Improved fermi performance in seti@home

[bryanW1995]

from reading through the lunatics forums I noticed that you can split up cuda wu's on a single fermi gpu. it takes up about 250 mb/wu, so be careful on weaker cards (I'm only running 2 for example). here's how you do it, at least on win 7:

go to: C(colon)\ProgramData\BOINC\projects\setiathome.berkeley.edu won't let me write c:\ ... grrrrr

you'll find a file called app_info. edit this file. it's a pretty large file, but you'll find the following near the bottom:

<type>CUDA</type>
<count>1</count>


change the 1 to .5 to run 2 wu's at a time, change it to .25 to run 4 wu's at a time, etc.

when you exit seti and restart the count defaults back to 1, don't know how to stop that.

anyway, I'm getting ~ 50&#37; better performance out of my gtx 460 from doing this.

edit: more like 25-50%, depending upon the wu.

 

[petrusbroder]

Thanks for the info!

 

[Ken g6]

I sort of expected that from higher-end Fermi, since those cards can run two processes at the same time. I didn't expect it from 460s and lower. Thanks for the insight! 🙂

 

[bryanW1995]

well, it's 50&#37; on some, 25% on others now. I jumped to the 50% conclusion after a huge trial of 2 wu's. :$

plus, my 460 is oc'd 24/7 to 880/4300, so it's very close to a stock 470 performance.