I tested various app_info.xml to run ABP2 CUDA and Seti CUDA concurrently three months ago. I was succesful at Windows (4CPu Xeon + GTX260 on XP 64bit, 2CPU C2D + 9600GT on XP 32bit) and Ubuntu (2CPU C2D + 9600GT on 9.04 64bit), but the things goes best when running ABP2 CUDA + Folding GPU3 on Win XP (2CPU C2D + 9600GT) using no app_info.xml. That is my experience.
Collatz, MW and DNETC I haven't tried as it is territory of ATI stream apps.
Hi Nenym,
happy to hear that! Could you please post your work here? I'm shure someone could adapt that to get it working.
BTW: MW has cuda-apps as well. And currently they are working on OpenCL-apps.
BTW: MW has cuda-apps as well. And currently they are working on OpenCL-apps.
Sure, but the hardware s*cks at that. GT200 chips can only use 12.5% of their single precision performance there, whereas ATIs can use 40$. GF100 improves this to 50%, but is artificially castrated to 12.5% by nVidia in all cards but the Quadros. For GF104 it should be 16.7% (not totally sure).
And since we've got enough ATIs to satisfy MW (and don't have any better project than MW for them, if one's not a DNET fan), running MW on nVidias (or CPUs) is inefficient, as their power could be used better elsewhere.
[/OT]
As I can remember, I had to add some folders to .conf file and run ldconfig or to make some symbolic links manually, ldd says everything.
App_info.xml Seti (Cruncher's 3.0 version http://calbe.dw70.de/mb/viewtopic.php?f=9&t=120) - one app version part
setiathome_enhanced
606
cuda
0.040000
0.040000
CUDA
0.89
setiathome-CUDA_3.0_6.09.x86_64_vlarkill
setiathome_enhanced
608
cuda
0.040000
0.040000
CUDA
0.89
setiathome-CUDA_3.0_6.09.x86_64_vlarkill
To hold only one Einstein task active I have modified trigggl's script. Script is not clear but works fine. Running more Einstein tasks concurently was not effective. It seems running one Einstein + one Seti/Setibeta task is better.
#!/bin/sh
cd /home/hajek/BOINC
until [ $(./boinccmd --get_results | grep -c Einstein) == 0 ] ; do
if [ !$(./boinccmd --get_results | grep Einstein -A 10 | grep -c active.state..1) ] ; then
TASK=$(./boinccmd --get_results | grep Einstein -B 2 -A 10 | grep GUI..yes -B 11 | grep "name" | head -n 1)
task=$(echo ${TASK#*:})
./boinccmd --result http://einstein.phys.uwm.edu/ $task resume
fi
until [ $(./boinccmd --get_results | grep Einstein -A 10 | grep report..no -A 7 | grep -c GUI..no) -lt 2 ] ; do
TASK=$(./boinccmd --get_results | grep Einstein -B 2 -A 10 | grep GUI..no -B 11 | grep " name" | tail -n 1)
task=$(echo ${TASK#*:})
./boinccmd --result http://einstein.phys.uwm.edu/ $task suspend
done
sleep 60
done
I had to copy and rename libs xxxx32_23.dll to xxx.dll, do not understand why.
Seti's app_info is similar to Linux version, I am running vlarkill V12 modified by Raimster using CUDA 23 libs. I have used Lunatic's installer http://lunatics.kwsn.net/index.php and changed only 1 to 0.89 in CUDA part.
As I do not know how to modify Linux script to VBA or something similar to be executable on Windows, I had to manage Einstein tasks via GUI manually to hold only one Einstein task active.
Although it works, I am running Einstein CUDA + Folding on Windows not using any app_info.xml as I have mentioned above.
Sorry for my English I hope it is understandable.
Hey guys, I have been reading this post and have changed the 'COUNT' parm in the client_state.xml member to .22 and even to .33 on my GTX260-216 with 896m of memory. I can get two ABP2 CUDA23 task running but all others have a status of waiting on memory. GPUZ shows less that 400m being used. SO I can't seem to get a 3rd ABP2 CUDA WU to start? My CPU is a AMD 940 QUAD, so I have cores to spare. Is the 419430400.000000 parm right after the COUNT parm in the client_state.xml limiting me to this value instead of the 896m that I actually have on the card. I also have a GTX470 Fermi with 1280m that has the same value and has the same 2 WU limit (waiting on memory). Do I need to either reset the project or even detach and then reattach to Einstein so that my card gets properly set??
