CPU/GPU Throttling

[sp33dball]

I'm adding an additional machine and since it's an older setup I'm not sure if I need to worry about the GPU being throttled by the CPU. It's duo core T9500 CPU with a geforce 8600m GT, should I limit one core so the GPU has support from the CPU, or with this setup is it better to have the CPU working 100%? Thanks!

 

[eaglescouter]

Laptop system?

I'd go with an optimized install then carefully watch the temperatures.
Without additional airflow you might see thermal throttling.

 

[sp33dball]

I'm trying to get it optimized, it's running OSX and so far the Lunatics version doesn't seem to be working with it. Trying another. But thanks for the heads up on the temps.

 

[sp33dball]

Oh and another question, how should I setup the the app_info for a 590? I have it set to do .5 CUDA so that it runs 4 at a time, it looks to be running at 100% GPU use. Is there a better way to have this running? can it handle more?

 

[The Wizard]

I have the same question for my 590. doing .5 CUDA also.
Around 95-97% load. Does the card benefit from running
3 instances per GPU ?


The Wizard

 

[Team kizb]

I don't have a 590, but when I was running my 470s I had them set to .33 to run 3 instances per GPU. Seemed to work well for me.

 

[eaglescouter]

Yes run 3 instances on 580/590
2 instances on 570
And reserve once CPU core to feed one GPU

 

[sp33dball]

yeah it looks like anything lower than a 570 you don't want to run more than 2. My 560Ti runs through them in a few minutes, but with three it takes significantly longer, upwards to an hour or more per.

Oh also searching around I found a Boinc Rescheduler, kinda interesting. I haven't had a chance to really look into since I'm at work, but I don't remember seeing anything on it in any of the stickies here. So just a heads up if anyone wants to look into it.

 

[4GHZ_or_bust]

yeah it looks like anything lower than a 570 you don't want to run more than 2. My 560Ti runs through them in a few minutes, but with three it takes significantly longer, upwards to an hour or more per.

It seems optimal setting is about 1 wu per 512mb of GPU's memory but it's always YMMV.

Oh also searching around I found a Boinc Rescheduler, kinda interesting. I haven't had a chance to really look into since I'm at work, but I don't remember seeing anything on it in any of the stickies here. So just a heads up if anyone wants to look into it.

Not stickied but it got mentioned in this forum from time to time.