Ok, several hours later I have some better data on these
2 at time is 1260 seconds (630 sec each)
3 at time is 1830 seconds (610 sec each)
4 at time is 2400 seconds (600 sec each)
Haven't tried them with any higher numbers, note temps of GPU are going up with each consecutive additional work unit.
Zalster
Edit
tolafoph, what are your temps on that 1070 when running 4 at a time?
Ty for your numbers, I am running my Gainward GTX1070 Phoenix GS at standard clock speeds. I think its a little OCed by Gainward. I looked for a cool card, do its also why less noisy than my old GTX580 which was like a hair dryer when I can CUDA taks or in games. The heatsink is 2.5 slots.
GPU-Z and the Gainward tool both give out 61°C to 63°C (about 145°F) when I run more than 1 task. I havent tried more than 4.
I bought the card for gaming, thats why I got a GTX1070 with 8GB of RAM and not a way cheaper GTX1060 with 3GB.
I just want to know if my card performs in the right ballpark figure.
I have a 1070 running 3x WUs. I haven't tried the new CUDA 55 version yet. They were completing 3x in about 44min which comes out to 880 seconds. The 1070 is running at 2.1ghz core and 9.4ghz mem. Take in the improvement in CUDA55 and you look good on those numbers,
I ran the CUDA 5.5 applications on both Windows 10 and Linux via 980 Ti Hybrid cards. With card frequencies of ~1400 MHz GPU core and 7905 MHz memory, this is what I am seeing after averaging available data:
FYI to all who are wondering why the app version page doesn't consider the beta versions: that's standard BOINC behavior and we still use the stock BOINC page for that purpose. It just shows the latest app version per app/platform pair, ignoring beta versions if there are any non-beta versions available. We're considering to change that though.
The Binary Radio Pulsar Search in Parkes data will not be extended.
Bernd, forgive my late response, I've been distracted with many other things :-( as I'm sure you have been too.
I have a question or few.
I assume we have sifted through all the 'of interest' frequencies in the Parkes data to hand, hence the fact this particular run on the data set has ended.
Will there be new data from Parkes (at a later date perhaps) or has it been deemed that information from Parkes does not fit the current project goals, for whatever reason? If so, why?
Basically, It intrigues me to know the scientific difference between the two telescopes radio data and to discover why the Parkes search was ended :-)
Is the project focus now solely on the GW search? and are there plans for a GPU app for that search?
Zalster wrote: Ok,
)
Ty for your numbers, I am running my Gainward GTX1070 Phoenix GS at standard clock speeds. I think its a little OCed by Gainward. I looked for a cool card, do its also why less noisy than my old GTX580 which was like a hair dryer when I can CUDA taks or in games. The heatsink is 2.5 slots.
GPU-Z and the Gainward tool both give out 61°C to 63°C (about 145°F) when I run more than 1 task. I havent tried more than 4.
I bought the card for gaming, thats why I got a GTX1070 with 8GB of RAM and not a way cheaper GTX1060 with 3GB.
I just want to know if my card performs in the right ballpark figure.
I have a 1070 running 3x WUs.
)
I have a 1070 running 3x WUs. I haven't tried the new CUDA 55 version yet. They were completing 3x in about 44min which comes out to 880 seconds. The 1070 is running at 2.1ghz core and 9.4ghz mem. Take in the improvement in CUDA55 and you look good on those numbers,
I ran the CUDA 5.5
)
I ran the CUDA 5.5 applications on both Windows 10 and Linux via 980 Ti Hybrid cards. With card frequencies of ~1400 MHz GPU core and 7905 MHz memory, this is what I am seeing after averaging available data:
Windows 10 - NV Driver 372.70
1x task - 677 seconds
2x task - 1284 seconds (642 seconds per task)
Linux 64-bit - NV Driver 370.28 (Beta)
1x task - 554 seconds
2x tasks - 1046 seconds (523 seconds per task)
I tried the CUDA 5.5
)
I tried the CUDA 5.5 applications on Windows 10 and get on my overclocked GTX 1070 FE following result:
3 tasks - exactly 30min
That is an improvement of 30% compared to the old application. The temperature increased by 3°C to 74°C.
Proud member of SETI.Germany
Would any of you have
)
Would any of you have measurements of the 1070 power consumption when crunching x2/x3/x4?
thanks!
Trotador wrote:Would any of
)
And x1 too ;-)
-----
Trotador wrote:Would any of
)
Not exactly what you want, and on an application for which new work is longer being issued, but you might review this post in the Pascal thread.
The post is harder to read than one would hope, as the current web site seems not to support the pre tag.
FYI to all who are wondering
)
FYI to all who are wondering why the app version page doesn't consider the beta versions: that's standard BOINC behavior and we still use the stock BOINC page for that purpose. It just shows the latest app version per app/platform pair, ignoring beta versions if there are any non-beta versions available. We're considering to change that though.
Best,
Oliver
Einstein@Home Project
Bernd Machenschalk
)
Bernd, forgive my late response, I've been distracted with many other things :-( as I'm sure you have been too.
I have a question or few.
I assume we have sifted through all the 'of interest' frequencies in the Parkes data to hand, hence the fact this particular run on the data set has ended.
Will there be new data from Parkes (at a later date perhaps) or has it been deemed that information from Parkes does not fit the current project goals, for whatever reason? If so, why?
Basically, It intrigues me to know the scientific difference between the two telescopes radio data and to discover why the Parkes search was ended :-)
Is the project focus now solely on the GW search? and are there plans for a GPU app for that search?
Oliver Bock wrote: We're
)
Thanks Oliver, would it be possible to include also
a) the name as described in the app_config.xml file e.g.
<name>einsteinbinary_BRP6</name>
b) and/or short code e.g. O1MD1TCV and O1MD1TG, which helps match up the columns on the server status page.