Has anybody looked at this guys tasks? According to Berkley, his Intel® Core™2 Quad CPU Q9450 @ 2.66GHz [x86 Family 6 Model 23 Stepping 7] completes work units that take an average of 1300-1400 seconds on my i7 in about 50 seconds.
Can anybody possibly explain this? I’m guessing it is a bug, but his top 2 computers run about the same, and they are #2 and #3 in the world.
FYI: The number 1 computer has 48 CPUs.
mr.kjellen!!!
Share with your brothers please?:sigh:
I have an idea that he has a custom application or something because his clocks and benchmark scores are nothing special.:worship:
[QUOTE=DoubleTop;432824]I’m sure Anton will be willing to let you know - pretty sure it’s just the optimised app and some serious overclock
DT.[/QUOTE]
I know Berkley can’t be reporting the times right or he would have a RAC of 139,000 or so on each system. It would be interesting to find out why the times are wrong. I’ve done some very serious overclocking myself and reporting programs of many kinds have problems with clocks that they do not understand.
Anton…so what is the actual time on those WU’s? I’m guessing 400-700 seconds?:jealous:
Yup, that CUDA app sure is an impressive thing… gotta love it.
The Q9450 IS oc’d by a bit (3.5GHz). It runs AP units only. But as pointed out the real deal is the lone GTX280 that scores almost as much as the quad on it’s own.
I’m impressed anyhow. So just couldn’t help myself. Got a core i7 that’s on the way with two GTX295s
If you go to the tasks page , select a completed WU and click on the task id column you will get the details of the device that crunched the WU , in these cases you get this sort of detail
stderr out
<core_client_version>6.5.0</core_client_version>
<![CDATA[
<stderr_txt>
setiathome_CUDA: Found 1 CUDA device(s):
Device 1 : GeForce GTX 280
totalGlobalMem = 1073741824
sharedMemPerBlock = 16384
regsPerBlock = 16384
warpSize = 32
memPitch = 262144
maxThreadsPerBlock = 512
clockRate = 1296000
totalConstMem = 65536
major = 1
minor = 3
textureAlignment = 256
deviceOverlap = 0
multiProcessorCount = 30
setiathome_CUDA: CUDA Device 1 specified, checking…
Device 1: GeForce GTX 280 is okay
SETI@home using CUDA accelerated device GeForce GTX 280
Rise priority modification by Raistmer based on rev380 of SETI@home sources
Priority of worker thread rised successfully
setiathome_enhanced 6.02 Visual Studio/Microsoft C++
libboinc: 6.3.22
[QUOTE=mr.kjellen;432834]Yup, that CUDA app sure is an impressive thing… gotta love it.
The Q9450 IS oc’d by a bit (3.5GHz). It runs AP units only. But as pointed out the real deal is the lone GTX280 that scores almost as much as the quad on it’s own.
I’m impressed anyhow. So just couldn’t help myself. Got a core i7 that’s on the way with two GTX295s
/Anton[/QUOTE]
How do I get the CUDA app? Does it replace my optimized app or is it in addition to it?
[QUOTE=RealPhenom;432851]How do I get the CUDA app? Does it replace my optimized app or is it in addition to it?
Thanks for your help.[/QUOTE]
You need an nvidia card which supports Cuda, together with updated drivers that support cuda. After that, check boinc recognises the card. In messages check for something like:-
22/01/2009 18:24:05||CUDA devices found
22/01/2009 18:24:05||Coprocessor: GeForce 9800 GT (1)
After that if you have changed nothing else you will just need to detach and reattach to seti. You will lose all downloaded units at this point so make sure you update first.
As long as you’re using the optimized app the CUDA app won’t download.
There is of course an optimized CUDA for you to download as well, Raistmer has been hard at work to make an optimized AP/CUDA combo, all is described in this thread over at the SETI forums.
Be aware that as long as you install BOINC as a service under vista it will not work with CUDA (NVidia/Vista driver limitation).
So in other words, when running BOINC as a service under Win Vista, you can’t run CUDA. Is that correct? If so, then I may as well keep the setup I currently am running on my quad as it is the only machine I have that’s got an NVidia card that supports CUDA.
[QUOTE=mr.kjellen;432918]As long as you’re using the optimized app the CUDA app won’t download.
There is of course an optimized CUDA for you to download as well, Raistmer has been hard at work to make an optimized AP/CUDA combo, all is described in this thread over at the SETI forums.
Be aware that as long as you install BOINC as a service under vista it will not work with CUDA (NVidia/Vista driver limitation).
edit: Gandelfs points are valid as well.
/Anton[/QUOTE]
I played with CUDA yesterday and all I did was lose 16 hours of production. About Intel® Core™2 Quad CPU Q9450 @ 2.66GHz [x86 Family 6 Model 23 Stepping 7]…Can you please tell us the following:
Also, what are your seti settings? What types of WU’s are you allowing to run? Are you running AP units as well as the normal? Do you have 5 jobs running at once? 4+1? I don’t understand how it’s supposed to work.
Please give any other info that is important for running CUDA successfully? You seem to be doing it better than others I’ve seen.:worship:
