The Scottish Boinc Team

So the question is; is it better to run 4 threads or 4 instances of 1 thread? Even running 1 thread I have a lot of WU that only last 2 minutes so that would be a lot of loading/unloading and doing prep work if you cut that time by 4.

I'm stuck on Asteroids for the next 3 days due to deadline issues. When I finish there I can play around.

I ran test with 2 threads per WU and 10 thread per WU. Credit per CPU sec was 80% higher running 2 thread per WU. Then I ran another test with 1 thread per WU and the credit per CPU sec was 150% higher than running 10 thread per WU. Now I've got to run multiple instances to see if I keep all threads busy.

EDIT: server will only give you 25 WUs per host so I'm running less than 20 WUs per client in order to keep a few ready to start and not wait for downloads.

scole250 wrote: ↑Fri Aug 11, 2017 5:45 pm Now I've got to run multiple instances to see if I keep all threads busy.

That's great Steve ... thanks! That is the testing I had planned on doing when I finish my load of Asteroids WU. I need another couple of days to finish those off and then I'll see whether it makes sense to put CPUs on it or just stick with the GPU.

I think I have all my rigs dialed in. I'm even going to run the GPUs here at home for a bit to see what kind of numbers I can pull down. Watch out Pete!

Hey Pete, what's a GPU?

Next week I may have to see about a remedy to my lack of GPUs. I've really been fighting the impulse but it is becoming too much to handle!

Pete it's a little more complicated than just adding an app_config (which is also needed). The problem is that the project will only give you 25 WU/machine. So if you setup your 56 thread machine to run single threaded, by using an app_config, then you will have 31 threads sitting idle. If you set it up to use 2 threads you'll still have 6 threads doing nothing. You could run 3 threads/WU and then all threads would be crunching but not as efficiently as only doing 1 thread.

What we are talking about is running multiple instances of BOINC on the machine. To "most" projects each instance will appear as a different machine even though it will show the same name. So to get all 56 of your threads running as a minimum you would have 3 instances of BOINC running with the ncpus set to 18, 18, 20. Due to another issue called NUMA, it would be best to run 4 instances of BOINC with the ncpus on each set to 14.

I think Scole put the directions for how to do multiple clients on TSBT somewhere. If not, check HERE on SUSA. The 1st post and also #7 give you everything you need to know. You monitor/control the extra instances with BoincTasks. It really isn't difficult! Once you've gone through it one time you'll be an expert from then on.

This is the app_config I'm using. Unless you install more instances you'll need to set the cmdline to nthreads 4 and avg_ncpus to 4

<app_config>
<app_version>
<app_name>xansons_cpu</app_name>
<plan_class>mt_windows</plan_class>
<avg_ncpus>4</avg_ncpus>
<cmdline>--nthreads 4</cmdline>
</app_version>
<project_max_concurrent>36</project_max_concurrent>
</app_config>

This app_config will use 4 threads/ WU and all threads will be active. That will generate many more credits than running 56 threads/WU but won't be anywhere near as good as running 1 thread/WU which requires multiple instances of BOINC.

scole250 wrote: ↑Fri Aug 11, 2017 11:34 pm I think I have all my rigs dialed in. I'm even going to run the GPUs here at home for a bit to see what kind of numbers I can pull down. Watch out Pete!

Uh, just an impartial observation but I think Scole may have bitten off more than he can chew

Bryan wrote: ↑Sat Aug 12, 2017 7:37 pm

scole250 wrote: ↑Fri Aug 11, 2017 11:34 pm I think I have all my rigs dialed in. I'm even going to run the GPUs here at home for a bit to see what kind of numbers I can pull down. Watch out Pete!

Uh, just an impartial observation but I think Scole may have bitten off more than he can chew

Don't know about you Bryan, but I think I'll get in some popcorn to watch this one unfold.

Nah, Pete has to much gear. He crunched double what I can do today.

I'm going to back off all but one rig. I'll come back before the project ends to get in a good finishing spot but it still has a couple months left.

As my machines are finishing up with Asteroids I'm moving them over. I'd like to hit 25M before it goes away. I should be able to do that.

There are 9 FB sprints between now and 11/23 so that is going to chew into fun time.

Very impressive numbers Pete and no slouch from Scole either. Dang, I think I need some GPUs.

McShane of TSBT wrote: ↑Mon Aug 14, 2017 2:58 am Bryan, 8 or 10 GTX1080Ti should do it. It will only cost you around 10grand.

I was thinking of adding 2 of em. Of course that means a new PSU also and then the current 1080 has to be moved to another machine and that means another PSU.

Then again maybe I'll get smart and do nothing

I finally got the last machine off of Asteroids so now I've got everything pointed at Xansons. I'm 1/2 way to my goal so it's downhill from here.

The project's active phase is 66% complete. You have been warned!

I'm going as fast as my old legs will carry me With any luck I'll be finished in 2-3 more days.