Boinc 6.10.58
4gig memory
WIN XP 32
AMD QUAD 940 and 955
Hey guys, I have been reading this post and have changed the 'COUNT' parm in the client_state.xml member to .22 and even to .33 on my GTX260-216 with 896m of memory. I can get two ABP2 CUDA23 task running but all others have a status of waiting on memory. GPUZ shows less that 400m being used. SO I can't seem to get a 3rd ABP2 CUDA WU to start? My CPU is a AMD 940 QUAD, so I have cores to spare. Is the 419430400.000000 parm right after the COUNT parm in the client_state.xml limiting me to this value instead of the 896m that I actually have on the card. I also have a GTX470 Fermi with 1280m that has the same value and has the same 2 WU limit (waiting on memory). Do I need to either reset the project or even detach and then reattach to Einstein so that my card gets properly set??
Boinc 6.10.58
4gig memory
WIN XP 32
AMD QUAD 940 and 955
Appreciate the help and info on this board.
Hi,
your computer is hidden so nobody can take a look on that.
The message 'waiting for mem' appears, when no more main mem is available. You have to subtract the video-ram from the 4GB.
This happend on one of my systems (with 2 GB Ram) when I tried to run 4 Cosmology@home. And it is depending on the other apps currently running.
In addition please check your settings regarding mem-usage in your boinc-manager.
Alexander, HMMMM... I will check main memory, but I think I have enough free at the time even though WIN 32 can only use 3.5 of my 4g main. I will also check Boinc memory prev as well. I figured that this message was that I ran out of memory on the CUDA. Also I am unhidden now.
I figured that this message was that I ran out of memory on the CUDA.
Yes, that's correct. BOINC checks for main memory usage and GPU memory usage separately. Running out of main memory while you still have enough left on the GPU, will not stop the GPU tasks. Vice versa, when you run out of memory on the GPU, only those tasks will stop, tasks running in main memory will happily continue.
Most GPU memory related messages and problems happen when some task is stuck in the video RAM. Only a full power cycle (reboot) can get rid of those.
Ageless, Thanks for the "stuck in video ram" suggestion. As I rechecked my main memory I see I have 2.4m free, and GPUZ shows the gpu at 18% used and 287m used. I was just going by the exact status on the waiting WU's (waiting for gpu memory) to mean just that, and with my card having 895m I knew I was not pushing that limit either. I will totally reboot and try again and report. The only other thing I was thinking about trying was to update my nvidia driver from 190.38 to the latest 258.96.
RE: I tested various
)
Hi Nenym,
happy to hear that! Could you please post your work here? I'm shure someone could adapt that to get it working.
BTW: MW has cuda-apps as well. And currently they are working on OpenCL-apps.
Kind regards,
Alexander
RE: BTW: MW has cuda-apps
)
Sure, but the hardware s*cks at that. GT200 chips can only use 12.5% of their single precision performance there, whereas ATIs can use 40$. GF100 improves this to 50%, but is artificially castrated to 12.5% by nVidia in all cards but the Quadros. For GF104 it should be 16.7% (not totally sure).
And since we've got enough ATIs to satisfy MW (and don't have any better project than MW for them, if one's not a DNET fan), running MW on nVidias (or CPUs) is inefficient, as their power could be used better elsewhere.
[/OT]
MrS
Scanning for our furry friends since Jan 2002
RE: BTW: MW has cuda-apps
)
I know that, but by my opinion it is better to use an ATI GPU than a CUDA GPU for these projects and CUDA use for GPUGRID or Seti now.
A. Linux Einstein +
)
A. Linux Einstein + Seti/SetiBeta
App_info.xml Einstein
einsteinbinary_ABP2
einsteinbinary_ABP2_1.11_i686-pc-linux-gnu__ABP2cuda23
libcudart32_23.so
libcufft32_23.so
einsteinbinary_ABP2
111
6.7.0
cuda
1.000000
1.000000
CUDA
0.1
einsteinbinary_ABP2_1.11_i686-pc-linux-gnu__ABP2cuda23
libcudart32_23.so
libcufft32_23.so
As I can remember, I had to add some folders to .conf file and run ldconfig or to make some symbolic links manually, ldd says everything.
App_info.xml Seti (Cruncher's 3.0 version http://calbe.dw70.de/mb/viewtopic.php?f=9&t=120) - one app version part
CUDA
0.89
setiathome-CUDA_3.0_6.09.x86_64_vlarkill
setiathome_enhanced
608
cuda
0.040000
0.040000
CUDA
0.89
setiathome-CUDA_3.0_6.09.x86_64_vlarkill
To hold only one Einstein task active I have modified trigggl's script. Script is not clear but works fine. Running more Einstein tasks concurently was not effective. It seems running one Einstein + one Seti/Setibeta task is better.
