No, not with current Apps. Actually so far we are quite happy we got it working at all.
Quote:
Any chance that an app_info.xml example can be posted here on the forum?
For now I wouldn't recommend to run Apps on anonymous platform here. We may need to release and deprecate apps at a pace that manual updates won't be able to keep up with.
We may ask you to try some manual tuning later, but only after the development has settled a bit.
Yesterday I ran the app 1.19 (atiOpenCL). My computer is AMD Phenom X4 925. Two ATI HD4850 1GB. Software is Windows 7 Ult x64. Boinc 6.13.12. AMD Driver 11.11 Screenshot
It seemed to me that the GPU load is very low. I looked at the Gigabyte HUD program and saw that the load on the GPU is ~ 40%. And progress of WU increased by 0,002% in the second.
I tried the following method. Priority of the process - real time. Screenshot 1
GPU load was ~ 80%. And progress of WU increased by 0,003%-0.004$ in the second.
Then I returned the priority - below average. GPU load returned to ~ 40%. Progress of WU increased by 0,002% in the second. Screenshot 2
Then I again tried real time priority.The result was repeated. GPU load ~ 80%.
And progress of WU increased by 0,003%-0.004$ in the second. Screenshot 3
Yesterday I ran the app 1.19 (atiOpenCL). My computer is AMD Phenom X4 925. Two ATI HD4850 1GB. Software is Windows 7 Ult x64. Boinc 6.13.12. AMD Driver 11.11 Screenshot
It seemed to me that the GPU load is very low. I looked at the Gigabyte HUD program and saw that the load on the GPU is ~ 40%. And progress of WU increased by 0,002% in the second.
I tried the following method. Priority of the process - real time. Screenshot 1
GPU load was ~ 80%. And progress of WU increased by 0,003%-0.004$ in the second.
Then I returned the priority - below average. GPU load returned to ~ 40%. Progress of WU increased by 0,002% in the second. Screenshot 2
Then I again tried real time priority.The result was repeated. GPU load ~ 80%.
And progress of WU increased by 0,003%-0.004$ in the second. Screenshot 3
Another interesting observation. When boinc only uses 2 cores for the calculations on the CPU, and the app 1.19 (atiOpenCL) process is running with realtime priority we have is Screenshot.
1.19 (atiOpenCL) app process uses 25% of the CPU (what is the total load of one core) and GPU load, thus increasing to a maximum of 95%.
My opinion: Too much demand for the CPU, because of what is too low speed of calculation. For comparison, the NVidia 9500 GT sredenee calculation time of one WU 2,5 hours.
Another interesting observation. When boinc only uses 2 cores for the calculations on the CPU, and the app 1.19 (atiOpenCL) process is running with realtime priority we have is Screenshot.
1.19 (atiOpenCL) app process uses 25% of the CPU (what is the total load of one core) and GPU load, thus increasing to a maximum of 95%.
My opinion: Too much demand for the CPU, because of what is too low speed of calculation. For comparison, the NVidia 9500 GT sredenee calculation time of one WU 2,5 hours.
If you look in the version 1,17 CPU was used as much as "CPU time 1,382.70". Which led to a greater load of the GPU as there Screenshot and much less runtime WU.(Run time 3,649.92)
If you look in the version 1,19 CPU was used only "CPU time 311.66". Resulting in significant lower loading of the GPU as you say(
Quote:
When I ran my 1.19 i saw a very small usage of my GPU for short periods of time.
), and much more runtime WU. (Run time 48,329.31)
If I had to pick between 1.17 and 1.19, I would choose 1.17.
There is great variations in the GPU run time(and the load) between the applications.
Sorry, but that's not caused by the different app versions - I know since this is my machine :-) Have a look again, the latest results of 1.19 are (almost) on par with 1.17 WRT performance. See this one for instance: http://albertathome.org/task/40854
The only difference between those long running task you mentioned is that this time I didn't allow any CPU tasks to run on the same core that the GPU app runs on. This is something we noticed earlier with CUDA apps also. If they don't have enough CPU support, their performance will degrade significantly. Our scheduler specifies that the OpenCL apps need 15-20% of a CPU core and we can just hope that the BOINC client respects that and schedules the host's resources accordingly... We'll keep an eye on this...
I set BOINC to use only 3 of 4 cores. Running no GPU task, total CPU usage was ~80%. Run an albert ATI task and CPU usage jumped to 100%.
Then set BOINC to use only 2 of 4 cores. Running no GPU task, total CPU usage was ~53%. Run an albert ATI task and CPU usage jumped to ~82%.
Summary: This ATI app, on this hardware, needs (or at least uses) ~29% of the total CPU power, or slightly more than a full core.
I am going to let it run a full task with BOINC using only two cores, then using three cores, and will report on run times.
Edit: One more data point. BOINCmanager says this task is using 0.15 CPU + 1.00 GPU. I am not sure where that number comes from, but it clearly way off the mark.
