If someone does, would they please run Fred's Performance program on it and post the results?

Thanks.

Seconded, please do if you do!

---

Just ordered 4 of them. The Palit overclocked variety (base clock 1006Mhz). Not quite as fast as the EVGA Superclocked ones. They will be replacing a bunch of GTX560Ti's.

---

BOINC blog

Just ordered 4 of them. The Palit overclocked variety (base clock 1006Mhz). Not quite as fast as the EVGA Superclocked ones. They will be replacing a bunch of GTX560Ti's.

I can't wait to hear about your experience with them. Thanks for sharing.

When do you expect them?

When do you expect them?

They arrived at shop this morning. I'll be picking up on Thursday (I have something on tomorrow evening).

---

BOINC blog

Nice!
I recently picked up a gtx 570 classified, but would liquidate that for a set of 660's if they perform well.

---

Okay first one is installed. Just downloaded a few multibeams for it and its running.

27/09/2012 8:37:28 PM | | NVIDIA GPU 0: GeForce GTX 660 (driver version 306.23, CUDA version 5.0, compute capability 3.0, 2048MB, 1930MB available, 2057 GFLOPS peak)
27/09/2012 8:37:28 PM | | OpenCL: NVIDIA GPU 0: GeForce GTX 660 (driver version 306.23, device version OpenCL 1.1 CUDA, 2048MB, 1930MB available)

Links to the first WU:
1st
2nd
3rd
4th

Looking slightly faster at the moment than the 560Ti that it replaced, but this is a very small sample and these are all shorties.

---

BOINC blog

Completed a couple of Einstein WU. 560Ti average is 46 min. On a miniscule sample size of 2 the 660 came in at 34.5 mins.

Need more samples to get a decent comparison

---

BOINC blog

Completed a couple of Einstein WU. 560Ti average is 46 min. On a miniscule sample size of 2 the 660 came in at 34.5 mins.

Need more samples to get a decent comparison

Can I convince you to run Fred's Performance tool V1.4 on them? (the automatic test)

Please?

I'd like to compare the results of my 660Tis with your 660s, especially since you seem to get better times with the 660s than with 560Tis. That's a surprise and if it holds true, you've discovered a gem.



Here are my results, 660Ti - reference cards at reference clocks, for .5 and .33 (two and three work units at a time; which are of most interest to most of us), using x41g included with the tool:



27 September 2012 - 11:30:38 Start, devices: 2, device count: 2 (0.50)
--------------------------------------------------------------------------- Results:
Device: 0, device count: 2, average time / count: 332, average time on device: 166 Seconds (2 Minutes, 46 Seconds)
Device: 1, device count: 2, average time / count: 330, average time on device: 165 Seconds (2 Minutes, 45 Seconds)





27 September 2012 - 11:36:16 Start, devices: 2, device count: 3 (0.33)
--------------------------------------------------------------------------- Results:
Device: 0, device count: 3, average time / count: 473, average time on device: 157 Seconds (2 Minutes, 37 Seconds)
Device: 1, device count: 3, average time / count: 477, average time on device: 159 Seconds (2 Minutes, 39 Seconds)



The differences in my two cards can be explained by slight differences in the automatic "Boost Clock" performance (bought from the same store shelf within days of each-other).




I'll throw-in these fresh results, using V1.4 for consistency; 560Ti - reference cards running reference clocks, for .5 and .33, using x41g included with the tool:



27 September 2012 - 11:59:50 Start, devices: 2, device count: 2 (0.50)
--------------------------------------------------------------------------- Results:
Device: 0, device count: 2, average time / count: 273, average time on device: 136 Seconds (2 Minutes, 16 Seconds)
Device: 1, device count: 2, average time / count: 275, average time on device: 137 Seconds (2 Minutes, 17 Seconds)




27 September 2012 - 12:04:28 Start, devices: 2, device count: 3 (0.33)
--------------------------------------------------------------------------- Results:
Device: 0, device count: 3, average time / count: 405, average time on device: 135 Seconds (2 Minutes, 15 Seconds)
Device: 1, device count: 3, average time / count: 407, average time on device: 135 Seconds (2 Minutes, 15 Seconds)




As you can see, my 560Tis beat the times of my 660Tis, so I am especially anxious to get your 660 numbers for comparison.

