I've got a system that is short a graphics card because I gave away the 9800GT that was in there. Which card gives the most performance for the money? I guess it should be somewhat more future proof than my GT 240 cards as well.

Looking at prices the GTX 590 and 580 are too much. The race is between the GTX570, 560, 560ti, 470 and 460.

Gigabyte has a GTX 560 ti with WINDFORCE 2X for around $205 that is clocked at 900MHz so it is super fast, but a the card gets bad reviews due to reliability issues.

Do not buy CC2.1 cards for GPUGrid, because only the 2/3 of their shaders can be used by GPUGrid. GTX570 or GTX470 is the best choice based on the performance/price ratio. I've bought a very cheap (probably factory renewed) MSI GTX 480 Twin Frozr II card. It might be an alternative for you as well.

Retvari, thanks for the reply. You have some nice equipment and your credit is very impressive.

It is cool that GPU grid allows for snooping of members computers and results.

Using a 24,990 credit work unit as a reference point I checked out your systems vs mine and BikerMatt's. Here is what I discovered:

Card--Time--Operating system

580--17,210--XP
480--16,782--XP
570--16,847--Linux
460--42,498--XP

Running Linux BikerMatt's GTX570 is roughly equal to your GTX480 or GTX580 cards on the 24,990 credit work unit. With the 570 being cheaper I'm temped to go with the 570, unless these results are in error somehow.

Do you think Linux accounts for some of what we are seeing here?

Totally agree with Zoltan on this; 570 or 470 is the GPU to get.

There are lots of comparison considerations:
Linux is using the 6.14app and Windows uses the 6.15app,
Is SWAN_SYNC being used?,
CPU type and speed,
PCIE bandwidth (X4, X8, X16),
What else is being processed,
Hard drive speeds sometimes,
Operating system updates.

But all that is irrelevant, a reference GTX570 is about 4.4% faster than a reference GTX480 (and 14% more power efficient):
GTX 570 GF110 40nm Compute Capable 2.0 1405 BoincGFlops peak (1896)
GTX 480 GF100 40nm Compute Capable 2.0 1345 BoincGFlops peak (1816)

NVidia GPU Card comparisons in GFLOPS peak

Thanks for the advice! I've pretty much settled in on the GTX-570. There is a highly rated 570 that has 3 fans but it is over 11 inches long and might not fit in the box. I'll most likely measure and order tonight. Thanks again!

Ordered! The 3 fan unit was just too big. Got a Galaxy GTX-570 inbound for about $290 after rebate. The Galaxy card is shorter than most at 9 or maybe 9.3 inches which is nice.

If you were wanting the 3 fan model because it would run cooler then don't worry. My GTX 570 has only 1 fan which never runs faster than 80% of max speed. Even on the hottest days this summer the GPU temp never went over 70 Celsius, usually it's 62 to 65 Celsius, 58 at night when the room really cools off. Mind you the hottest day we saw this summer was only 32 Celsius which isn't at all hot by some standards.

The only problem I've had with mine is that the fan speed does not increase when the GPU temp rises. At 90 Celsius the fan speed stays at 40% if left on "automatic speed control". I've forced it to run at 80% all the time so in the end it's not a big deal and I'm very happy with the card. Your card is from a different manufacturer (mine is by Asus) so your fan may work properly but I suggest you have your temp monitor running when you get that first GPU task just in case.

The gtx-570 finally shipped today.

gtx-570 has arrived. Will install tonight, late.

Galaxy gtx-570 installed and running just now. Sinking its teeth into a 291-MJHARVEY_NPC2-14-50-RND6391_0 work unit. It is running a little faster than one percent a minute.

Until work unit 291-MJHARVEY_NPC2-14-50-RND6391_0 today 10-30-2011 that same system was running a GT-240 card.

Here's a script I run on my Linux box to monitor GPU temperature, fan speed, clock frequencies, GPU load and memory load. It updates every 2 secs. I shrink the terminal I run it in and stick it in the corner.

watch -t -n 2 "nvidia-settings -t -q [gpu:0]/GPUCoreTemp                  
               nvidia-settings -t -q [fan:0]/GPUCurrentFanSpeed           
               nvidia-settings -t -q [gpu:0]/GPUCurrentClockFreqs         
               nvidia-settings -t -q [gpu:0]/GPUCurrentProcessorClockFreqs
               nvidia-smi -a | grep GPU
               nvidia-smi -a | grep Memory"

bash: watch: command not found

From nVidia X Server Settings

fan speed 49%
temp 72 C

Was set to "adaptive" power savings, just changed to "max performance".

Dagorath

Comparing the first couple of work units to yours my gtx-570 is slightly slower, but that might be expected.

My system is being run by a 2.33 GHz Core-2-Duo, which is much slower than your CPU.

I could trade the 2.33 GHz Core-2-Duo for a 3.0 GHz Core-2-Duo which is currently in another system, do you think it would be worth the trouble?

Also my GPU core is running 732 MHz, did you once say yours was running 742 MHz?

Dagorath

Comparing the first couple of work units to yours my gtx-570 is slightly slower, but that might be expected.

My system is being run by a 2.33 GHz Core-2-Duo, which is much slower than your CPU.

I could trade the 2.33 GHz Core-2-Duo for a 3.0 GHz Core-2-Duo which is currently in another system, do you think it would be worth the trouble?

Also my GPU core is running 732 MHz, did you once say yours was running 742 MHz?

The script in my last post outputs the numbers in the following order, my readings are in brackets:

temperature (60 Celsius)
fanspeed (85%)
graphics clock, memory clock (742 MHz, 1900 MHz)
processor clock (1484 MHz)
GPU load (80% to 99%)
Memory load (~26%)

Your 732 MHz number might be for the graphics clock?

I see you've got SWAN_SYNC=0 but have you also limited BOINC to only 1 core to leave the other core free to feed the GPU? If not then try that first before switching the CPU. I doubt the 3.0 GHz CPU will make much of a difference but I have no experience to base that opinion on.

I'm using NVIDIA driver version 260.19.44 and X server 11.0. Could there be a difference there? My NVIDIA driver is old, I don't know why I'm using it, just too lazy to change.

I've gone in and looked at all the settings I could find in an attempt to use both CPU cores on that machine. I did find one spot where I changed 20% to 100% CPU usage but the thing is still only using one core after an update?? My other Linux box uses both cores, but it has 2 graphics cards. The Windows XP system with the single GTX-460 card is also only using 1 core...

In your preferences, on Processor Usage tab....

1) Set "While processor usage is less than... " to 0 (equivalent to 100).

2) Set "On multiprocessor systems, use at most... " to 50% on a 2 core system that is attached to more projects than just GPUgrid. That leaves 1 core to the other project and the second core to service the GPU. I assume the machine that has 2 GPUs is attached only to GPUgrid so leave this setting at 100% on that machine.

So I think only using one core is normal behavior? BikerMatt checked his systems and they only use one core (plus a little more) per GPU. BikerMatt's better results are mostly a result of having overclocked cards while mine runs the stock speed of 732 MHz.