I was running a GT430 for some months and it was of course quite slow... but it was steady like a train, even quite a bit overclocked (just for fun). Tasks never ended up with "validation inconclusive" with that card. At that time it was running BOINC version 7.3.19 (development version). Never any problems with that combination and it run 24/7.

Then a couple of days ago I upgraded to GTX660 card. Maybe two days later I also updated the BOINC version to new development version, 7.4.8.

Now, I see that although that GTX660 has resulted in many succesfully completed tasks, there has been about 8 "validation inconclusive" results already. It's too much in two days. There must be a problem.

I use BOINC "Windows 64-bit" version 7.4.8 & Win 7 64-bit. Nvidia drivers are latest version, 33788.

So, after a couple of days now, some tasks end up still succesfully but some just don't.

Any ideas?

I forgot to mention...

Einstein@Home is the only project my computer is participating in.

All of those tasks which have ended up "validation inconclusive" are this type:
Binary Radio Pulsar Search (Perseus Arm Survey) v1.39 (BRP5-cuda32-nv301)

So, they all are cuda-tasks... (I don't accept any opencl-tasks).

Einstein@Home is the only project my computer is participating in.

Then Einstein@Home is the project you should ask this at, as it is their application and their validator that deems your work as inconclusive, and in the end presumably marked as invalid.

So, they all are cuda-tasks... (I don't accept any opencl-tasks).

Not sure how you determine that you don't accept any OpenCL tasks, as it's not a choice in the Einstein project preferences to do so. But luckily for you, the OpenCL applications for BRP5 at Einstein are for AMD GPUs only. Were you e.g. to run the FGRP (Gamma-ray pulsar search #3) application, it would only be OpenCL.

---

Thanks for directing me into right direction. I will post my problem at the Einstein@Home forum. I'm sorry for messing up here.

Ps. You are perfectly right about the OpenCL-thing. It's not possible to avoid getting OpenCL tasks. They keep loading on the task list... but I reject them manually.

My own reason is that when I started to use BOINC, I was basicly just eager to see if I managed to use my old video card for this... and watch it really complete some tasks. I felt it was very interesting how GPU can be used for those things, these days.

I just don't want to stress the main processor of my computer too much (heat aspect + other work to do at the same time). It's doing some tasks... but I have dedicated only 33% cpu time and only one core for this. CUDA tasks use CPU only very little. It doesn't build up heat.
But, if I let OpenCL task to run... it beats the hell out of the CPU (no matter what the settings are). Of course that could be done without problems, but I would need to arrange my computer cooling in different way. I like it how it is now... and basicly: I'm only interested in "using GPU" for this science work. OpenCL seems to be extremely far away from that.

Ps. You are perfectly right about the OpenCL-thing. It's not possible to avoid getting OpenCL tasks. They keep loading on the task list... but I reject them manually.

You can of cause deselect 'Gamma-ray pulsar search #3' then you GPU won't be sent OpenCL tasks for that type of Wu,
and if you deselect 'Run beta/test application versions?' or 'Gravitational Wave S6 Directed Search (CasA)' you won't be sent OpenCL CasA Wu's.

Claggy

... but I reject them manually.

Maybe that's why they're rejecting you. ;)

Checks temperature every minute (also GPU):
http://efmer.eu/boinc/

---

My reasons to participate in grid computing:
1. I'm using true renewable energy (German accreditation; no certificate purchasing)
2. Production and disposal (will) amount to about 95% of the energy "used" during the lifetime of my PC
3. Helping

Ps. You are perfectly right about the OpenCL-thing. It's not possible to avoid getting OpenCL tasks. They keep loading on the task list... but I reject them manually.

Well, it is possible to avoid getting them, by either setting the project preferences to an application that only has CUDA, such as the BRP5 app (see the Einstein applications page for which apps they have).

Or you can use the <exclude_gpu> option in cc_config.xml's Options section.

and basicly: I'm only interested in "using GPU" for this science work. OpenCL seems to be extremely far away from that.

OpenCL is just the latest in (open source) GPU and CPU computation; where Nvidia first supported it, however, they have stopped doing so in favor of their own proprietary CUDA. So with support for OpenCL dwindled, and it having stopped at OpenCL 1.1 for all Nvidia GPUs out there, any newer updates in code and hardware won't be used by Nvidia GPUs and thus therefore OpenCL seems slow or erratic. Not OpenCL's fault, but purely Nvidia who don't support it anymore, hardware wise.

My host runs only work on the AMD HD 7870 GPU, it switches between Seti and Einstein. It only runs OpenCL applications on the GPU. Nothing on the CPU cores.

And when I game, and that I do a lot lately, then I switch BOINC off. My % of time BOINC client is running is 15.5306%. :-) But that's better than it was last week, when it was only 8%. :P

Great, since I am admin, I can't post unescaped XML anymore.

---

Thank you guys, all of you!

I learned so much from those answers. Well, the account preferences, huh. This is ridiculous and very newbie... but I had pretty much forgot that whole thing. Must be I haven't visited my BOINC web account page after the first days I entered using BOINC. Or at least, I can't remember seeing there anything options related to choosing tasks. I feel ashamed. I just remember like reading some post where a user asked for possibility to choose only CUDA and reject OpenCL. Maybe it was too old stuff or might be I remember totally wrong anyway.

And yeah, I had also thought about that... if it was unpolite to reject those tasks manually every day :D I don't wanna be unpolite. Peace and love for everybody. But finally I came into conclusion: there is no human on the other end, watching if I keep rejecting them. Propably just a machine... which doesn't get angry of my actions.

I would add to my defence also that I haven't stocked a gigantic amount of tasks on my waiting list. So, any rejected task was delayed only for a few hours before getting another chance with another host :)

I made some adjustments through my account page now. Seems to work! No OpenCL landing here anymore. Thanks :)

I will check out that throttling software. It seems interesting, because it hookes directly with BOINC. I'll test it, at least... and maybe I'll raise the workload for my CPU :)

Years ago I was always like AMD + Radeon guy. Then at some panic point I happened to change system and CPU to Intel Q6600 (earlier motherboard broke down and had to get something to replace it, quickly). At the same time I changed the GPU from Radeon to Nvidia. Then I replaced the CPU with Q9550 and now I got the GTX660. Far away from new... but I've always felt comfortable to follow the newest technical development about 5-7 years late :D
I didn't play games for some years, but recently I wanted to start gaming again, a little bit... and at the moment I'm into games that are about that old too :)

I felt already comfortable with Nvidia this year and was not interested in AMD (no CUDA). But the additional background information that you gave about the OpenCL, it might change my view :)

Actually I was reading some reviews of R9 270X/280X yesterday. They seem nicely balanced products (price/power) and some models are quiet too. I'll keep my mind open again :)

Ouch, too much out of context already, but hopefully nobody got stressed.