I just acquired (thanks, Craigslist!) a 2GB MSI Twin Frozr III GTX 680 card, and am wondering how it should stack up against my twin GTX 460s running 2 WUs each (for a total of 4 GPU threads). The 460s run in the mid-700MHz range, while the 680 is as high as 1100+ under high usage turbo boost.

I hope to use the 680 to replace both 460s (to save on my electric bill) without decreasing my actual WUs crunched.

Will the 680 do this?

Any helpful hints/info would be appreciated!

Preliminary results show that (one 460 with the 680, so 2 threads each) the 680 is running at 95+% GPU usage with 2 WUs, and doesn't seem to be putting out work 2x as much as the 460. Am I missing something here?

I should add that they are running Cuda42, as that's what the 460s support. Would reinstalling Lunatics and picking Cuda50 help throughput on the 680 significantly?

---

It might get close.
My number 2 rig has 2 GTX680s.
Doing about 59k RAC. They are OCd a little.
Running the Lunatics 50 Cuda apps. Not sure 50 is much faster than 42.
2/per. You could try 3/per on a single 680 to try to push the utilization closer to 99%.

A single 680 should do a fine job by itself. Leave a core or two free to support it. Regardless, you should see some power and heat savings from the 680 vs dual 460s.

---

"Freedom is just Chaos, with better lighting." Alan Dean Foster

It might get close.
My number 2 rig has 2 GTX680s.
Doing about 59k RAC. They are OCd a little.
Running the Lunatics 50 Cuda apps. Not sure 50 is much faster than 42.
2/per. You could try 3/per on a single 680 to try to push the utilization closer to 99%.

A single 680 should do a fine job by itself. Leave a core or two free to support it. Regardless, you should see some power and heat savings from the 680 vs dual 460s.

I'm running 3 ea. cuda50 on my GTX660SCs @ 99%, & temps. are running approx. 76c, so 3 on a 680 should not be a problem. Right now both GPUs are running between 1159 & 1176 MHz. I would note that the run times have gone from approx. 20 min. each to approx. 33 each.

---

I don't buy computers, I build them!!

Sorry to hijack the thread, but where might one find information as to how to get your GPU to run more than 1 task at a time? Is this generally something to strive for?

[...] information as to how to get your GPU to run more than 1 task at a time? [...]

You're using BOINC Version 7.0.64, OK... BUT your GPUs are not going to take it. Please don't do the following on your rigs:

Go to your '..\BOINC\projects\setiathome.berkeley.edu' directory and create a file named 'app_config.xml' with a content like this:

<app_config>
	<app>
		<name>setiathome_v7</name>
		<gpu_versions>
			<gpu_usage>0.5</gpu_usage>
			<cpu_usage>0.06</cpu_usage>
		</gpu_versions>
	</app>
	<app>
		<name>setiathome_enhanced</name>
		<gpu_versions>
			<gpu_usage>0.5</gpu_usage>
			<cpu_usage>0.06</cpu_usage>
		</gpu_versions>
	</app>
	<app>
		<name>astropulse_v6</name>
		<max_concurrent>6</max_concurrent>
		<gpu_versions>
			<gpu_usage>0.5</gpu_usage>
			<cpu_usage>1.00</cpu_usage>
		</gpu_versions>
	</app>
</app_config>

As '<gpu_usage>0.5</gpu_usage>' stands for 2 tasks at a time. Change it to '<gpu_usage>0.33</gpu_usage>' for 3 WUs at a time etc.

Hope that helps.

/EDIT:
PS: Then stop and restart BOINC. Or at least go to the Extras menu and click something like 'read in config file'.

---

.:morpheus:.

It might get close.
My number 2 rig has 2 GTX680s.
Doing about 59k RAC. They are OCd a little.
Running the Lunatics 50 Cuda apps. Not sure 50 is much faster than 42.
2/per. You could try 3/per on a single 680 to try to push the utilization closer to 99%.

A single 680 should do a fine job by itself. Leave a core or two free to support it. Regardless, you should see some power and heat savings from the 680 vs dual 460s.

I was thinking of upgrading my linux PC which now runs a gtx 260 with 2 cards. Would you say I should go for 2x 680 or, since there comes new cards often, there's a better option?

What size PSU should I go for?

Be careful, i dont think it will work on his GPU`s.
2 are notebooks and one 8800.
All with only 512 MB RAM

---

With each crime and every kindness we birth our future.

Be careful, i dont think it will work on his GPU`s.
2 are notebooks and one 8800.
All with only 512 MB RAM

I was about 2 minutes away from giving the same warning.
I'm not sure 2/per would be a good thing for those rigs.

---

"Freedom is just Chaos, with better lighting." Alan Dean Foster

Be careful, i dont think it will work on his GPU`s.
2 are notebooks and one 8800.
All with only 512 MB RAM

I was about 2 minutes away from giving the same warning.
I'm not sure 2/per would be a good thing for those rigs.

