GTX 580 running at 85-95% for the duration of crunching

I tried crunching with my GTX 580 yesterday, and I really didn't like how the card was perma working near 100% while crunching. I thought that at this rate I was going to run the card into the ground pretty fast, so I switched off GPU crunching.

Is there any way you folks know where one is able to make the card operate/crunch while working at less than 50% usage?

This is my only rig, so I really don't have the luxury to blow it up. GTX 580 is also an expensive card (I own Gigabyte GTX 580 Super Overclock Windforce 3X).

What do you guys recommend?

Use TThrottle:
http://www.efmer.eu/boinc/

It will not allow to set exact GPU load% - it works on temperature.
You "Select a temperature, check "Auto active" and TThrottle handles the rest."
('the rest' means: TThrottle pauses the apps (computing threads) in short intervals and keeps the temperature at the max you selected/typed
If the temperature is below the max you selected - the device (CPU/GPU) will run at 100%
If it gets 'hot' - pauses in app will occur (which will be visible by less CPU/GPU load) to keep the temperature at the max selected (e.g. 55°C for CPU and 65°C for GPU)
)


____________



- ALF - "Find out what you don't do well ..... then don't do it!" :)

Thanks. I had a look at the software you recommended.

BTW, I am really impressed at the rate which my GTX 580 is completing SETI tasks compared to my Core 2 Extreme QX 9650 running @ stock.

Both GPU and CPU are running at stock.

GPU processes an ASTRO PULSE task in about 30-35 minutes. My CPU needed for the same task 23 hours.

GPU processes normal SETI tasks in about 1.5 to 6 minutes. My CPU needed for the same tasks between 6 and 7 hours.

In fact, the difference in processing time between CPU & GPU in SETI tasks is so unbelievably huge, that I stopped my CPU from processing SETI tasks altogether and assigned it to do some humanitarian work from World Grid. World Grid tasks seem to run faster on my CPU.
____________

Those are the miracles of parallel computing.

Today's GPUs are (almost) as fast as supercomputer of only decade ago:
- your GPU NVIDIA GeForce GTX 580 have 1581.1 GFLOPs "theoretical shader performance in single-precision floating point operations":
http://en.wikipedia.org/wiki/GeForce_500_Series#Products

1500 GFLOPs = 1.5 TFLOPs (TeraFLOPs)
http://en.wikipedia.org/wiki/Tera-

And "As of November 2004, the Cray X1 had a maximum measured performance of 5.9 teraflops"
http://en.wikipedia.org/wiki/Cray#Cray_Inc.:_2000_to_present

"The Cray X1 is ... sold by Cray Inc. since 2003 ... offering a peak speed of 12.8 gigaflops per processor."
"The largest unclassified X1 system was the 512 processor system at Oak Ridge National Laboratory"
http://en.wikipedia.org/wiki/Cray_X1

512 * 12.8 = 6553.6 gigaflops = 6.55 teraflops (theoretical)

And the systems from 'Top hosts' can compute faster (if programed in CUDA and algorithm/code is highly parallel)
http://setiathome.berkeley.edu/top_hosts.php

"Coprocessors [8] NVIDIA GeForce GTX 690" (= 4 video cards):
http://setiathome.berkeley.edu/show_host_detail.php?hostid=5948046


____________



- ALF - "Find out what you don't do well ..... then don't do it!" :)

To expand on what you said, I made a few calcs:

I had originally joined SETI@Home on 25 Oct 2005. From that date since I stopped participating on 28 Aug 2006, my Pentium 4 Dual Core@3.20 GHz had crunched 3720 units.

I returned on 1st Feb 2013, and since then my GTX 580 GPU alone, working from morning til evening (not 24/7) has crunced about 30000 units in less than 8 days. :-)
____________