Has something changed? I know it has on my end. I realized that my wife's computer has an onboard 8200 Geforce that isn't being used. I was pleased to see this as CUDA capable.

So, I enabled it in bios, and the 8200 is working (no monitor plugged into it) along with the main GPU 8600 GT. Not in SLI, which is fine. I understand this, forget about SLI, please :)

BOINC starts and doesn't mention the 8600 at all. It says, "CUDA devices found"
Geforce 8200 coprocessor (2).
Hmm, ok, but I know that the 8600 is working too due to video lag and elevated temperature.

But then I noticed something else; on this triple core processor all 3 cores are almost maxed out. So I looked under, "tasks" and there are 4 separate files that are being worked on, as opposed to only 1 at a time last night. The CPU is under almost a full load. What's different?

BOINC starts and doesn't mention the 8600 at all. It says, "CUDA devices found"
Geforce 8200 coprocessor (2).
Hmm, ok, but I know that the 8600 is working too due to video lag and elevated temperature.

A fix for this problem is coming, eventually.
As per David Anderson:

The latest client handles multiple GPUs, but not quite correctly:
it picks the fastest one, then acts as if all the GPUs are that kind.

The main problem with this is memory: e.g.,
if the faster board has >= 256MB but the slower board has < 256MB,
BOINC will try to run the SETI@home/CUDA app on the slower board.

We'll fix this at some point, but probably not for a couple of months.

-- David

With the latest client, he meant 6.6.4, but do steer clear of that one as it has other problems. Wait until the 6.6 range is recommended before going there.

But then I noticed something else; on this triple core processor all 3 cores are almost maxed out. So I looked under, "tasks" and there are 4 separate files that are being worked on, as opposed to only 1 at a time last night. The CPU is under almost a full load. What's different?

Did you use a cc_config.xml file with the ncpus flag? The only other explanation I can come up with is that it runs 2 tasks per GPU. Seeing how you use about 5 CPU minutes per GPU task, that might be it. (The CPU is only used to move data to and from the video card's memory. It looks like it's swapping a lot here).

---

Seeing how you use about 5 CPU minutes per GPU task, that might be it. (The CPU is only used to move data to and from the video card's memory. It looks like it's swapping a lot here).

Well, I'm certainly no expert, but I know the onboard video is using shared system memory, it doesn't have it's own memory. Perhaps that accounts for the swapping.

Where is that happening? On the X3 or X4? What numbers should I expect in that regard?

I haven't done any kind of file alteration. But it appears like both GPU and the CPU is in use. That machine works on 4 tasks at a time now. 2 tasks specify CUDA, the other two do not.

Well, I'm certainly no expert, but I know the onboard video is using shared system memory, it doesn't have it's own memory. Perhaps that accounts for the swapping.

That could well be it. Many embedded GPUs still use the CPU for some of its functions as well, plus of course the main memory.

I haven't done any kind of file alteration. But it appears like both GPU and the CPU is in use. That machine works on 4 tasks at a time now. 2 tasks specify CUDA, the other two do not.

And the other two show as Seti Enhanced as well? I didn't think it was possible, it's one of the fundamentals in the FAQ that says that you can't do Seti Enhanced on the CPUs and GPUs at the same time. At least, not without the Raistmer optimized application. But you don't show as using that one.

So check again, aren't the other ones Astropulse?

---

I'll check again to be sure here shortly. Wife is on that box playing world of warcraft right now :(

Hey, also I've added a miniitx system to the mix. It's a wimpy little system with a 1.4ghz Geode processor, but it was just sitting around anyway. Once I get stats on it I want your opinion on whether it's worth the juice to run it forever.

Here we go, analyze this for me;

unmodified GPU and CPU working on S@H?

Here we go, analyze this for me;

unmodified GPU and CPU working on S@H?

Most likely you had the 6.03 already in your cache when you first downloaded the CUDA app. It's just finishing up what you had left over of the 6.03. Have you recieved any more 6.03 tasks?

I think once you complete the ones you have you won't get anymore unless you install Raistmer's app ( and those will be marked as 6.08 but will be able to be processed by the cpu as well as the gpu), from what I remember reading in other threads. This kind of behaviour where you can process Enhanced and CUDA MB at the same time, has been seen before when a system is transitioning from 6.03 to 6.08.

What I think is really cool though is you have 2 gpu's going, which has been very hard for some to get to work, but I don't understand the 4 tasks running, without a cc_config.xml file or a Boinc above 6.4.5.

Edit: though after thinking about it again, it mkaes sense. Two cpu tasks and 2 gpu tasks on a 3 core system with two gpu is how it's suppose to work without the cc_config.xml file

---

Thanks for the information there. After those tasks are completed I'll see how it acts afterwords.

The only change I made to that system was enabling the onboard Geforce 8200 graphics for the purpose of more crunching, and those were the results. In any case, I find it's highly efficient to be using that onboard video that was just sitting there disabled, even if the 8200 is relatively slow.

