first impression
log in

Advanced search

Message boards : Cafe : first impression

Author Message
Renato
Send message
Joined: 8 Jul 17
Posts: 5
Credit: 3,834,993
RAC: 0
Message 334 - Posted: 15 Jul 2017, 19:32:49 UTC

Since a few days I calculate XANSONS
I have noticed the following things:
A) the tasks are much too short (6-77 Sec) and generate too much pauses / overhead
B) there are many error messages, especially for fast computers / graphical cards
C) the revenue much the computing time is very low, so no interesting project.
D) The tracking of the results on Boincstats is deficient.

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 322
Credit: 103,382
RAC: 0
Message 335 - Posted: 16 Jul 2017, 11:28:55 UTC - in response to Message 334.
Last modified: 16 Jul 2017, 13:33:46 UTC

Since a few days I calculate XANSONS
I have noticed the following things:
A) the tasks are much too short (6-77 Sec) and generate too much pauses / overhead
B) there are many error messages, especially for fast computers / graphical cards
C) the revenue much the computing time is very low, so no interesting project.
D) The tracking of the results on Boincstats is deficient.

First of all, I apologize for the bad experience you have had participating in this project. This is my first project and I started it with no experience in BOINC at all.

The problem (A) has a workaround. You may place this app_config.xml file to the project directory. It tells the client to run 3 WUs in parallel on a single GPU for the GPU app with the line
<ngpus>0.33</ngpus>
and to use only 2 CPU cores per WU for the CPU app with the lines
<cmdline>--nthreads 2</cmdline> <avg_ncpus>2</avg_ncpus>
On fast CPU and GPU this will allow to use the computational resources more efficiently, because starting and finishing the tasks as well as the data transfer will be overlapped with the computations. This will also allow you to gain more credits. You can change the values of ngpus and nthreads/avg_ncpus as you want (note that the values of nthreads and avg_ncpus should be equal), but I do not recommend running more than 4 tasks in parallel on a single GPU.

Regarding the problem (B), all the errors you got so far are related to file transfer and not related to CPU or GPU performance. I don't know what causes this problem but only 0.2% of your WUs (including all 3 hosts) ended with this error.

In this project, the default BOINC credit system is used. It was not modified in any way to reduce the credit gain. The projects may introduce their own mechanisms for granting credit. However, the BOINC developers state that "projects should use the default credit system unless there is a clear reason not to."

Renato
Send message
Joined: 8 Jul 17
Posts: 5
Credit: 3,834,993
RAC: 0
Message 341 - Posted: 17 Jul 2017, 19:30:47 UTC - in response to Message 335.

Hello Vlad

To a) a "normal" user can not make such xml settings, complete examples would help.

To c) I ran a benchmark today on a Radeon 260x.
Compensated points at Collatz Conjecture 15.13 / sec, for XANSONS for COD 0.28 / sec

That means 53.6 times more points.


Many greetings from Switzerland
Renato

Win10
Send message
Joined: 10 Jul 17
Posts: 4
Credit: 47,870,721
RAC: 0
Message 342 - Posted: 17 Jul 2017, 23:11:28 UTC
Last modified: 17 Jul 2017, 23:21:02 UTC

Maybe your GPU is not optimized for this project. (or the Benchmark gives wrong points)
my gtx 1080 gives me around 5 points / sec for this project

Renato
Send message
Joined: 8 Jul 17
Posts: 5
Credit: 3,834,993
RAC: 0
Message 343 - Posted: 18 Jul 2017, 8:39:23 UTC - in response to Message 342.

Hello Win10

My 1080 GTX not optimized, delivers with Collatz 75-79 points / sec

Profile [AF>Amis des Lapins] Phil1966
Send message
Joined: 12 Nov 16
Posts: 6
Credit: 33,584,218
RAC: 0
Message 419 - Posted: 14 Aug 2017, 18:43:24 UTC

Hello,

I don't understand the credit difference between machines running the same GPU ie GTX 1080 :

1st table is my machine, ID 231

11792616 5804656 14 Aug 2017, 18:24:02 UTC 14 Aug 2017, 18:33:42 UTC Terminé et validé 21.26 5.12 40.59 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)
11792621 5804659 14 Aug 2017, 18:24:02 UTC 14 Aug 2017, 18:33:25 UTC Terminé et validé 24.54 1.00 47.24 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)
11792482 5804600 14 Aug 2017, 18:23:49 UTC 14 Aug 2017, 18:28:36 UTC Terminé et validé 17.22 3.92 73.80 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)
11792518 5804618 14 Aug 2017, 18:23:49 UTC 14 Aug 2017, 18:31:02 UTC Terminé et validé 15.79 4.07 69.54 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)
11792525 5804621 14 Aug 2017, 18:23:49 UTC 14 Aug 2017, 18:30:36 UTC Terminé et validé 12.83 0.72 45.09 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)


