Setizens, with your highly optimized computers ...
10 May 2020 2:13:05 UTC
Topic 222413
(moderation:
Hi Setizens,
Could We focus our GPUs on the gravitational wave search tasks? (Why?)
I know anyone can and is allowed to compute whatever they feel. (But ?)
The old gamma ray binary search can do well with modern phones and lesser GPUs and they R E A L L Y need those tasks if and when they are available. And Yes, I know, some tasks pay better, but the computing is not for the Top List... It is for fun.
I do not know if it is a problem when some other people come to another project with their computers and begin cherry-picking. I feel it might be.
Happy Mothers Day!
--
Petri33
I'm going to find out how I
)
I'm going to find out how I can set my computer not to accept those easy tasks.
Sorry I did not do that before I posted.
[EDIT] I think it is done now. Only gravitational wave search allowed. [/EDIT]
You have actually been able
)
You have actually been able to get any GW work in the past couple of days? I haven't, only been receiving GR work regularly. The GW tasks ready to send have been only double digits for several days now. I get just the occasional GW task. If I didn't also do GR work, the computers would be only running MW since GPUGrid has been shut down for over a day now.
I assumed it had something to do with the planned maintenance and upgrades, but it was announced those have been completed and yet still very little GW work is available.
I only work here at Einstein
)
I only work here at Einstein as a backup for when GPUGRID is out of work. but it is now, so...
7x RTX 2080's now switched to crunching GW for you :)
the CPU is too weak to effectively feed GW tasks in my 7x 2070 system. (E5-2630Lv2, 60W, 2.6GHz 6c). So it will stay on Gamma Ray.
maybe you can do something about the GW efficiency. only seeing 60-75% GPU utilization on these kinds of tasks. and I can't explain why the 2080s seem to run slower (longer run times) than the other system running 2070s.
edit, I don't seem to have any problem getting GW tasks, all 17 GPUs are crunching them (7 on GR). settings only allow the GW app though, GR turned off for the GW systems, and GW turned off for the GR system.
_________________________________________________________________________
petri33 wrote: Hi
)
. . Hi Petri, nice to see you posting.
. . You make good points and I do wonder if we offend anybody by suddenly swooping on the project in large numbers and gobbling up so many tasks. I do like consistency on my gear and the GRs run in under one hour on all my rigs, GWs take much longer but maybe it would be good etiquette to do some of the heavy lifting.
. . On that subject have you considered attacking a CUDA version of the E@H app? if you were to achieve similar increases in efficiency here as you did at S@H then look out GW tasks, prepare to be gobbled up :)
Stephen
:)
. . Ruh Roh! . . I
)
. . Ruh Roh!
. . I changed one rig over to GW tasks but something got all twisted up and they all bombed out so I aborted the remaining tasks rather than waste time and money then restored the original settings.
. . It seems to be working OK again but I will monitor it to be sure ... :(
Stephen
:(
Don't run GW on cards with
)
Don't run GW on cards with less than 4GB of video memory. probably about half of the GW tasks need more than 3GB video memory. so dont run it on your GTX 950 or 1050 that only have 2GB. and don't run it on GTX 1060 3GB, some will succeed, some will fail.
_________________________________________________________________________
Hi Ian, . . That would
)
Hi Ian,
. . That would explain several things, firstly why I had such erratic results with GW tasks when I first joined E@H, it would have been mostly on the 2GB cards, which was why I dropped the GW project. And also why when I now tried GW on the 1050ti (4GB) it seems to be working AOK. I thought the PC itself might have been an issue so I didn't even try it on the Core2Quad (4GB) with the GTX950 (2GB) but while the Core2Duo only has 6GB ram compared to the i5-6400 with 8GB it is not running any CPU tasks so it seemed fairly equivalent but the advantage is the 1050ti has the extra ram.
. . When the overall RAC plateaus out I will try the 'big boys' [ :) ] with GW too.
Stephen
Keith Myers wrote:You have
)
I only run GW gpu tasks on one box. I have not experienced any "shortages."
https://einsteinathome.org/host/12830231
I read in another thread that when you run both of the active gpu tasks (Pulsar Search #1 and GW) you will have a tendency to get one or the other in blocks. And will not get a smooth combining unless you intervene manually switching from one to the other on blocks of time.
I might be confused about this but I had a similar experience when I tried to run both GPU tasks. I would get a block of one or the other. But very seldom did I get both.
I do have a pending list of several hundred that I am waiting on my wingman to get back to me on the GW tasks.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Ian&Steve C. wrote: Don't
)
You can use the "exclude" gpus in the cc_config.xml file to avoid gpus without enough memory. But I have had trouble trying to mix the smaller Pulsar#1 gpu tasks with the larger GW tasks. I just couldn't get them to "flow" together.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Ian&Steve C. wrote: maybe you
)
While you can push utilization up into the 99+% range by running two tasks per gpu with a cpu thread pushing each task, so far I haven't gotten any actual gains when I do.
You also need to run with "several" cpu threads idle (I have 4 idle) otherwise the gpu utilization continues to tank on my system.
Each task is now taking slightly more than twice as long as it did under single tasking on my Nivida RTX 2060 Supers.
I still have two other ideas I want to try. 2 cpu threads for 1 gpu task. And 1 cpu per 0.33 gpu tasks.
So far, I think the 1 task per gpu is slightly more productive than 2 tasks per gpu. On my Amd 3950x it looks like the 2 tasks per gpu might crunch about 20,000 RAC less than the 1 task per gpu.
If anyone has another idea and/or needs an alpha/beta tester I am available.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!