Message boards : Number crunching : Ryzen 7 3700X numbers.
Message board moderation

To post messages, you must log in.

AuthorMessage
lazlo_vii
Avatar

Send message
Joined: 20 Nov 19
Posts: 21
Credit: 1,074,330
RAC: 0
Message 40757 - Posted: 2 Dec 2019, 19:52:43 UTC
Last modified: 2 Dec 2019, 20:18:31 UTC

I have two nearly identical computers. The have the same models of CPU, RAM, and motherboards with the same BIOS settings (Eco mode is turned on which lowers the TDP of each CPU from 65W to 45W) and the same OS (Ubuntu 18.04). They are both set to run BOINC 24 hours a day, 7 days a week. Where they differ is what software I have installed, how I use them, and how BOINC and LHC are configured.

Host 1 (laz-ubtop) is my destop system. Almost all of the software installed on it comes from the Ubuntu repositories with the following exceptions: Steam and the games I play through Steam, a few games I bought through The Humble Store, and Singularity. I had to update Singularity by compiling the source code because some Atlas tasks were failing because (I think) the disk images were made with a newer version than the one in the Ubuntu repository. I have this system set to accept any and all work task from all LHC projects. I have it set to use 12 "cores" and until a few minutes ago it was set to use up to 8 "cores" per task. I just changed it that 4 "cores" because running tasks kept getting suspended so the Atlas task could start.

Host 2 (rebelhostess-v2) is my home server. With it I run my NextCloud install, my squid proxy server (only used for LCH), NFS shares, hostapd for my wifi, the occasional VM if I want to explore what's happening with other distros or BSD. I also have the this system set to use 12 "cores" but it only does Atlas native for LHC. It also has a LXD container with GPU pass-through running Einstein@Home.

According boincstats.com Host 2 is outperforming Host 1 by about 10K points a day:

https://www.boincstats.com/stats/-1/host/list/0/0/bb6dd2820ecefbe946a788a41fe15d11

Are the Atlas task really worth so much more scientifically? If so should I go pure Atlas on my desktop as well? I don't really care about the points. I just want to help with the science. If the point values are not a reflection of the science value that's OK, too. I am just ignorant about it all works and I know that questions can lead to answers and answers are the only real cure for ignorance.

EDIT: OK, so I just realized where my error was. Boincstats isn't breaking down the numbers by project and is including my Einstein@Home points with my LHC numbers.
ID: 40757 · Report as offensive     Reply Quote
computezrmle
Volunteer moderator
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 15 Jun 08
Posts: 1428
Credit: 73,045,305
RAC: 106,240
Message 40758 - Posted: 2 Dec 2019, 20:46:40 UTC - in response to Message 40757.  

Identical hosts?
See your computer details pages:

https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10623665
Measured floating point speed 6.05 billion ops/sec
NVIDIA GeForce RTX 2070
driver: 440.26

https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10624568
Measured floating point speed 6.24 billion ops/sec
NVIDIA GeForce GTX 970
driver: 435.21


Although the floating point speed is not directly used for credit calculation it is used as a starting point.
Now, don't wast your time trying to sync the benchmark results.

The best you can do is to setup a reliable and healthy system.
90% reliable throughput would be better than 98% interrupted by crashes or heatproblems...
ID: 40758 · Report as offensive     Reply Quote

Message boards : Number crunching : Ryzen 7 3700X numbers.


©2020 CERN