Re: difference between cuda and stream

Mike Reed <[email protected]> Tue, 8 Jan 2013 23:10:28 +0000
Newsgroups gmane.comp.distributed-net.rc5
Message-ID <CACwkm_MX=wJbrc9D3wE6h_35K07tobSar-x_M9byXeqToyfz8A@mail.gmail.com>
Hi,

To follow on from my earlier notes..

If you look at bug 4512 in bugzilla, you will see that the AMD HD 7870
chip running at 1 Ghz can turn over RC5 keys at a rate of 1.9 Gk/sec.
This is a speed of over 7 stats units per second. At these speeds, I/O
becomes a limiting factor, which is one of the reasons that we have
moved towards larger packets of work.

Kind regards,
Mike

On 8 January 2013 20:05, Mike Reed <[email protected]> wrote:
> Hi,

> One stats unit / block is 2^28 keys in length. Not nearly something or maybe
> something. For graphics cards, iter 64 (i.e. 64 x 2^28) is common. Proxies
> give out packets up to iter 256.

> Kind regards,
> Mike
_______________________________________________
rc5 mailing list
rc5-Ra3b/QYEcJ3d140v2zMXi0fjHoOT/h/[email protected]
http://lists.distributed.net/mailman/listinfo/rc5