Seti@Home optimized science apps and information
Optimized Seti@Home apps => Windows => GPU crunching => Topic started by: Frizz on 15 Sep 2010, 04:01:30 pm
-
I just got a GTS 450 (MSI Cyclone) to do some testing and wanted to share some benchmark results with you.
Summarized: It's nothing world-shattering. My GT 240 crunches one typical WU in about 20 minutes, my GTX 460 needs 11 minutes. The GTS 450 needs 16 minutes. Power consumption while crunching: ~70 watt (only for the GTS 450).
But the MSI GPU cooler is awesome - even under load it's never over 35°C.
Running app : Lunatics_x32f_win32_cuda30_preview.exe with -verb -st
with WU : 04jn10ab.1063.3339.11.10.97.wu
Started at : 21:15:06.046
Ended at : 21:30:54.296
948.219 secs Elapsed
45.109 secs CPU time
------------
Running app : Lunatics_x32f_win32_cuda30_preview.exe with -verb -st
with WU : 31ja10aa.18666.17828.14.10.107.wu
Started at : 21:30:54.359
Ended at : 21:46:44.375
949.984 secs Elapsed
45.813 secs CPU time
------------
Running app : Lunatics_x32f_win32_cuda30_preview.exe with -verb -st
with WU : PG0009.wu
Started at : 21:46:44.437
Ended at : 21:50:03.562
199.094 secs Elapsed
12.438 secs CPU time
------------
Running app : Lunatics_x32f_win32_cuda30_preview.exe with -verb -st
with WU : PG0395.wu
Started at : 21:50:03.625
Ended at : 21:51:14.296
70.641 secs Elapsed
12.844 secs CPU time
------------
Running app : Lunatics_x32f_win32_cuda30_preview.exe with -verb -st
with WU : PG0444.wu
Started at : 21:51:14.343
Ended at : 21:52:15.812
61.438 secs Elapsed
12.766 secs CPU time
------------
Running app : Lunatics_x32f_win32_cuda30_preview.exe with -verb -st
with WU : PG1327.wu
Started at : 21:52:15.875
Ended at : 21:52:59.765
43.875 secs Elapsed
12.906 secs CPU time
[EDIT] Added results of GT 240 benchmark for comparison plus WUs.
-
Have you got any comparisons with other cards and these wu's?
Depending on price it may not be a bad entry level fermi card
-
Looks like I will be able to see some decent SETI speeds about Monday, that should be when my GTX 460 arrives.
-
Sorry if this was discussed before, but I didn't wont to open another thread for GTS 450.
If I test it with cuda 2.3 GTS 450 throws invalid results, so it looks like she is not backward compability, or that is only temporary?
Thanks for any answer.
I have same card as member that open this thread.
-
Hi Pepi. Cuda 2.3 has no support for Fermi architecture. Only Cuda 3.0 onwards.
Your choices are stock 610 or x32f from the latest installer 0.37 . x32f is better than stock (less errors & a tad faster)
-
From CUDA 3.1 log
* Adds support for CUDA Toolkit 3.1 which includes significant performance increases for double precision math operations
It look like a boost when CUDA 2.3 was released, and she was way faster then 2.2 :)
I was testing Lunatics verse 6.10 stock, and Lunatics was only faster 10 seconds ( on 364 seconds result)
So hope that 3.1 or even 3.2 and more optimized application from Lunatics will make this difference larger .
Also must notice one thing: 240 has only 96 cores and 450 has 192. But 450 is not double faster then 240. And has also much more clocks of shaders, memory and gpu. Why?
-
Boost not for SETI cause it's single float, not double
-
And second one?
....240 has only 96 cores and 450 has 192. But 450 is not double faster then 240.....
-
Also, Cuda 3.1 has bugs (when running Multi-GPU) that makes it not useful here. Cuda 3.2 fixes these
but is not released to the public yet. (Guess it is)
-
And second one?
....240 has only 96 cores and 450 has 192. But 450 is not double faster then 240.....
Also Fermi optimisation has only just begun. There is a long way to go. For now the focus has been on getting a build that runs with less errors.
-
So hope is not vanished for GTS 450? :)
That is great news!
-
So hope is not vanished for GTS 450? :)
That is great news!
It's a brand new card/chip that isn't even in the Cuda documentation yet. I can't predict what the final capability will be, but it should be more than what we get now ;)
-
So all I need is waiting :) And hope that SETI will live enough time :)