Posts by mikey

1) Questions and Answers : Getting started : What's the Difference (Message 1552)
Posted 19 Jan 2021 by mikey
Post:
I think the answer is "yes" Mikey.
A 1950X will use all 32 threads to crunch a 32t task.


I figured that but am wondering if I NEED a 32 thread machine to crunch them? Can I for example fudge the numbers and use a 24 thread to make it THINK I have a 32 thread pc or do I even need too.
The short answer is I can't seem to get any of the 32 thread tasks and me not having a 32 thread machine seems to mean I will never run them.
2) Message boards : Number crunching : YAFU running with GPU's (Message 1547)
Posted 23 Dec 2020 by mikey
Post:
I'm running 4 GPU's
3-RX480's (collatz) and
1 motherboard GPU (Einstein) also WUprop and iThena non-intensive cpu
Boinc Mgr states each RX480 is using 0.994 CPU and the motherboard CPU is using 1.0.
YAFU runs 4T and 8T workunits at the same time the GPU's are running. Does this affect the processing in anyway?


Maybe, it depends on your machine and how the Collatz and Einstein apps deal with your 480's, some gpu's require a cpu core free for each gpu so they can run at their max of they are limited while others do not. The best thing to do is run a few cpu units using all the cpu cores and track the time used to crunch the gpu tasks, then suspend all the cpu tasks and track the time to run some gpu tasks again...which way is it faster. Since gpu tasks do over 10 times the work cpu's do it's better to optimize them and limit the cpu tasks. And THAT is why people end up with muplitple pc's, people hate to slow down so you get more pc's spread the gpu's out into separate machines and bang out a bunch of both.
3) Questions and Answers : Getting started : What's the Difference (Message 1546)
Posted 23 Dec 2020 by mikey
Post:
The big ones use every core they can find if not constrained.


So are you saying I need a 16 or 32 core machine to run the 16 and 32 core workunits? I know how to constrain them thru the settings but am asking as I do not own a 32 core machine.
4) Message boards : Number crunching : Credit reduced (Message 1502)
Posted 12 Apr 2020 by mikey
Post:
Here is another example of incorrect CPU time -vs- run time:
http://yafu.myfirewall.org/yafu/result.php?resultid=5592660

The task was running on a Pentium(R) Dual-Core CPU E5300 @ 2.60GHz
Run time 16 hours 55 min 58 sec
CPU time 3 hours 6 min 56 sec

I abandoned task because it quit accruing CPU time despite the Run Time increasing every second..
I am guessing CPU Time should of actually been close to 1/2 of the Run Time or aprox 8.5 hours.

Appears somehow the BOINC wrapper that the actual science tasks runs under lost track of what the science app was doing.


It seems your dual core pc's only have 2GB to 3GB of ram, I wonder if the task is spinning it's wheels due to not enough ram?

Have you tried going into the Boinc Manager to see how much each task is using? Click on a running task and then to the left click on properties and near the bottom it will show something like this:
Virtual memory size 33.12 MB
Working set size 22.60 MB

That's the number for a NON Yafu task as I don't have one running on this pc right now but you get the idea.
5) Message boards : Number crunching : How do I get YAFU to stop sending me tasks that use up 100% of my cores? (Message 1405)
Posted 17 Jul 2019 by mikey
Post:
Hi YoYo, When you define your own class plans (<plan_class>16t</plan_class> & <plan_class>32t</plan_class>) don't you have to pass the client your plan_class_spec.xml file???
https://boinc.berkeley.edu/trac/wiki/AppPlanSpec

This app_config.xml is not working to control number of threads used:
[code] <app>
<name>yafu-16t</name>
</app>
<app_version>
<app_name>yafu-16t</app_name>
<plan_class>16t</plan_class>
<cmdline>--nthreads 16</cmdline>
<nthreads_cmdline>1</nthreads_cmdline>
</app_version>
<app>
<name>yafu-32t</name>
</app>


BTW, can't see the check marks in my Yafu Preferences Run only the selected applications.


