I don't see FGRP work on my 9173342 host. I have WinXP x32 and 2 NVidia GPUs there, GT640 and 250GTS. Now I don't see any GPU work there. Is there something happened?
I don't think that FGRP GPU will be available for 32Bit systems. The FFT library that we are using (clFFT, originally from AMD, now open source) doesn't even compile for 32Bit. If at some point we get bored we could port the application to CUDA, and then possibly build a 32Bit version for NVidia only, but I strongly doubt that this will happen.
It looks like we do have the remaining problems with the "full GPU" Windows version fixed (thanks again, Christophe!). Releasing the new version (1.16) will take a bit, but I already deprecated the 1.15 version to avoid more tasks being sent out with that.
On a Fury X GPU usage jumps from 0-100% every other second, but the times have dropped from 19k sec to ~330 s. One Wu already validated. Great job!
On a Tesla K20 after initial high-CPU startup, the GPU is utilized at 90-100% most of the time, but the entire task seems to take too long (currently 21m and still only 22%). Also the CPU usage is still very high (>99%), so this looks quite odd. This GPU should support FP64 very well, so any idea why is this ? Perhaps crippled OpenCL support on NV ?
EDIT: Tesla has finished the 1st task, it took 1500 s. That's quite high I think.
EDIT2: HD 7950, GPU usage ~50% most of the time, CPU usage ~60%. WU finished in 460 s.
1.16 is a huge improvement on Windows. My times have gone down to under 6 minutes per task. The progress bar moves along slowly until it reaches a little over 13% and then jumps to 100%.
This is running 1 task at a time. I'll start trying 2 and see what that does.
EDIT: Running two of these at once has a very nice effect. Times go from around 5 min 50 sec running one task to around 8 min running two tasks.
1.16 is a huge improvement on Windows. My times have gone down to under 6 minutes per task. The progress bar moves along slowly until it reaches a little over 13% and then jumps to 100%.
This is running 1 task at a time. I'll start trying 2 and see what that does.
EDIT: Running two of these at once has a very nice effect. Times go from around 5 min 50 sec running one task to around 8 min running two tasks.
Matt,
What GPU are you using? Mine, (in my XP Pro x64 system), is an EVGA GTX-760. I run two Units at a time on all Unit Types. BRP4G currently takes about 40 Min per Unit. I haven't gotten to the new 1.16 Units, yet.
I don't see FGRP work on my
)
I don't see FGRP work on my 9173342 host. I have WinXP x32 and 2 NVidia GPUs there, GT640 and 250GTS. Now I don't see any GPU work there. Is there something happened?
FGRP on GPUs was released for
)
FGRP on GPUs was released for x64 systems only.
I'm too wondering whether there will be an x86 version, as there are still plenty such systems.
-----
I don't think that FGRP GPU
)
I don't think that FGRP GPU will be available for 32Bit systems. The FFT library that we are using (clFFT, originally from AMD, now open source) doesn't even compile for 32Bit. If at some point we get bored we could port the application to CUDA, and then possibly build a 32Bit version for NVidia only, but I strongly doubt that this will happen.
BM
It looks like we do have the
)
It looks like we do have the remaining problems with the "full GPU" Windows version fixed (thanks again, Christophe!). Releasing the new version (1.16) will take a bit, but I already deprecated the 1.15 version to avoid more tasks being sent out with that.
BM
It's fully utilize my GPU and
)
It's fully utilize my GPU and everything is laggy such as browsing (while scrolling a page) etc.
See picture here for specs and utilization.
How can I deal with it?
Running 1.16.On a Fury X
)
Running 1.16.
On a Fury X GPU usage jumps from 0-100% every other second, but the times have dropped from 19k sec to ~330 s. One Wu already validated. Great job!
On a Tesla K20 after initial high-CPU startup, the GPU is utilized at 90-100% most of the time, but the entire task seems to take too long (currently 21m and still only 22%). Also the CPU usage is still very high (>99%), so this looks quite odd. This GPU should support FP64 very well, so any idea why is this ? Perhaps crippled OpenCL support on NV ?
EDIT: Tesla has finished the 1st task, it took 1500 s. That's quite high I think.
EDIT2: HD 7950, GPU usage ~50% most of the time, CPU usage ~60%. WU finished in 460 s.
-----
[Update:] Final time at 100%
)
[Update:]
Final time at 100% on the two 1.15 Units on Win XP Pro x64: 6 Hours, 6 Min. and 18 Sec.
I have just aborted all the other 1.15 Units. (300+ of them...)
Just picked up two 1.16 Units; after finishing the BRP4G Units some time late morning today, the 1.16 Units should start.
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
One more result: GTX 1050 Ti
)
One more result: GTX 1050 Ti running 1.16 took 730 s.
There must be something odd with the Tesla K20, which takes 1500 s.
-----
1.16 is a huge improvement on
)
1.16 is a huge improvement on Windows. My times have gone down to under 6 minutes per task. The progress bar moves along slowly until it reaches a little over 13% and then jumps to 100%.
This is running 1 task at a time. I'll start trying 2 and see what that does.
EDIT: Running two of these at once has a very nice effect. Times go from around 5 min 50 sec running one task to around 8 min running two tasks.
Matt_145 wrote:1.16 is a huge
)
Matt,
What GPU are you using? Mine, (in my XP Pro x64 system), is an EVGA GTX-760. I run two Units at a time on all Unit Types. BRP4G currently takes about 40 Min per Unit. I haven't gotten to the new 1.16 Units, yet.
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees