Just updated nVidia drivers, and Boinc Manager to 6.5.0

Have two 9600gso's running on two separate boxes.

I was "hopeful" with update to BM 6.5.0, it would allow cpu's to Boinc, while 0.10 of a cpu was "set aside" for gpu to crunch for gpugrid.

Both boxes previously cpu Boinc'ed and gpu F@H'ed with no problems.

Maybe just a n00b, and this is normal, but I see that for about every 10 seconds of "real time", the "cpu time" on gpu increases by 1 second.

On Boinc + F@H, I'm used to seeing close to a 1:1 correspondence.

What am I missing here?

It's a different program. All the programs that use the GPU (graphics card) need some CPU time - to manage the reading and writing of data to disk for instance. The proportions of time spent in the processing within the graphics card compared to the CPU will vary depending on the application for any given machine. GPUgrid is generally reckoned to make the most use of the graphics card of the Boinc / Nvidia gpu applications seen to date.

My most recent wu with the GTX260 card spent 0.116 of the elapsed time in the CPU so I'd suggest you are doing OK.

Phoneman1

Actually what you are seeing is the miricle of proper operation ... :)

Because the GPU Grid application needs so little CPU intervention, you can, and should run other tasks on the machine. The GPU Grid application will "steal" that moment of time every so often to "groom" the GPU and you will slowly accumulate time on the CPU side ...

In my case on my "busiest" system I see from 10-30 minutes of time accumulated on tasks that take hours to run. But that is because we really and truly only consume that much CPU time ... yet the GPU is running nearly full time on the problem ...

When you look at the returned result the CPU time will be the time registered, but you can get the actual GPU time in the error out part ...

This is one of the accounting issues that we have raised with the developers to work on in their copious free time ... so that we do have better accounting of the time applied to tasks...

I don't have any experience with FaH, but if they are not actually doing computing on the CPU, then they are showing the compute time of the GPU as CPU time ...

I mean, it is 6 of one half dozen of the other ... in the BOINC world CPU time is CPU time only ...

Without a screen-saver, just looking to Boinc Manager for some sort of visual confirmation that my gpu is crunching.

With gpugrid, especially as a noob, how do i KNOW that the gpu is actually doing something?

With F@H on gpu, there are options to display the % of work done, and eta of when wu is expected to be completed. I KNOW at any given time whether or not my gpu has a wu queued up and running or not, and after what point in time if it is not completed to check for a problem.

gpugrid doesn't appear to have this. if there is a problem, i.e. stuck in a loop, how will i know this? do i just let it run forever? when do i know to suspend and abort?

Good questions all ...

GPU Grid is good about updating the progress meter so that is the indication. The good news is that I have not seen a "hanging" task here (knock wooden head) ... The bad news is that with your cards being on the slow end this will not be something that updates very fast.

Like all projects you have to watch it for awhile to get used to the ebb and flow of the work ...

IN 6.7 and later the tools to monitor the GPU are supposed to increase ... but we do not know what that wil look like yet.

On my 9800GTX+ it's just a couple of seconds until the progress bar is incremented, which tells you if a WU is stuck (very rare case). After running for some time the ETA shown in BOINC manager should be quite accurate, at least it is for my system. You could also check your GPUs temp to see if it's doing anything.

MrS

---

Scanning for our furry friends since Jan 2002