Article: Parallelism at HotPar 2010
By: Groo (charlie.delete@this.semiaccurate.com), August 17, 2010 8:18 pm
Room: Moderated Discussions
Steve Underwood (steveu@coppice.org) on 8/17/10 wrote:
---------------------------
>13x or more is a pretty nice speed up for a low cost solution. Even 2x is not insignificant
>if the code is very heavily used by a large audience(e.g. video codecs), which can
>justify the extra development effort. However, MCML is pretty much a best case problem.
>Most number crunching tasking will not fit the hardware nearly as well.
>
For heavily used code, or code running on racks of machines, performance per watt tends to have a lot of sway. In this case, a 2x speedup for adding a GPU over say, integrated graphics, may not be worth it.
For the home user, you already have the CPU. Is a 2x speedup worth adding a GPU for? Debatable, but for things like video codecs, if you can decode at full frame rate, 2x is not a selling point. :)
-Charlie
---------------------------
>13x or more is a pretty nice speed up for a low cost solution. Even 2x is not insignificant
>if the code is very heavily used by a large audience(e.g. video codecs), which can
>justify the extra development effort. However, MCML is pretty much a best case problem.
>Most number crunching tasking will not fit the hardware nearly as well.
>
For heavily used code, or code running on racks of machines, performance per watt tends to have a lot of sway. In this case, a 2x speedup for adding a GPU over say, integrated graphics, may not be worth it.
For the home user, you already have the CPU. Is a 2x speedup worth adding a GPU for? Debatable, but for things like video codecs, if you can decode at full frame rate, 2x is not a selling point. :)
-Charlie
