CPU % + GPU %

[Gandalf]

The BoincTasks Settings > Tasks provides for CPU %, but does not provide for GPU %.  With the advent of the app_config.xml file, more and more attention is being paid to the efficiency of how the GPU is working.  WCG's HCC1 tasking is being completed years ahead of time partially do to the setting of the GPU configuration.  Displaying the GPU % could go a long way to improving the app_config.xml settings.

Sincerely,
Gandalf

[fred]

The BoincTasks Settings > Tasks provides for CPU %, but does not provide for GPU %.  With the advent of the app_config.xml file, more and more attention is being paid to the efficiency of how the GPU is working.  WCG's HCC1 tasking is being completed years ahead of time partially do to the setting of the GPU configuration.  Displaying the GPU % could go a long way to improving the app_config.xml settings.

Sincerely,
Gandalf

I will check, but I don't think it's possible.

[Pepo]

The BoincTasks Settings > Tasks provides for CPU %, but does not provide for GPU %. [...] Displaying the GPU % could go a long way to improving the app_config.xml settings.

I will check, but I don't think it's possible.

SysInternals' ProcessExplorer and wj32's ProcessHacker are capable of displaying utilization of all GPU nodes for Win7+

I do not say it is doable O:-)

[fred]

SysInternals' ProcessExplorer and wj32's ProcessHacker are capable of displaying utilization of all GPU nodes for Win7+

I do not say it is doable O:-)

TThrottle shows the GPU usage.

The problem is Boinc runs several tasks on on GPU.
And there is no way to determine how much one task uses up the GPU.

My favorite in this case is MSI Afterburner.
It shows the utilization, but that isn't saying everything.
Most cards scale down the clock,  if they get above a set temperature.

My GTX 480 cards clock down just over 80C. So I set a fan profile to keep it just below that.

As the clock goes down the utilization may be 90 %, but with a clock at 50% it actually is 45%.

[Gandalf]

The concern for me is not its temperature, but its productivity.

A CPU does many things, but you provide a percentage for its utilization.  Even if the GPU percentage is not absolutely accurate, an approximation would be better than what we have now (which is 0).

If I set the app_config.xml to 0.5C + 0.5ATI GPU, is the GPU functioning within 50% of its capability for that task.  I need to know if I need to increase the GPU percentile or decrease the percentile in order to improve the time based rational.  Example:  If I am seeing Elapsed Time as 00:10:00 (00:01:00), then something is not being very productive, but WHAT?  How do I answer that question?  I have one Project now that reports Elapsed Time as 00:10:00 (00:09:30).  This tells me to keep my hands off its settings and let it do its thing.

I have no idea how you guys do what you do, and if I am insulting anyones intelligence, my apologies.
I am the one here who is mentally challenged.

[fred]

The concern for me is not its temperature, but its productivity.

A CPU does many things, but you provide a percentage for its utilization.  Even if the GPU percentage is not absolutely accurate, an approximation would be better than what we have now (which is 0).

If I set the app_config.xml to 0.5C + 0.5ATI GPU, is the GPU functioning within 50% of its capability for that task.  I need to know if I need to increase the GPU percentile or decrease the percentile in order to improve the time based rational.  Example:  If I am seeing Elapsed Time as 00:10:00 (00:01:00), then something is not being very productive, but WHAT?  How do I answer that question?  I have one Project now that reports Elapsed Time as 00:10:00 (00:09:30).  This tells me to keep my hands off its settings and let it do its thing.

I have no idea how you guys do what you do, and if I am insulting anyones intelligence, my apologies.
I am the one here who is mentally challenged.

Temperature is productivity, the harder it has to work the warmer it gets. Its a good indication of how much the GPU is used.

0.5C + 0.5ATI GPU
Is just a number that is guesswork and is supposed to be an average number, but has nothing to do with your system.
The tasks may use 0.25 or 0.75 of the GPU or whatever.

The elapsed time gives you some info, but all tasks are not equal, so it may be almost impossible to see how productive the card is that way.
00:10:00 (00:01:00) indicates how much the CPU has to work to keep your GPU working. Meaning a 10% CPU usage.
A low CPU usages means the CPU can do more work on other CPU tasks.

If you have a 4 core CPU the most you can get is 0.25C so 0.5 can never be obtained.
100% of one core / 4 = 25% = 0.25C.

I've seen may values but they are all off, way way off.

And for the GPU it depends on the GPU, some GPU are more efficient when they run 3 tasks at the same time. They all run slower than 1 tasks, but together they are still more efficient.

[Gandalf]

I've added a CPU-Z in my Signature so you can best identify the components of my computer.

