EPYC server seems to have incorrect data with a new BOINC
log in

Advanced search

Message boards : Number crunching : EPYC server seems to have incorrect data with a new BOINC

Author Message
Technologov
Send message
Joined: 27 Jan 22
Posts: 36
Credit: 301,883,610
RAC: 1,509
Ukraine
Message 3229 - Posted: 29 May 2023, 4:03:44 UTC
Last modified: 29 May 2023, 4:07:52 UTC

Hardware: Dual AMD EPYC 7742; 128-cores; 256-threads; I have committed 160-threads to BOINC.

Software: It is running Debian's version of BOINC 7.20.x + Linux Debian 12 O.S. (from Debian 12 "bookworm" repo)

1.
benchmark
Measured floating point speed 1000 million ops/sec
Measured integer speed 1000 million ops/sec
---
why those numbers? Seems like an over-flow.

1.a. Threadripper 3990X shows a better numbers:
Measured floating point speed 7370.21 million ops/sec
Measured integer speed 140558.86 million ops/sec
(I have not validated if those are correct or not; but a TR-3990X should perform like 3x AMD Ryzen 9 3950X or 4x AMD Ryzen 9 3900X systems)

2. Coprocessors NVIDIA NVIDIA GeForce RTX 2060 (5926MB) driver: 525.99 OpenCL: 3.0

Should write NVIDIA only once. (on older drivers with 430.xx series there was no such problem) -- NVIDIA drivers have changed since the old times, so now it is both a vendor string and part of the GPU model string.

3.IP address = ""
returns me a blank string. Why ? (Computer 86352)

3.a. Threadripper 3990X, which has an older Debian (v11) and older BOINC (7.16.16) works correctly, and returns a correct string. (Computer 83458)

Cross-posted to BOINC development chat on Discord
https://discord.com/channels/605398931479724074/608037526396993554

Profile Keith Myers
Send message
Joined: 26 Jun 20
Posts: 63
Credit: 15,281,375
RAC: 48
United States
Message 3230 - Posted: 29 May 2023, 7:44:40 UTC

The

Measured floating point speed 1000 million ops/sec
Measured integer speed 1000 million ops/sec

is just the placeholder value for new hosts until the host is actually benchmarked using the benchmark menu item in the Manager.

Run it and you will see the value change.

But I agree, it does a poor job on large core count hosts.

The benchmarking code is very old in the client from back in the days of single core cpus.
____________

A proud member of the OFA (Old Farts Association)


Post to thread

Message boards : Number crunching : EPYC server seems to have incorrect data with a new BOINC


Main page · Your account · Message boards


Copyright © 2024 CNR-TN & UniTN