Error Report - FYI - Access Denied for all BRP4G-opencl-ati tasks

Also does anyone think a BRP5 task using a Turks Ati GPU should take

111000 seconds per task?? even though it is sucessful??

PB0046_04881_328_1 199950861 27 Sep 2014 6:38:57 UTC 28 Sep 2014 21:12:05 UTC Completed and validated 111,651.12 3,739.92 26.59 3,333.00 Binary Radio Pulsar Search (Perseus Arm Survey) v1.39 (BRP5-opencl-ati)

for RoboRoy

There are two "CPU settings" - the number of CPUs to use and the proportion of time to use them for. The OP states that he is currently using 100/75 and he would probably be better off using 75/100, since there is no indication he is having any sort of a heat issue. In any case, the number of cores available to BOINC or the % of time to use them for has nothing to do with the OP's actual problem.

The other point is that you shouldn't just make the blanket statement that EAH tasks always need a free CPU thread. If you are also interested in the CPU only runs like FGRP4, you may get better overall performance by NOT leaving any free cores - particularly if the GPU is nvidia rather than AMD. What we should emphasise is the need to experiment to see what works best for the individual, their particular requirements, and the hardware mix being used. There are just too many variables to be able to correctly predict outcomes without experimentation.

The OP clearly stated that the CPU was an AMD FX-4300 quad core - ie no HT, so there can't be 8 threads. The best advice, given the low end GPU (HD 6570) would be to set the available cores to 75% and just run 1 task at a time on the GPU since it's quite likely that the GPU crunch time would just double (or worse) if he tried to run 2x. However, he should try the experiment if he wished to test that out.

What evidence do you have for this? Sure, trying to push low end hardware beyond its capabilities can cause tasks to give compute errors, but decent hardware, running at full load and properly cooled, can give error free operation. I have a bunch of GTX650s, some running 2x and some 3x, and with ALL CPU cores loaded with CPU tasks and I'm not seeing compute errors. If I free up a CPU core, all that happens is that I gain a very small reduction in the GPU crunch time which doesn't make up for the loss of the CPU task.

It's a bit different with AMD GPUs. I have a bunch of HD 7850s running BRP5 tasks 4x in quad core hosts. I still have my BOINC settings to use 100% of the CPUs but the default settings within the task (0.5 CPUs) causes BOINC to reserve 2 full CPU cores to support the 4 concurrent GPU tasks. My experiments show that this is pretty much optimal for my situation. If I reserve a third core there is a small reduction in GPU crunch time which doesn't make up for the loss of the CPU task. Some people may deliberately want to run GPU tasks only and that is fine for them. My personal requirements are to maximise my chances of finding either a radio pulsar or a gamma ray pulsar so leaving extra CPU cores free is not an option for me. This is another reason why everybody needs to make their own decision about how they configure things rather than just blindly following some set recipe.

Why? The OP has 8GB RAM and plenty of swap space. He is running FGRP4 tasks on CPU cores. These tasks don't checkpoint all that often and seem to have a variable but usually quite long period at the end before finally finishing. If a task happened to be suspended and NOT kept in memory, you could lose quite a bit of crunch time if it had to be restarted from the last saved checkpoint. Even if you had little physical RAM and had to reload from a disk image, it would be far preferable to do that rather than perhaps losing 10s of minutes of crunch time by reloading from a previous checkpoint.

In my post I clearly started with the words "Just as a test", however that part was not quoted. The intent was to encourage the OP to try different settings and find what works.

It's possible that I could have worded some things in my post differently, such as my example about 8 threads on a cpu when the OP has only 4. It was an example, not a statement about the OPs machine.

I'm not going to address every statement you made regarding my post, some of which may have merit from your own perspective. However, while I am not the most experienced cruncher here, by any means, I don't just make up stuff to respond to someone. I either learned it on my own machines, or learned it right here in these forums.

I could be wrong with some of my statements, I don't think I am, but I have an open mind about it. I also am not going to be taken to task by being partially quoted and accused of telling someone to blindly follow some recipe (your words, not mine).

I would expect better from a forum moderator. If this is how you treat other folks experiences, folks who try to help, then I want no part of it.

I'm outta here.

Phil