It wouldn`t even work with 512 MB VRAM.
Evenso 8800 is pre Fermi and 1 instance only card.

---

With each crime and every kindness we birth our future.

Be careful, i dont think it will work on his GPU`s.
2 are notebooks and one 8800.
All with only 512 MB RAM

Oh yeah, I missed that. Sorry. Could be some trouble around...
But at least he knows what to do if the 8800 GPU gets an upgrade. ;)

---

.:morpheus:.

It might get close.
My number 2 rig has 2 GTX680s.
Doing about 59k RAC. They are OCd a little.
Running the Lunatics 50 Cuda apps. Not sure 50 is much faster than 42.
2/per. You could try 3/per on a single 680 to try to push the utilization closer to 99%.

A single 680 should do a fine job by itself. Leave a core or two free to support it. Regardless, you should see some power and heat savings from the 680 vs dual 460s.

I was thinking of upgrading my linux PC which now runs a gtx 260 with 2 cards. Would you say I should go for 2x 680 or, since there comes new cards often, there's a better option?

What size PSU should I go for?

I hope jravin is not bothered by the wandering away from his original question.

2 x 680 makes a solid rig, but depending on your budget, there certainly are newer 700 series cards to consider.
As to the PSU, the more the merrier. My dual 680 rig is usually in the 600-700 watt range. I am running a 910w PC Power and Cooling Silencer. I also have in other rigs a couple of 950w Silencers. Excellent PSUs. Kittyman's choice.

---

"Freedom is just Chaos, with better lighting." Alan Dean Foster

Back from a 2-day mini-vacation, so I'll take my thread back, if you please:

I'm running 3 ea. cuda50 on my GTX660SCs @ 99%, & temps. are running approx. 76c, so 3 on a 680 should not be a problem. Right now both GPUs are running between 1159 & 1176 MHz. I would note that the run times have gone from approx. 20 min. each to approx. 33 each.

In which case, you are losing, NOT gaining work done/RAC:

3 threads: in 33 minutes, you do 3 x 1 WU = 3.00 WUs
2 threads: in 33 minutes, you do 2 x 1 WU + 2 x 13/20 WU = 3.3 WUs

You are doing 10% less work and getting 10% less RAC by running 3 threads instead of 2!!!

---

Back from a 2-day mini-vacation, so I'll take my thread back, if you please:

I'm running 3 ea. cuda50 on my GTX660SCs @ 99%, & temps. are running approx. 76c, so 3 on a 680 should not be a problem. Right now both GPUs are running between 1159 & 1176 MHz. I would note that the run times have gone from approx. 20 min. each to approx. 33 each.

In which case, you are losing, NOT gaining work done/RAC:

3 threads: in 33 minutes, you do 3 x 1 WU = 3.00 WUs
2 threads: in 33 minutes, you do 2 x 1 WU + 2 x 13/20 WU = 3.3 WUs

You are doing 10% less work and getting 10% less RAC by running 3 threads instead of 2!!!

Now that's strange as I get around a 30-40% increase in production on my GTX660's by running 3 over 2, but then I guess YMMV.

I am a bit alarmed about those temps though, mine only run at 1100MHz, but they only get mid to high 60's during summer here (currently they are sitting in the low to mid 50's but it is winter here now).

Cheers.

Wiggo - running 2 threads on my GTX 680 gives me about 57-60C, and it is running at 95%+ usage. I may try 3 WUs later on, but I don't foresee much temp increase, since runniing nearly flat out now. And it is 93F outside here in Newton, MA, USA. And, since there is not much headroom on useage, I don't expect any increase in throughput.

---

runniing?

http://en.wikipedia.org/wiki/IEEE_754-1985

Sorry, jravin, for sending everyone off topic :)

But I do appreciate the information everyone, I will wait for an upgrade before I take on this task!

Wiggo - running 2 threads on my GTX 680 gives me about 57-60C, and it is running at 95%+ usage. I may try 3 WUs later on, but I don't foresee much temp increase, since runniing nearly flat out now. And it is 93F outside here in Newton, MA, USA. And, since there is not much headroom on useage, I don't expect any increase in throughput.

I get the same temps here usually during summer and I was more concerned by Cliff's temps for his 660's.

Your temps sound fine to me (I'd never have the $'s to own a high end card, but I can use 2 or 3 cheaper 1's), but looking at the numbers I'd have thought that 3 concurrent workunits would be the simple "sweet spot" for a 680 (though I'd be tempted to see what would happen with 4).

Cheers.

Now that's strange as I get around a 30-40% increase in production on my GTX660's by running 3 over 2, but then I guess YMMV.

YMMV indeed - it is very hard to compare WUs for SETI - there's no way of being sure about the actual "size" of the computations in each. Hence the comparison he made of 20 min in 2 threads vs. 33 min in 3 threads may be off, and by a lot.

I go by GPU utilization - I figure if 2 have 95-99% utlization, a 3rd one has little, if anything, to work with, and will therefore show no improvement. BICBW. (But I Could Be Wrong).

---