@William

Nice is not the command you are thinking about. Nice only deals with the short-term scheduler, and it will change something only when there are more processes then a CPU can handle (the process with higher niceness will receive less time than that with lower), I currently have cores that are barely used.

I will research if there is some sort of downclocking for a single process, that is available.

I'm wondering if anybody on the seti@home team is paying any attention to this thread. There is a thread above this one that is talking about cuda and GPU as if they are the only people being affected by this problem...

Anybody? I've shut down my linux box to wait this out.

(edit) My win 8.1 box is not affected.

Boinc Manager (on Linux) has (in Tools-> Computing Preferences) both

On multiprocessor systems use at most ___ % of the processors

and

Use at most ___% cpu time.

It's not clear from the posts which control you have been using.
Set the multiprocessor % to 50 as a start.
Leave the Use-at-most field at 100

That way you will get half the cores (whether you have 2 or 4)
running boinc stuff. I find that the default GPU setup uses
a small percent of an additional core.

Not that this is necessarily going to help...
Oh. On one machine I was having flaky issues (GPU invalids, 30 spikes)
and replacing the power supply, replacing the GPU, and fixing the heat
sink on the Northbridge chip got things working ok again. Dunno whether it
was one or all of those that mattered. Saw some notes on the web
about PSUs that
are ok except have short glitches ...
which got me to replace the PSU.

I've been using 12% of processors at most (1 out of 8 virtual cores) and 40% of cpu time. Still the invalids didn't disappear.

I've temporarily stopped computing on this machine (sorry aliens, you had your chance), but still have two more machines running all the time, but they are at least 10 times slower than this one... however, they produce no invalid results except in rare cases of a crash or something similar.

I've had one instance where using a low value of "Use x% of CPU time" has caused problems. I think it might be that the operating system gets itself in a twist continuously loading and unloading stuff.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

You mention 8 virtual cores so I guess that is 4 cores with
hyperthreading on (set by BIOS) so Linux counts 8.
I never could figure out if
hyperthreading helps with boinc.
Or if it hurts or is irrelevant.
In the end I sort of worried that Linux would not
clearly distinguish whether tasks
were sharing a real core when that was
pointless (if that even makes any sense)
so I turned off Hyperthreading in my BIOS.
Thinking (probably incorrectly) that it could be
better to run on distinct real cores rather than
sharing a real core across 2 tasks.

I agree with Rob that it's
good to set the cpu time % at 100. It means 100% of the
time of the cores-you-allow, but I guess that is obvious.
With those 4 real cores I would think running 100% on one
would be unobtrusive.

I'm not that knowledgeable, really, so I should shut up
and hope you get useful advice from...someone.

Using Ubuntu 13.10 here, on an over clocked PC. Here are some observations.

Make sure that your CPU and internals are being cooled well. SETI will stress your PC/Laptop more than pretty much anything out there. What I did to help this from Day 1 was putting a much better heatsink/fan on my Intel i7 2600K - which is overclocked.

Secondly (and two years later when the second video card went in) I replaced all but one of the case fans with higher air flow fans to ensure that cool air is going in and the hot (and I mean HOT) air is pumped out ASAP.

Now a laptop is going to be different from a heat perspective as you generally must make do with what you are supplied with, which means you need to throttle back how SETI is using your CPU.

Firstly, if its a CPU that uses Hyperthreading, i.e. 2 threads per core, then Ubuntu/Linux will see DOUBLE the cumber of usable cores (as it should be). However when it comes to SETI thats a disadvantage to throttling the CPU back unless you know what you are doing. Why? Because lets say you reduce the number of cores for SETI to use to just 50% - guess what - you are still using ALL YOUR CPU CORES! So you haven't throttled back the CPU at all O.O

So to reduce the cpu load go for 25% CPU usage. Linux actually shifts the usable cores around as well to make use of that cores turbo feature until the core gets hot, then shifts it to a cooler core and starts again. Now keep in mind that one hot core does not make for a hot CPU, and when they are being juggled every sever seconds it averages out the chip heat.

I have never cut CPU usage by using the "Use at most" feature simply due to my (untested) belief that it would be more process intensive leaving less processing for the computer overall and depending on the algorithm used by SETI (or whatever boinc project) may not work as well at reducing temps.

It could be power supply related, but, generally I think bad power supply usually results in hard computer lockups. So unless thats occurring too I doubt its power related.

If all else fails reduce number of CPU's to use to 13%, this will knock the CPU down to 1 core (assuming your CPU has 4 physical cores) and see how that goes. It should also keep CPU temps below 70c.

Hope the above helps a bit.

Cheers

Mark

---

Firstly, if its a CPU that uses Hyperthreading, i.e. 2 threads per core, then Ubuntu/Linux will see DOUBLE the cumber of usable cores (as it should be). However when it comes to SETI thats a disadvantage to throttling the CPU back unless you know what you are doing. Why? Because lets say you reduce the number of cores for SETI to use to just 50% - guess what - you are still using ALL YOUR CPU CORES! So you haven't throttled back the CPU at all O.O

While Boinc may run half the amount of apps, each app uses fftw and you'll find that is run on a different core, meaning there is some load on all the cores
(depending how many app instances you run, and how many real and hyperthreaded cores you have)

Claggy