up an array with pointers for each length >> accordingly. I think this would be fairly cheap in this context. > > Why not just run the per-length kernels sequentially, out of one call to > crypt_all()? Do you prefer to run them in parallel because there would > be less than the full number of passwords with some lengths, and you > want to compensate for that? Maybe only run 2+ kernels at a time in > this special case (too few passwords with a given length)? I'd have to experiment with that, but it seems natural to me to just launch all of them without waiting for any previous. I guess I should interleave transfers and launches though. Context switches in GPU are not only cheap, they are a fundamentally good thing if I understand things right. Actually I would think running 4 kernels in parallel with 4K each would hide latency just as well as 1 kernel with 16K so I might not even have to buffer more keys. But I could very well be wrong. magnum
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.