other machines running GTX1080

pc n° 325
11783264 5800060 14 Aug 2017, 17:27:41 UTC 14 Aug 2017, 18:29:03 UTC Terminé et validé 57.27 9.95 784.65 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)
11782857 5799862 14 Aug 2017, 17:26:33 UTC 14 Aug 2017, 18:28:06 UTC Terminé et validé 66.17 10.23 1,123.31 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)
11782826 5799846 14 Aug 2017, 17:25:00 UTC 14 Aug 2017, 18:26:59 UTC Terminé et validé 129.21 15.69 1,938.61 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)

et pc n° 1298
11784611 5800728 14 Aug 2017, 17:36:19 UTC 14 Aug 2017, 18:32:13 UTC Terminé et validé 68.91 7.59 914.85 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)
11784102 5800474 14 Aug 2017, 17:33:24 UTC 14 Aug 2017, 18:29:58 UTC Terminé et validé 40.05 7.48 654.15 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)
11784071 5800458 14 Aug 2017, 17:32:27 UTC 14 Aug 2017, 18:29:39 UTC Terminé et validé 105.59 11.22 1,361.50 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)
11783552 5800204 14 Aug 2017, 17:30:07 UTC 14 Aug 2017, 18:26:52 UTC Terminé et validé 38.20 6.50 629.21 XaNSoNS BOINC for GPU v1.05 (cuda65_windows)

???

Thank You.

Phil1966

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 322
Credit: 103,382
RAC: 0
Message 420 - Posted: 14 Aug 2017, 21:16:57 UTC - in response to Message 419.

I don't understand the credit difference between machines running the same GPU ie GTX 1080

Hello,
The default BOINC credit system is very complex. When calculating the credits it implements several averagings and normalizations based on the statistics collected for each host. Also, all Wus are different and it's hard to compare them. But the total number of credits per day obtained by the similar hosts should be more or less the same.

By the way, the credits you got for the last 100 WUs between WU#5806276 and WU#5812034 look normal.

Profile [AF>Amis des Lapins] Phil1966
Send message
Joined: 12 Nov 16
Posts: 6
Credit: 33,584,218
RAC: 0
Message 441 - Posted: 27 Aug 2017, 18:13:02 UTC - in response to Message 420.

I don't understand the credit difference between machines running the same GPU ie GTX 1080

Hello,
The default BOINC credit system is very complex. When calculating the credits it implements several averagings and normalizations based on the statistics collected for each host. Also, all Wus are different and it's hard to compare them. But the total number of credits per day obtained by the similar hosts should be more or less the same.

By the way, the credits you got for the last 100 WUs between WU#5806276 and WU#5812034 look normal.


Hello Vlad,

Thank you for your answer.

NB How comes the daily credit went down so fast since 3 days ?

Re : https://boincstats.com/fr/stats/169/project/detail/lastDays

On a given machine, the distributed credits went from x -> x/3.

See ie machine ID 231

Did the production really went down ? Are the items calculated different ?

Just interested to know why this decrease happens. Credit level itself not so important, as the same apply for all of us.

Thank you for all your har work

Best Regards

mikey
Avatar
Send message
Joined: 10 Aug 17
Posts: 3
Credit: 14,852,526
RAC: 0
Message 442 - Posted: 28 Aug 2017, 1:34:57 UTC - in response to Message 343.

Hello Win10

My 1080 GTX not optimized, delivers with Collatz 75-79 points / sec


Comparing credits from one project to another is not a fair comparison as each has their own agenda in how they give us the credits. Some projects use credits to entice crunchers to come there, other projects give out fewer credits to keep their servers running as a bigger workload can make them crash. In the past some projects sold their results so paid a ton of credits to get people to crunch for them, I don't know of any projects that do that right now though. Seti tried to enforce a 'credit new' process a while back but since each project does it's own thing, only using the Boinc software as the common denominator between them that didn't work and for the most part was dropped.

BTW 1 Million credits and $10 can get you a cup of coffee at Starbucks, but without the $10 you are NOT getting a cup of coffee!!

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 322
Credit: 103,382
RAC: 0
Message 443 - Posted: 28 Aug 2017, 20:58:37 UTC - in response to Message 441.
Last modified: 28 Aug 2017, 20:59:02 UTC

NB How comes the daily credit went down so fast since 3 days ?

Re : https://boincstats.com/fr/stats/169/project/detail/lastDays

Hello,

This seems like the decrease in production in terms of processed WUs (because utilization of the server also decreased) but probably not in terms of real computational performance. Such behavior is possible If some relatively heavy WUs were processed in these 3 days (COD contains CIF files of 10 KB as well as the CIF files of 10 MB and sometimes they are grouped). I noticed that the best days (in terms of granted credit) were the days when small WU were processed. So, it's quite possible that the BOINC Credit Systems grants a little more credit to small WUs than it should. At first glance, this may seem unfair, but it's not, as everyone is in the same boat.

Also, compare the lowest bar charts for this project, Rosetta, LHC and Universe. They all show nonmonotonic behavior.

Message boards : Cafe : first impression


Main page · Your account · Message boards


© 2020 Vladislav Neverov (NRC 'Kurchatov institute'), Nikolay Khrapov (Institute for Information Transmission Problems of RAS)