Spot the New CPU Application einstein_O1Spot1THi
5 May 2017 20:00:37 UTC
Topic 207346
(moderation:
Looks like we have a new application....(or two)
Continuous Gravitational Wave search Galactic Center Tuning highFreq v1.00- nothing validated yet.
Curiosity is bad for cats....
edit: penguins however are ok ... Linux and Mac versions at the moment. see app page
edit:+: First validated task with @MPW wingpenguin https://einsteinathome.org/workunit/293148843
.... and no mention in the
)
.... and no mention in the technical news or science forums yet.
I'm ready to bring my CPUs back in a few days or a week when it becomes general availability. I've managed to get my standings in all my other projects except Seti out of the nosebleed section anyway. (And since breaking 10k rank there would require ~5M credits that needs an extended GPU outage to be possible; a month or two of CPU time is nowhere near sufficient for that.)
Thanks for the tip. The
)
Thanks for the tip. The "highFreq (AVX)" are running fine on my i7-4770 Ubuntu machine, but no sign of the lowFreq yet.
DanNeely wrote:.... and no
)
.... and right up against a weekend. That's pretty brave :-).
The validator looks to be quite happy with things - 6,000 valid so far, 10,000+ pending and not a single invalid or inconclusive result. There are quite a few failed tasks though.
Cheers,
Gary.
Got some Hi and some Lo
)
Got some Hi and some Lo ones.
The Lo are sent to the lesser CPU's. Memory usage about 270Mb and taking about 6 to 6 and a half hours on my i3-6100T's.
The Hi seem to go to faster machines. I have a batch of 8 running on an i7-6700 that are almost done after 12 and a half hours. Memory usage around 1.4Gb, but probably similar to the last gravity wave run which went up to 1.8Gb depending on work unit.
BOINC blog
There was no announcement for
)
There was no announcement for this testrun as I was busy making sure it gets started on Thursday. Basically it should work the same as the most recent multi-directional search and it is focused on the galactic center this time.
Unfortunately I only found out after starting it that the Windows app is broken and we have to look into fixing it next week. There was also a problem with memory estimations for some of the high frequency tasks. I corrected that and the peak memory requirement is now a bit over 2GB for some of the 1400 Hz and up tasks. We also need to find out what went wrong and why our estimate was off by almost 200MB.
There will be more information once we fixed the two problems we see now and the scientific results we gather now check out with our estimations. Hopefully this will be resolved within next week and the real search can start the week after that.
I also want to add that everything been said in this thread so far, is correct.
Thanks Christian for the
)
Thanks Christian for the update, a couple of strange things.
Some tasks take a lot longer than others.
https://einsteinathome.org/task/641053075
https://einsteinathome.org/task/641169355
running on the same host at the same time but quite different run times. 61Ks and 36Ks resp.
The later deadline ones seem to be faster.
Running tasks have two typical memory sizes 1.44 GB and 0.997 GB.
edit: i have one host running 1418.75Hz tasks, these peak at 1.75GB
Other than that i don't see any errors, and the fat penguin is still happy.
Christian Beer
)
Yikes! Assuming the windows app is as ram hungry as the other two I'm going to need to be ram shopping for most of my systems. I'll probably hold off until you figure out if the higher than expected memory use is a bug or not; since it combined with a future GPU app that doesn't eat a core would mean I'd need to upgrade to 24/32GB on my desktops instead of just to 16 if I don't want to permanently give some of their CPU cores over to other projects.
I was thinking the same thing
)
I was thinking the same thing about ram usage. i upgraded all my i7-6700's for the last gravity wave run so they have 16Gb. The two Ryzen 7's I have being built only have 32Gb. Hopefully they'll get the ram usage down or I will have to use an app_config to limit the number running at the same time.
BOINC blog
MarkJ wrote:I was thinking
)
My buildup's been more ad-hoc. My main personal desktop (i7-4790k) has 32GB in it. It's slightly slower twin (i7-4770k) only has 8. My elderly i7-930 has 12GB of slower ram.
In some ways my personal desktop may end up being the most difficult problem; with science apps wanting 2GB core now (and presumably more in the future) the assumption I made when building it that 32GB would be enough that I'd never need to worry about ram usage may no longer hold true as games and browsers also keep eating more as well. The problem there is that it uses DDR3 and 64GB kits of that are stupidly expensive (too the point that it'd almost make sense to buy a new CPU+Mobo that took DDR4 instead).
Still a better deal than my laptop with 8 non-upgradable GB I guess.
The majority of these tasks
)
The majority of these tasks are running under 1Gb for me, i'd say < 20% over 1.7GB.
I have not seen any reach 2GB but the tasks seem to have sticky locality, seems you get a batch the same frequency and location and these have the same memory profile.
If memory was a problem for me (it's not) i'd just let boinc schedule around it or just crunch FGRPB1 at 0.5GB per task. One fat penguin is running about 30 of these tasks and still shows 27GB free (of 64GB).