This may be stating the obvious, but I have not seen it discussed here. My GTX460 crashed over the weekend and dumped all 90 pending tasks as 'error while computing'. When I got to work this moring I rebooted the PC and all was OK, but for the first hour SETI refused to give me any GPU tasks, saying none were available. To give the GTX460 something to do I started experimenting with boinc rescheduler to move tasks from CPU to GPU. This allows many of the issues with the current limits to be overcome. When I reached 200 CPU tasks (max for quad core), I moved all except the VLARS and APs to the GPU and SETI promptly gave me 40 more CPU tasks. As far as I can tell, this will enable me to have more than 400 GPU tasks, just keep filling up the CPU then moving them to the GPU, though I have not got there yet as the shorties are processed almost as quickly as I can download and reschedule them.

By the way, what is the current view on running VLARs on a GTX460? I have not tried yet. I have the 768MByte version and have it set to run 2 x MB tasks at a time which typically uses 97% of the GPU but only 75% of its memory.

---

This may be stating the obvious, but I have not seen it discussed here. My GTX460 crashed over the weekend and dumped all 90 pending tasks as 'error while computing'. When I got to work this moring I rebooted the PC and all was OK, but for the first hour SETI refused to give me any GPU tasks, saying none were available. To give the GTX460 something to do I started experimenting with boinc rescheduler to move tasks from CPU to GPU. This allows many of the issues with the current limits to be overcome. When I reached 200 CPU tasks (max for quad core), I moved all except the VLARS and APs to the GPU and SETI promptly gave me 40 more CPU tasks. As far as I can tell, this will enable me to have more than 400 GPU tasks, just keep filling up the CPU then moving them to the GPU, though I have not got there yet as the shorties are processed almost as quickly as I can download and reschedule them.

By the way, what is the current view on running VLARs on a GTX460? I have not tried yet. I have the 768MByte version and have it set to run 2 x MB tasks at a time which typically uses 97% of the GPU but only 75% of its memory.

I have occasionally when almost out of GPU WU's Rescheduled from CPU to GPU.

If you decide to shift VLAR's to GPU they will run (maybe after a fashion) but it is a desperate measure, You will have to reduce processing to a single unit before rescheduling and expext at times some video slowdown dependent on your card. I am running 470's.

If you are that desperate for work it may be easier to step back to Boinc version 6.10.**.

---

Kevin

I`m running a pair of 460`s.
I recently tried a reschedule of VLAR`S to GPU, as I had run out GPU tasks.
Don`t do it!
My machine was just clogged with them running very slowly. RAC went into a terminal dive from around 37000 to 29000 until I hit the abort button on remaining tasks. Won`t try that again.
Regards,

A

---

I prefer SETI credit to other credit, I do ocasionally do other (CPND or Einstein) but I need a good reason to do so.

---

Kevin

I`m running a pair of 460`s.
I recently tried a reschedule of VLAR`S to GPU, as I had run out GPU tasks.
Don`t do it!
My machine was just clogged with them running very slowly. RAC went into a terminal dive from around 37000 to 29000 until I hit the abort button on remaining tasks. Won`t try that again.
Regards,

A

yes the standard CUDA app doesnt play well with VLAR's yet. They are working on one at Lunatics that will run OpenCL for CUDA and be able to run the VLAR WU's

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

I`m running a pair of 460`s.
I recently tried a reschedule of VLAR`S to GPU, as I had run out GPU tasks.
Don`t do it!
My machine was just clogged with them running very slowly. RAC went into a terminal dive from around 37000 to 29000 until I hit the abort button on remaining tasks. Won`t try that again.
Regards,

A

yes the standard CUDA app doesnt play well with VLAR's yet. They are working on one at Lunatics that will run OpenCL for CUDA and be able to run the VLAR WU's

Doesn't need OpenCL to be able to run VLAR on a NVidia card.

My normal mode of operation is to only download task on the CPU and then move them to the GPU. Simply due to all the weird things that occur with the scheduler normally. At the CPU & GPU are not that fast I would have 150-200 CPU task. Then every few days I would dump 50-100 on the GPU. Then the CPU tasks would fill up again.