Question:  An Intel CPU has cores and threads.  Does that not effect what a 0.5C means?
 

[fred]

I've added a CPU-Z in my Signature so you can best identify the components of my computer.

Question:  An Intel CPU has cores and threads.  Does that not effect what a 0.5C means?

It does, it should mean 0.5 Cpu. On an Intel you get 1 tasks / thread. 4 Thread = 0.25 Cpu each, so 0.5C isn't possible.

[Gandalf]

It does, it should mean 0.5 Cpu. On an Intel you get 1 tasks / thread. 4 Thread = 0.25 Cpu each, so 0.5C isn't possible.

If it isn't possible, then please explain this... (sorry to use your own tool against you  )

[fred]

If it isn't possible, then please explain this... (sorry to use your own tool against you  )

The first two run on the same NV card so 0.5 is right.
the first one runs at 91% = is close to 11% of the CPU (8 threads) = 0.11C

On my computer I have task 0.482C + 1NV that runs at 10% so it's actually 0.0125C

But I see you get some serious issues on the first computer.
The CPU tasks should run as close as possible to 100%, typically 95%-100%.

This will cause the GPU to run less that optimal.

Set in Extra->BOINC preferences -> Processor
On multiprocessor systems, use at most: from 100% to 87,5%
75% will probably be even better.

This looks counter productive but it gives the GPU feeder thread more room to keep the GPU busy.

[Gandalf]

I've already set the multiprocessor systems to 87.5.  Since this picture was taken, a member of XtremeSystems gave me that word.  I must admit the more I communicate with you the less I understand.  I'm just a simple wizard.  You've heard of the saying one step forward two steps back?  That's where I feel I am right now.  That is why I try to get you Grand Wizards to just tell me what to do.  Every time I ask why, I start feeling the quicksand pulling me under.  I'm retired.  Retirement isn't suppose to be this hard.   

I've just figured out how to get Milkyway@Home to run with WCG on this computer.  For right now, I'm still running WCG HCC1, but when HCC is completed, I'll be letting M@H use the 2 GPUs while WCG will have to rest of the computer to crunch with.

I think if I keep reading your posts to me I'll eventually understand some of it.  They do say repetition is the best way to learn.

Thanks for the instruction. 
Gandalf

[fred]

I've already set the multiprocessor systems to 87.5.  Since this picture was taken, a member of XtremeSystems gave me that word.  I must admit the more I communicate with you the less I understand.  I'm just a simple wizard.  You've heard of the saying one step forward two steps back?  That's where I feel I am right now.  That is why I try to get you Grand Wizards to just tell me what to do.  Every time I ask why, I start feeling the quicksand pulling me under.  I'm retired.  Retirement isn't suppose to be this hard.   

Why "'ve already set the multiprocessor systems to 87.5"

It's partly because BOINC isn't following it's own rules, it knows it has a GPU task that requires more CPU time.
But it's still running all the 8 CPU tasks. That mean there is simple no CPU that can handle this all.
So Windows is assigning work to the CPU and GPU tasks more or less evenly.
As GPU tasks are way more effective, 87.5 and probably even better 75, stops a CPU task so that the GPU task have a GPU left to feed them.

The GPU needs continues feeding, if the feeding by the CPU stops or doesn't keep up, the GPU isn't busy full time.

Best startup the Window Task manager and check the overall CPU load. It should be < 90% most of the time, so there is some extra for the feeding.

[Gandalf]

Reference: http://support.amd.com/us/kbarticles/Pages/AMDSystemMonitor.aspx

Take a look a the AMD System Monitor. Does this not represent what I was talking about for your GPU%?

[fred]

Reference: http://support.amd.com/us/kbarticles/Pages/AMDSystemMonitor.aspx

Take a look a the AMD System Monitor. Does this not represent what I was talking about for your GPU%?

I was still thinking about it. 

Getting the GPU usage is no problem, TTrottle already shows it.
But there may be 3 different BONC tasks running on the GPU.
And usage will vary from project to project.

So knowing the GPU usage is meaningless to a specific task.
As the GPU usage may be:

Project 1: 25%
Project 2: 50%
Project 3: 15%

Using the CPU for 90%.

[Gandalf]

I never said it would be easy.  Determining what Project is using the GPU should go into the equation as to how much of the GPU should be being used.  My problem is, while I can put one of these system monitors on each of my computers, I can't monitor them from one location.  I had one GPU stop crunching and it took me half a day to realize it.  It turned out ok, because apparently Propriety CPU WUs will out rank GPU WUs.  I didn't know that before.  I was running around like the proverbial chicken until I understood what was happening.