This is the app_config.xml file I use at another project to run 1 workunit at a time and use 4 cpu cores per workunit.
It is a 6 core machine so only 1 unit can run at a time.

Obviously you will need to change the "llrsr5" line to the app used here and put it in the project folder for Yafu.

<app_config>
<app>
<name>llrSR5</name>
<fraction_done_exact/>
<max_concurrent>1</max_concurrent>
</app>
<app_version>
<app_name>llrSR5</app_name>
<cmdline>-t 6</cmdline>
<avg_ncpus>4</avg_ncpus>
</app_version>
</app_config>
6) Message boards : News : Badges! We have badges. (Message 1265)
Posted 14 Oct 2018 by mikey
Post:
Thank you very much!!!
7) Message boards : Number crunching : Too many WUs! (Message 1244)
Posted 25 Aug 2018 by mikey
Post:
That cache setting relies on BOINC calculating an accurate completion time. Yafu jobs run much longer than BOINC calculates. The only way would be to limit actual number of jobs retrieved and I don't see a setting for that in BOINC or in cc_config.xml options. <fetch_minimal_work> comes close, but it only restricts the number of jobs to the number of CPUs.


If you use a zero resource share it will only get one workunit per cpu that you allowed to get work in Boinc. For instance I have a 6 core cpu but am only allowing to use 99% of the cpu's so it gets 5 workunits. I reserve one cpu core to keep the gpu fed for the gpu workunits, from another project, I also do.


Right... but they've got at least 16 cores and don't want more than 2-3 jobs at a time (definitely not 16)


They could use this setting under 'Preferences' "Max # of jobs for this project".
I also use the setting "Max # of CPUs for this project" and set it to 2 so only 2 cpu cores are used per workunit.
8) Message boards : Number crunching : Too many WUs! (Message 1242)
Posted 24 Aug 2018 by mikey
Post:
That cache setting relies on BOINC calculating an accurate completion time. Yafu jobs run much longer than BOINC calculates. The only way would be to limit actual number of jobs retrieved and I don't see a setting for that in BOINC or in cc_config.xml options. <fetch_minimal_work> comes close, but it only restricts the number of jobs to the number of CPUs.


If you use a zero resource share it will only get one workunit per cpu that you allowed to get work in Boinc. For instance I have a 6 core cpu but am only allowing to use 99% of the cpu's so it gets 5 workunits. I reserve one cpu core to keep the gpu fed for the gpu workunits, from another project, I also do.
9) Message boards : Number crunching : Too many WUs! (Message 1239)
Posted 23 Aug 2018 by mikey
Post:
I noticed Yafu recently decided to give me more than 3 jobs. Since I only had 16T selected, Yafu gave me more jobs than I'll be able to complete. Since each job takes 1-2 days, can I just cancel out ones that will expire with no credit loss? I'm worried that canceling jobs will lower my credit for finished jobs somehow.

Side note, I had some jobs give me, "Error While Computing" results. Any tips on how to alleviate this?


Yes you can abort workunits you can't finish in time. To alleviate this try lowering your cache in the Boinc Manager, open the Boinc Manager which is down by the clock, then go to options, computing preferences, then in the bottom section are your cache settings, set them to say 0.5 days for the 'store at least' and then 0.5 for the 'store an additional' days of work, be sure to click save at the bottom of that page so your settings are saved. That will give you roughly one day of work and when you need more Boinc will automatically get more.
10) Message boards : Number crunching : Slower and slower, then sticks at 100% (Message 1068)
Posted 15 Oct 2017 by mikey
Post:
have a similar one ... but while waiting the system is not putting other task ? Have that with 8C tasks under Ubuntu with Ryzen


The admin said they are still running when they are done you will start new tasks, the 'finish time' is just a guess by the Boinc software and only after a few hundred or even a few thousand workunits will it be even somewhat close to reality.







Datenschutz / Privacy Copyright © 2011-2021 Rechenkraft.net e.V. & yoyo