RE: Is there any way that
)
No, not with current Apps. Actually so far we are quite happy we got it working at all.
For now I wouldn't recommend to run Apps on anonymous platform here. We may need to release and deprecate apps at a pace that manual updates won't be able to keep up with.
We may ask you to try some manual tuning later, but only after the development has settled a bit.
BM
BM
Thank you for that
)
Thank you for that answer.
Will you have a thread in the forum about application updates when they arrive and some information of what can be expected from it?
RE: Is there any way that
)
I'll try to answer this question.
Yesterday I ran the app 1.19 (atiOpenCL). My computer is AMD Phenom X4 925. Two ATI HD4850 1GB. Software is Windows 7 Ult x64. Boinc 6.13.12. AMD Driver 11.11
Screenshot
It seemed to me that the GPU load is very low. I looked at the Gigabyte HUD program and saw that the load on the GPU is ~ 40%. And progress of WU increased by 0,002% in the second.
I tried the following method. Priority of the process - real time.
Screenshot 1
GPU load was ~ 80%. And progress of WU increased by 0,003%-0.004$ in the second.
Then I returned the priority - below average. GPU load returned to ~ 40%. Progress of WU increased by 0,002% in the second.
Screenshot 2
Then I again tried real time priority.The result was repeated. GPU load ~ 80%.
And progress of WU increased by 0,003%-0.004$ in the second.
Screenshot 3
The screenshots you have to look at Image
Hi. I see you have a runtime
)
Hi.
I see you have a runtime of aprox. 10hours on your 4850.
I use a 5850 and have a runtime of 10hours as well.
Did the runtime decrease with 40-50% when you ran the task in "real-time mode"?
I tried to look at your computer but it hadn't any completed task so i couldn't look at it. I'll get back to your computor in a day or 2.
http://albertathome.org/host/1393/tasks
Interesting point there with increased priority.
//TQ
Another interesting
)
Another interesting observation. When boinc only uses 2 cores for the calculations on the CPU, and the app 1.19 (atiOpenCL) process is running with realtime priority we have is Screenshot.
1.19 (atiOpenCL) app process uses 25% of the CPU (what is the total load of one core) and GPU load, thus increasing to a maximum of 95%.
My opinion: Too much demand for the CPU, because of what is too low speed of calculation. For comparison, the NVidia 9500 GT sredenee calculation time of one WU 2,5 hours.
RE: Another interesting
)
Interesting.
Here is another computor with app. 1.17 http://albertathome.org/task/38871
Her is the same computor with app. 1.19 http://albertathome.org/task/43443
There is great variations in the GPU run time(and the load) between the applications.
And only a smaller variation in CPU run time(and the load)
When I ran my 1.19 i saw a very small usage of my GPU for short periods of time.
As of what I can tell from the people that ran 1.17 they had a higher GPU load then with the new 1.19
I have downgraded to boinc manager 6.12 and I will wait for 6.13 to be the recomended version before running any more tasks.
And I will try the high priority thing you've shoved me when I run the new tasks.
RE: Here is another
)
If you look in the version 1,17 CPU was used as much as "CPU time 1,382.70". Which led to a greater load of the GPU as there Screenshot and much less runtime WU.(Run time 3,649.92)
If you look in the version 1,19 CPU was used only "CPU time 311.66". Resulting in significant lower loading of the GPU as you say(
), and much more runtime WU. (Run time 48,329.31)
If I had to pick between 1.17 and 1.19, I would choose 1.17.
RE: If I had to pick
)
Me too.
RE: Here is another
)
Sorry, but that's not caused by the different app versions - I know since this is my machine :-) Have a look again, the latest results of 1.19 are (almost) on par with 1.17 WRT performance. See this one for instance: http://albertathome.org/task/40854
The only difference between those long running task you mentioned is that this time I didn't allow any CPU tasks to run on the same core that the GPU app runs on. This is something we noticed earlier with CUDA apps also. If they don't have enough CPU support, their performance will degrade significantly. Our scheduler specifies that the OpenCL apps need 15-20% of a CPU core and we can just hope that the BOINC client respects that and schedules the host's resources accordingly... We'll keep an eye on this...
Best,
Oliver
About CPU usage and run
)
About CPU usage and run times.
Q9550 w/ 5870 , 11.11
I set BOINC to use only 3 of 4 cores. Running no GPU task, total CPU usage was ~80%. Run an albert ATI task and CPU usage jumped to 100%.
Then set BOINC to use only 2 of 4 cores. Running no GPU task, total CPU usage was ~53%. Run an albert ATI task and CPU usage jumped to ~82%.
Summary: This ATI app, on this hardware, needs (or at least uses) ~29% of the total CPU power, or slightly more than a full core.
I am going to let it run a full task with BOINC using only two cores, then using three cores, and will report on run times.
Edit: One more data point. BOINCmanager says this task is using 0.15 CPU + 1.00 GPU. I am not sure where that number comes from, but it clearly way off the mark.
Dublin, California
Team: SETI.USA