#!/bin/sh cd /home/hajek/BOINC until [ $(./boinccmd --get_results | grep -c Einstein) == 0 ] ; do if [ !$(./boinccmd --get_results | grep Einstein -A 10 | grep -c active.state..1) ] ; then TASK=$(./boinccmd --get_results | grep Einstein -B 2 -A 10 | grep GUI..yes -B 11 | grep "name" | head -n 1) task=$(echo ${TASK#*:}) ./boinccmd --result http://einstein.phys.uwm.edu/ $task resume fi until [ $(./boinccmd --get_results | grep Einstein -A 10 | grep report..no -A 7 | grep -c GUI..no) -lt 2 ] ; do TASK=$(./boinccmd --get_results | grep Einstein -B 2 -A 10 | grep GUI..no -B 11 | grep " name" | tail -n 1) task=$(echo ${TASK#*:}) ./boinccmd --result http://einstein.phys.uwm.edu/ $task suspend done sleep 60 doneHere's my inspiration http://www.gpugrid.net/forum_thread.php?id=2243&nowrap=true#18285, trigggl is the great wizard.
B. Windows Einstein + Seti/SetiBeta
App_info.xml Einstein
einsteinbinary_ABP2
einsteinbinary_ABP2_3.11_windows_intelx86__ABP2cuda23.exe
cudart32_23.dll
cufft32_23.dll
einsteinbinary_ABP2
311
6.7.0
cuda
1.0
1.0
CUDA
0.1
einsteinbinary_ABP2_3.11_windows_intelx86__ABP2cuda23.exe
cudart32_23.dll
cufft32_23.dll
I had to copy and rename libs xxxx32_23.dll to xxx.dll, do not understand why.
Seti's app_info is similar to Linux version, I am running vlarkill V12 modified by Raimster using CUDA 23 libs. I have used Lunatic's installer http://lunatics.kwsn.net/index.php and changed only 1 to 0.89 in CUDA part.
As I do not know how to modify Linux script to VBA or something similar to be executable on Windows, I had to manage Einstein tasks via GUI manually to hold only one Einstein task active.
Although it works, I am running Einstein CUDA + Folding on Windows not using any app_info.xml as I have mentioned above.
Sorry for my English I hope it is understandable.
RE: Sorry for my English I
)
Thank you very very much!
And don't worry about your english!
Kind regards,
Alexander
Hey guys, I have been
)
Hey guys, I have been reading this post and have changed the 'COUNT' parm in the client_state.xml member to .22 and even to .33 on my GTX260-216 with 896m of memory. I can get two ABP2 CUDA23 task running but all others have a status of waiting on memory. GPUZ shows less that 400m being used. SO I can't seem to get a 3rd ABP2 CUDA WU to start? My CPU is a AMD 940 QUAD, so I have cores to spare. Is the 419430400.000000 parm right after the COUNT parm in the client_state.xml limiting me to this value instead of the 896m that I actually have on the card. I also have a GTX470 Fermi with 1280m that has the same value and has the same 2 WU limit (waiting on memory). Do I need to either reset the project or even detach and then reattach to Einstein so that my card gets properly set??
Boinc 6.10.58
4gig memory
WIN XP 32
AMD QUAD 940 and 955
Appreciate the help and info on this board.
RE: Hey guys, I have been
)
Hi,
your computer is hidden so nobody can take a look on that.
The message 'waiting for mem' appears, when no more main mem is available. You have to subtract the video-ram from the 4GB.
This happend on one of my systems (with 2 GB Ram) when I tried to run 4 Cosmology@home. And it is depending on the other apps currently running.
In addition please check your settings regarding mem-usage in your boinc-manager.
Regards
Alexander
Alexander, HMMMM... I will
)
Alexander, HMMMM... I will check main memory, but I think I have enough free at the time even though WIN 32 can only use 3.5 of my 4g main. I will also check Boinc memory prev as well. I figured that this message was that I ran out of memory on the CUDA. Also I am unhidden now.
Thanks for your help and suggestions.
RE: I figured that this
)
Yes, that's correct. BOINC checks for main memory usage and GPU memory usage separately. Running out of main memory while you still have enough left on the GPU, will not stop the GPU tasks. Vice versa, when you run out of memory on the GPU, only those tasks will stop, tasks running in main memory will happily continue.
Most GPU memory related messages and problems happen when some task is stuck in the video RAM. Only a full power cycle (reboot) can get rid of those.
Ageless, Thanks for the
)
Ageless, Thanks for the "stuck in video ram" suggestion. As I rechecked my main memory I see I have 2.4m free, and GPUZ shows the gpu at 18% used and 287m used. I was just going by the exact status on the waiting WU's (waiting for gpu memory) to mean just that, and with my card having 895m I knew I was not pushing that limit either. I will totally reboot and try again and report. The only other thing I was thinking about trying was to update my nvidia driver from 190.38 to the latest 258.96.
Appreciate the suggestion and help.
Regards Mike