well I seemed to get a bunch of 7-8 second units but somehow they taxed the CPU not the GPU...

But then nothing...

well I seemed to get a bunch of 7-8 second units but somehow they taxed the CPU not the GPU...

But then nothing...

I noticed mine stopped for about 10 minutes half an hour ago. I pressed 'update' and it's flying again...

23/02/2009 22:00:41|Milkyway@home|Sending scheduler request: To fetch work. Requesting 1745280 seconds of work, reporting 3 completed tasks
23/02/2009 22:00:46|Milkyway@home|Scheduler request completed: got 0 new tasks
23/02/2009 22:02:01|Milkyway@home|Sending scheduler request: To fetch work. Requesting 1745280 seconds of work, reporting 0 completed tasks
23/02/2009 22:02:11|Milkyway@home|Scheduler request completed: got 0 new tasks
23/02/2009 22:11:07|Milkyway@home|Sending scheduler request: Requested by user. Requesting 1745280 seconds of work, reporting 0 completed tasks
23/02/2009 22:11:17|Milkyway@home|Scheduler request completed: got 15 new tasks

---

Well I just got some more, but still hitting the CPU not the just the GPU... Though it is hitting the GPU as well. I would just like MW to be GPU only.

Well I just got some more, but still hitting the CPU not the just the GPU... Though it is hitting the GPU as well. I would just like MW to be GPU only.

I must admit I've tried to run it with two other projects, but no luck. It all slows down and then it's off in fury when I suspend the other project. Perhaps someone else is having luck (did someone say they were running ABC with it) ?

I'm not really too bothered, I can crunch other tasks on another PC, but it would be nice to get MW GPU to play with other tasks.

---

workkk I need work...

---

Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.

If I understand the lectures of the other thread that CP gave me, running 2, 4, 6 ,8 tasks is not actually doing more work. It is an illusion. If you have one task showing that it is running you are accomplishing as much work per unit time as if you have 8 "running".

While it shows 8 running at the same time, you really only have one running at a time and it just takes 8 times longer to complete all 8 ... but each takes the same amount of time ...

As best as I can tell this is true. When I was running 8 on the GPU I was running tasks in 8-14 seconds as reported to the site. At the moment, and for the last several days I have been running only one task at a time .. but now, instead of taking a minute to clear the 8 tasks, it takes the 12 seconds to complete each one ...

SO, I am running three RaH, one CPDN and one FreeHAL tasks while burning through MW tasks ...

I also see fewer outages of tasks where I burn through the whole pile and have none on hand (lasts one to 5 minutes) before the next download happens.

That being said, I am also, as near as I can tell, running nearly full time with MW tasks and pulling in 50K per day or so on only the one system (all others I made NNT for this test) ... you can look me up on Willy's site to see my numbers if you wish ...

well right now I am crunching ABC as a main source on all PC's. Then secondary ones are CPDN and SH1-collision. I just added MW to this one computer because it was said that it would use the GPU instead of the CPU.
Right now I can tell when it is running, the GPU fan changes speed, but it still loads up the CPU and kicks off my ABC.
It goes to .50 CPU's used or something like that. So it still dips into my CPU which I do not like. I got it to run on the vid card so I could have two projects running at the same time on different components.

A side note I am trying to add my boincstats link to sig.
ahh there is goes...

---

All GPU applications will use some, to all, of one core of the CPUs installed per GPU core. We had a miserable time on GPU Grid over the holidays because the application was using a really bad idle poll loop and I was losing 22% of one core on my Q9300 and 7% per core on my i7 for each GPU core I was running... bad news ...

So, I have not measured the CPU load running MW, though I know it is likely higher than I would like. It is just the cost of doing business. But, I am running the full number of tasks I would expect on the system, it is just that we lose efficiency on at least one running CPU task (or lose that entire core) ...

THAT said, I will also note that the difference between the loss on a "pure" core and an HT CPU is significant. Look at those numbers 22% or nearly an entire core vs, 7% (of 12%) or a little more than half of virtual CPU (I was and am running 4 GPU cores in 2 GTX 295 cards on that system).

What this tells *ME* is that the best way to run GPU systems is with an HT enabled Intel CPU ... like the one I am running MW on ... :)

While it shows 8 running at the same time, you really only have one running at a time and it just takes 8 times longer to complete all 8 ... but each takes the same amount of time ...

I think I can confirm this as being true. On it's own I can have 4 MW tasks running concurrently. I've found a stopwatch on my mobile phones and found that on average I am completing a WU every 10 to 12 seconds.

Now I have 2 MW tasks running and one CPDN task. MW is still completing a WU in 10 to 12 seconds on average, but now I am now also processing other project's WU.

---

What this tells *ME* is that the best way to run GPU systems is with an HT enabled Intel CPU ... like the one I am running MW on ... :)

That's handy. That's how I'm running my GPU :)

---