Making a thread for GTX 650 owners to post their results and experiences, and for general information on running the GTX 650 on this project.

I have ordered an EVGA Superclocked GTX 650 1GB today, it should arrive by Wednesday the 19th. Will try out both short runs and long runs and report results.

Anybody get the 650 yet? I know it just released though.


____________
XtremeSystems.org - #1 Team in GPUGrid

Got this installed next to a GTX 660 Ti.

The GTX 650 has started on its first work unit and preliminary estimates are looking around 17-21 hours on a long task (one of the new PAOLA tasks, GPU usage 98-99% on winxp64, driver 306.23)

First long run took 18.4 hours. http://www.gpugrid.net/result.php?resultid=5864445 Started another one!

Not bad for such a little card but it took more than twice as long as the 660Ti at 2/3 the price?
____________
Thanks - Steve

Yes, my numbers were off so I looked them up before making this post. Based on New Egg pricing from a few days ago the card EVGA Superclock card
http://www.newegg.com/Product/Product.aspx?Item=N82E16814130831
which is the card that werdwerdus posted results for, here in the US was 1/2 the cost of the 660Ti I was comparing to.

If you go with a lower cost variant and could overclock it as well as the factory OC card it would be a better deal and edges towards being slightly better in performance per initial cost. What I have no information on is performance per watt.


____________
Thanks - Steve

Yes, my numbers were off so I looked them up before making this post. Based on New Egg pricing from a few days ago the card EVGA Superclock card
http://www.newegg.com/Product/Product.aspx?Item=N82E16814130831
which is the card that werdwerdus posted results for, here in the US was 1/2 the cost of the 660Ti I was comparing to.

If you go with a lower cost variant and could overclock it as well as the factory OC card it would be a better deal and edges towards being slightly better in performance per initial cost. What I have no information on is performance per watt.

Hey. Is gtx 650 faster in gpugrid.net tasks than GTX 260 (216 sp)? I understand that the gtx 650 card has a better performance / watt ratio.

I am thinking to upgrade the gtx 260 video card to gtx 650 graphics card, and I keep thinking which of these is better raw performance in gpugrid.net tasks.

The 650 will outperform the 260 substantially!
(roughly 1.5x to 2x the perfomance)

It can return all of the long WU types within the 24 hr. bouns window but I doubt a 260 could even return the smallest of the large types.
____________
Thanks - Steve

The 650 will outperform the 260 substantially!
(roughly 1.5x to 2x the perfomance)

It can return all of the long WU types within the 24 hr. bouns window but I doubt a 260 could even return the smallest of the large types.

Pretty sure ... using a NATHAN_N variety on WinXP x64, we can see it took just about 17.75 hours to complete. When I was running my GTX295 a few months ago it was hard pressed to finish any long WU in under 24 hours and when the NATHAN_N and original PAOLAs were released it would not finish them within 24 hour bonus and I got tired of tasks swapping so removed it from the project. The individual cards in the 295 are faster than a 260. The second example is one that werdwerdus provided a few posts up ... this is for one of the PAOLA_ARP models that runs really long but finished in 18.4 hours. In my 1.5x - 2x estimation I'm not even taking into account the time bonus, I'm just talking straight up runtime.

http://www.gpugrid.net/result.php?resultid=5879977
http://www.gpugrid.net/result.php?resultid=5864445

I did try searching around a little bit but didn't find any actual 260 results for LONG WUs (I could only find standards). If someone runing a 260 and has actual LONG WUs we could compare precisely (please post w/OS also).
____________
Thanks - Steve

Pretty sure ... using a NATHAN_N variety on WinXP x64, we can see it took just about 17.75 hours to complete. When I was running my GTX295 a few months ago it was hard pressed to finish any long WU in under 24 hours and when the NATHAN_N and original PAOLAs were released it would not finish them within 24 hour bonus and I got tired of tasks swapping so removed it from the project. The individual cards in the 295 are faster than a 260. The second example is one that werdwerdus provided a few posts up ... this is for one of the PAOLA_ARP models that runs really long but finished in 18.4 hours. In my 1.5x - 2x estimation I'm not even taking into account the time bonus, I'm just talking straight up runtime.

http://www.gpugrid.net/result.php?resultid=5879977
http://www.gpugrid.net/result.php?resultid=5864445

I did try searching around a little bit but didn't find any actual 260 results for LONG WUs (I could only find standards). If someone runing a 260 and has actual LONG WUs we could compare precisely (please post w/OS also).

So we can compare similar PAOLA_ARP WUs between 650 and 260
There are a couple of things to keep in mind that will (perhaps) skew the results a bit ...

- The 650 is running under WinXP which generally gets higher utilization compatred to Win7
- The 260 is running at a slightly higher CPU frequency which will help some
- The CUDA 4.2 app is more efficient that the 3.1
- I'm guessing here but because of the CPU models and the fact that both rigs are running multiple cards I'm guessing they are both at PCIE2 x8
- There is some variance between WUs of the same family but the runtimes are considerably less that 5% difference on the same machine.

The 650 = 66,275.70 seconds or 18.41 hrs
The 260 = 130,799.94 seconds or 36.33 hrs

hey lookey here ... you can't get much closer to 2x performance, and that is just straight up runtime.

The 650 also returned well within the runtime bonus timeframe so if you calculate the points per second the ratio goes to 3x better for the 650 @1.61 vs 260 @.53

Less electricity required, much better performance (and I'm guessing) quieter ... the 650 is heads above the 260 and all things considered a pretty nice card. All this is not a surprise as it is multiple generations newer.

Happy crunching,
Steve

So we can compare similar PAOLA_ARP WUs between 650 and 260
There are a couple of things to keep in mind that will (perhaps) skew the results a bit ...

- The 650 is running under WinXP which generally gets higher utilization compatred to Win7
- The 260 is running at a slightly higher CPU frequency which will help some
- The CUDA 4.2 app is more efficient that the 3.1
- I'm guessing here but because of the CPU models and the fact that both rigs are running multiple cards I'm guessing they are both at PCIE2 x8
- There is some variance between WUs of the same family but the runtimes are considerably less that 5% difference on the same machine.

The 650 = 66,275.70 seconds or 18.41 hrs
The 260 = 130,799.94 seconds or 36.33 hrs

hey lookey here ... you can't get much closer to 2x performance, and that is just straight up runtime.

The 650 also returned well within the runtime bonus timeframe so if you calculate the points per second the ratio goes to 3x better for the 650 @1.61 vs 260 @.53

Less electricity required, much better performance (and I'm guessing) quieter ... the 650 is heads above the 260 and all things considered a pretty nice card. All this is not a surprise as it is multiple generations newer.

Happy crunching,
Steve

Are you sure?

216 Shaders at ~1.3 GHz versus 256 Shaders (can't use the super-scalar ones at GPU-Grid) at 1.1 GHz is not much of a difference, performance-wise. Running the CUDA 4.2 app versus the 3.1 app will help a lot, though. And power efficiency is a complete homerun, of course.

If you mean the 650ti is "slightly better" than the 650 as in having twice as many cores for $30 more then you are correct.

My 650ti is now posting slightly better crunching averages than my 560 and has not peaked yet.