With the current limits I found it was better for that machine to let the GPU fetch work. At the moment it is sitting on 98 CPU and 340 GPU tasks.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

if you have no GPU work, if there are no
Gpu WUs being given out (think Tuesday), if
you don't want to crunch another project on
your card then yes you can. It won't be fast
and it does make the system lag but it does work.

I`m running a pair of 460`s.
I recently tried a reschedule of VLAR`S to GPU, as I had run out GPU tasks.
Don`t do it!
My machine was just clogged with them running very slowly. RAC went into a terminal dive from around 37000 to 29000 until I hit the abort button on remaining tasks. Won`t try that again.
Regards,

A

yes the standard CUDA app doesnt play well with VLAR's yet. They are working on one at Lunatics that will run OpenCL for CUDA and be able to run the VLAR WU's

Doesn't need OpenCL to be able to run VLAR on a NVidia card.

I am only using Lunatics x39e with Nvidia 280.26 driver on 3 GTX470's.

VLAR's 1 per card runs fairly well, sometimes slight lag when viewing video, approx 60 min per WU.

VLAR's 2 per card will run but you can't do much else on the machine.

VLAR's 3 per card don't go there, causes lock ups errors etc.

The main thing I have to be careful with is to suspend all vlar's (or Reschedule them back to CPU) before restarting 3 WU's per card because all 3 will be grabbed by card 0 and cause problems, I have never had a problem with rescheduling a part run VLAR either way.

Running VLAR's on GPU's is now a lot easier than it used to be on the earlier drivers and software.

---

Kevin

VLAR's 3 per card don't go there, causes lock ups errors etc

Thats what I tried. Disaster. Next time I'll just wait for more GPU units. Nothing like learning the hard way.

Regards,

A

---

Slightly off topic, but decided to start using my HD 5570 with Lunatics OpenCL for Seti a month ago. After about 40 APs noticed they were all either invalid or validation inconclusive but with help from Claggy and Raistmer got it sorted. Wanted to just crunch APs but as we are all aware there just aint enough, so started doing MB as well. Now to the point, Crunching AP has no visble affect on video. Crunching MB about 50-50 Vlar I get real screen lag and usually have to snooze GPU to run any thing else on machine.

---

Slightly off topic, but decided to start using my HD 5570 with Lunatics OpenCL for Seti a month ago. After about 40 APs noticed they were all either invalid or validation inconclusive but with help from Claggy and Raistmer got it sorted. Wanted to just crunch APs but as we are all aware there just aint enough, so started doing MB as well. Now to the point, Crunching AP has no visble affect on video. Crunching MB about 50-50 Vlar I get real screen lag and usually have to snooze GPU to run any thing else on machine.

Hi Jim.

For MB r177 change period iterations Num in your appinfo section to 20.
Should do the trick.

---

With each crime and every kindness we birth our future.

OK Mike will try that

---

actually the idea is to incrementally increase the period iteration until you don't get the lag. That way you get top performance from your GPu while watching videos and also to be crunching seti with the least interuptions

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

actually the idea is to incrementally increase the period iteration until you don't get the lag. That way you get top performance from your GPu while watching videos and also to be crunching seti with the least interuptions

Wouldn't a better idea be to start with something large, safe and lag-free: then gradually reduce period_iteration until the lag starts to get annoying, then back off a step or two?

actually the idea is to incrementally increase the period iteration until you don't get the lag. That way you get top performance from your GPu while watching videos and also to be crunching seti with the least interuptions

Wouldn't a better idea be to start with something large, safe and lag-free: then gradually reduce period_iteration until the lag starts to get annoying, then back off a step or two?

On my experience 20 is a value most cards works fine.

---

With each crime and every kindness we birth our future.

Going with 20 for now, and its looking good. Mind you its a problem that will solve itself later this afternoon if they don't get it fixed. Plenty of cpu but gpu will be having a rest soon lol

---