Please for the sake of future research relocate this conversation to Freds thread...
http://setiathome.berkeley.edu/forum_thread.php?id=69136&sort_style=6&start=150

Thanks guys :)

---

Please for the sake of future research relocal this conversation to Freds thread...
http://setiathome.berkeley.edu/forum_thread.php?id=69136&sort_style=6&start=150

Thanks guys :)

If he answers here, I will repost it over there.

That thread is getting a little... difficult to find anything in.

I hear ya there :) seems to be the nature of these things...

---

As you can see, my 560Tis beat the times of my 660Tis....

Heh? WHAT? Why?

Is this the case in real-world crunching too? I totally expected the 660Ti to be around 30% better @seti than the 560Ti...

As you can see, my 560Tis beat the times of my 660Tis....

Heh? WHAT? Why?

Is this the case in real-world crunching too? I totally expected the 660Ti to be around 30% better @seti than the 560Ti...

I don't know. I haven't been able to keep work on the 660Ti cards since I got them.

But the test results using Fred's Performance tool are consistent from one test to the next, even when I moved the cards from one machine to another, so I suspect it's true.

There may be exceptions and "gotchas," but like I said, I haven't been able to develop a stable "RAC" on the 660Tis because I can't keep work on them.

Everything I read indicated that the Fermi cards (5xx series) are better at number crunching than Kepler cards (6xx series). But that may be just for Floating Point calcs? I am also interested in some real test results.

Btw. My GPUs have been starving for four days. How about you?

Cheers
Chris

And freds performace tool v1.4 run on one of them. More info on my blog. This is a Palit factory OC'ed one (Base clock 1006Mhz, Boost 1072Mhz). Details posted in the other thread as well.


Starting automatic test: (x41g)
29 September 2012 - 08:56:38 Start, devices: 1, device count: 1 (1.00)
--------------------------------------------------------------------------- Results:
Device: 0, device count: 1, average time / count: 234, average time on device: 234 Seconds (3 Minutes, 54 Seconds)
Next :---------------------------------------------------------------------------
29 September 2012 - 09:00:34 Start, devices: 1, device count: 2 (0.50)
--------------------------------------------------------------------------- Results:
Device: 0, device count: 2, average time / count: 319, average time on device: 159 Seconds (2 Minutes, 39 Seconds)
Next :---------------------------------------------------------------------------
29 September 2012 - 09:05:56 Start, devices: 1, device count: 3 (0.33)
--------------------------------------------------------------------------- Results:
Device: 0, device count: 3, average time / count: 465, average time on device: 155 Seconds (2 Minutes, 35 Seconds)
Next :---------------------------------------------------------------------------
29 September 2012 - 09:13:45 Start, devices: 1, device count: 4 (0.25)
--------------------------------------------------------------------------- Results:
Device: 0, device count: 4, average time / count: 619, average time on device: 154 Seconds (2 Minutes, 34 Seconds)
>> The best average time found: 155 Seconds (2 Minutes, 35 Seconds), with count: 0.33 (3)

---

BOINC blog

Everything I read indicated that the Fermi cards (5xx series) are better at number crunching than Kepler cards (6xx series). But that may be just for Floating Point calcs?

The GTX*** Kepler boards all have limited double precision support. As it's not used for Seti crunching, it doesn't hinder them in any way.
The beefed up double precision support will be coming out in the Kepler Tesla cards.

---

Grant
Darwin NT

And freds performace tool v1.4 run on one of them. More info on my blog.

Thank you.

I have been planning to get a GTX680 for one of my rigs, but after reading this post, I am having doubts. Does this mean that the 5xx series perform better than the higher priced 6xx ones? I havent come across any comparisons for the 680 or 690 cards. Would anyone have details?