Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Project Staging Area :
duration of cuda wu
| Author |
Message |
|---|
|
|
|
hi, i am very frustrated about my oced 2 gtx 460s. they need around 27 minutes for PPS (Sieve) v1.39 (cuda23) WUs while running almost at 900 mhz and not even at default clocks.
in the table provided by scott here it should only take them 19 minutes.
that is a huge difference.
what exactly am i doing wrong?
i use windows 7 and my cards share 16 pci lanes (so 8x/8x) | |
|
|
|
|
8x is not 16x, so you can expect longer runtimes. Before the length was increased I needed around 24min / WU. | |
|
 Michael Goetz Volunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,244,369
RAC: 880,036
                               
|
hi, i am very frustrated about my oced 2 gtx 460s. they need around 27 minutes for PPS (Sieve) v1.39 (cuda23) WUs while running almost at 900 mhz and not even at default clocks.
in the table provided by scott here it should only take them 19 minutes.
that is a huge difference.
what exactly am i doing wrong?
i use windows 7 and my cards share 16 pci lanes (so 8x/8x)
Since Scott posted those numbers, the size of the WUs has increased by 50%, so you need to multiply his numbers by 1.5 to be relevant to the WUs you're seeing today.
That, alone, accounts for the difference in the numbers you're seeing.
I might have a suggestion that could make your card even faster. Maybe.
I have a 460 at stock clocks, and I noticed that the GeneferCUDA program got about a 10% speed improvement when I went from the 275.33 drivers to to 285.86 drivers.
I don't know if that will help you or not because GeneferCUDA uses double precision floating point math and also uses the CUDA FFT libraries. I'm not certain, but I think the sieves use integer arithmetic and don't use FFTs, so the programs are very, very different.
____________
My lucky number is 75898524288+1 | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,244,369
RAC: 880,036
|
8x is not 16x, so you can expect longer runtimes. Before the length was increased I needed around 24min / WU.
The thinking is that the bandwidth is more or less irrelevant for our purposes, and the GPUs could probably run in a 1x slot without a noticeable affect on performance.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
@ Michael Goetz how long do your gtx 460 take to complete tasks? i couldn't find any infos in your profile | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,244,369
RAC: 880,036
|
|
Hmmmm......
Those WUs aren't in the database anymore, so going by memory I think it was about 33 minutes with the 275.33 drivers. I have not tried PPS Sieve with the 285.86 drivers. Also, I sometimes have all the CPU cores crunching, which may sometimes slow down the GPU. With overclocking and a free CPU core you could go faster.
____________
My lucky number is 75898524288+1 | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,244,369
RAC: 880,036
|
|
At 27 minutes, you're running almost 20% faster than my GPU, which sounds about right for the overclocking.
I have heard that for this application it is not necessary to overclock the memory. Keep it at stock speed and reduce the power, heat, and noise with a negligible affect on speed.
____________
My lucky number is 75898524288+1 | |
|
Message boards :
Project Staging Area :
duration of cuda wu |