Now having said that, perhaps you can clear something up for me; are some tasks larger/more complex than others? I now have a 1.4Ghz mini itx (no CUDA) running and after 1.5 hours it's at like 30% done on a task. But on my 9800GT CUDA, after an hour I'm only at a few percent. Is this to be expected?

Thanks in advance.

Now having said that, perhaps you can clear something up for me; are some tasks larger/more complex than others? I now have a 1.4Ghz mini itx (no CUDA) running and after 1.5 hours it's at like 30% done on a task. But on my 9800GT CUDA, after an hour I'm only at a few percent. Is this to be expected?

Thanks in advance.

The short answer is yes, some tasks do take longer to process than others on the cpu and gpu apps. The angle range as one example, (among other things) of the task can have a dramatic effect on gpu tasks.

Now the longer version of the answer is probably best left to someone like Jord to answer. He has a much wider range of knowledge than most people on here, and can give you a much more detailed explaination than I could even begin to think of.

---

Ah, ok. Much appreciated.

I think part of the problem was that I was using the 9800GT system, and to top it off the task progress is not refreshing in real time because it's dragging down the graphics card so hard that it's like a slide show. Since I last posted the task is close to 50% done now.

For the time being, I'm using the mini itx to surf/post/etc. It's utilizing 100% cpu but is smoking fast in comparison to the Phenom II right now lol in regard to web browsing.

Ah, ok. Much appreciated.

I think part of the problem was that I was using the 9800GT system, and to top it off the task progress is not refreshing in real time because it's dragging down the graphics card so hard that it's like a slide show. Since I last posted the task is close to 50% done now.

For the time being, I'm using the mini itx to surf/post/etc. It's utilizing 100% cpu but is smoking fast in comparison to the Phenom II right now lol in regard to web browsing.

That 9800GT is a good card, and should get you by with CUDA very nicely. I have one on a gaming rig that I may put on my dual core later for SETI, but for now, I'm content with my little 8500GT on the AMD 64, which the card seems to be faster than the cpu in some tasks even with an opti cpu AK_v8 app.

When I browse on that system, I don't notice any problems or slow downs, though I don't do anything video intensive on it. The CUDA app isn't ready for that yet, ( such as games, video watching etc..) and it's usually suggested you stop the gpu processing when you have something like that going on.

Happy Crunching.

Edit: additions.

---

Ok update;

those tasks completed, and it started new CPU tasks. So I'm still getting 2 CUDA related tasks and 2 CPU tasks crunching simultaneously. Before with one CUDA device, it would only do one task at a time. Now with onboard video enabled, it's doing 2 CPU tasks and 2 CUDA.

I don't understand why this is happening when the consensus has been that it can't. I'm glad it's doing it don't get me wrong, but if we can find out why this is happening perhaps we can exploit it for other systems.

Ok update;

those tasks completed, and it started new CPU tasks. So I'm still getting 2 CUDA related tasks and 2 CPU tasks crunching simultaneously. Before with one CUDA device, it would only do one task at a time. Now with onboard video enabled, it's doing 2 CPU tasks and 2 CUDA.

I don't understand why this is happening when the consensus has been that it can't. I'm glad it's doing it don't get me wrong, but if we can find out why this is happening perhaps we can exploit it for other systems.

Unless you still have a completed task that you haven't reported, according to your task list the only 6.03 task you've completed was task 1150457359. If that's the case, you're still working on one of the ones you were doing before, and have started what should be the last one. Do you have more than two 6.03 tasks in the Boinc Manager task list?

Edit: punctuation

---

I reported one, and the other is finished, and there are 2 more it's working on. So, when I go to report (will hop over and do that now), there will be two reported tasks and two tasks working from CPU to replace them.

edit; ok both cpu tasks completed and reported. It's crunching two more CPU tasks.

I reported one, and the other is finished, and there are 2 more it's working on. So, when I go to report (will hop over and do that now), there will be two reported tasks and two tasks working from CPU to replace them.

Yep two tasks there now for 6.03.

Though I can't explain it, the onboard video using shared system memory seems like a place to start looking, so I'd be curious to what would happen if you added the cc_config.xml file to get 3 cpu's going with the two gpu's.


Edit: but to be honest, it's working and if you're happy with it, don't mess with it. :)

Edit: also I think your statement

no monitor plugged into it

might have something to do with it too. It seems very strange to me too, that the computer summary page shows you having two 8200's but the app being used is clearly using your other card with a monitor pluged into it. As well as the 8200 showing less memory than the CUDA app is supposed to need to run in the first place (I believe it's 254).

---

Well, I got up this morning and now it's running 3 CPU tasks, and 0 CUDA

:(

I notice that the CPU apps are 6.03, and the CUDA were all 6.08. It's not getting new tasks right now, and the ones it is running are all 6.03. Should I start a new thread? I want to get back on using the video cards. I'm told on 4 machines now when I request tasks that none are available.

It appears there was a change to the scheduler which makes it now possible for CPUs and GPUs to work simultaneously on Seti Enhanced work. So the FAQ is now wrong. I'm still waiting for details on what changed.

---

Yes I'm running a 6.03 & 6.08 & Climate Prediction all at the same time, on a dual core with an 8500GT. KOOL!!

---