mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Software

Reply
 
Thread Tools
Old 2016-12-13, 20:52   #1
PerformanceTest
 
Dec 2016

1 Posts
Default mprime benchmark tests backwards?

Hello,
In utilizing the mprime Linux code to benchmark some PC units, I accidentally stumbled upon what appears on the surface to be a simple coding oversight.

Unless I'm confused (entirely possible!), when running the "Options/Benchmark" in mprime (#16 from Main Menu), the order the test sections are run in, and the order they are printed/reported as running in, appears inverted.

Inotherwords, the output states at first "Timing FFTs using 2 threads on 1 physical CPU.", however, if htop is run simultaneously, it is revealed that 4 cores are busy running.

Later on, the output states, "Timing FFTs using 8 threads on 4 physical CPUs.", however htop reveals only 1 core is busy.

It also follows that times (I assume per thread) get better as fewer cores are active and Intel Turbo-Boost can further increase the clock of a single core. (Instrumenting for core clock rates and voltages corroborates these findings.)

Although mprime is not intended as a benchmarking tool, it may be quite suitable for general benchmarking and verification of relative vectorized processor performance, and would be convenient if the printed results matched actual tests run.

I expect this is the result of a simple typo in a for loop, however, I am open to the possibility of user error.

Best regards,
PerformanceTest is offline   Reply With Quote
Old 2016-12-15, 17:21   #2
thyw
 
Feb 2016
! North_America

10011112 Posts
Default

Can confirm, tried out 64 bit mprime/prime95 29.7/9/10, when starting mprime bench at 1 thread the "top" shows 399% cpu, when jumps to 2 thread, it's 300%...
Windows: 100% then 75%, and times improve from "1 thread" to "2 thread".
But when looking into my past logs, the benchmarks show that the timings are improving as they should be. So definitely a mix.


I'm just guessing ahead with bs theories:
Maybe it "fills" the threads with nothing? Or those OSs cannot comprehend the load on 1 thread. Maybe it sees it as full utilization?? Or maybe this was always the case. How nobody noticed it before? Is this normal. Maybe it's intended?

Last fiddled with by thyw on 2016-12-15 at 17:24
thyw is offline   Reply With Quote
Old 2016-12-15, 19:51   #3
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

32·823 Posts
Default

mprime/prime95 does fill the empty cores with "do nothing" tasks. I don't remember why I did that. It's been a while, maybe something to do with turbo/speedstep.

I'll look at this later, I could easily have messed something up.
Prime95 is offline   Reply With Quote
Old 2017-03-01, 02:32   #4
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

32×823 Posts
Default

I've improved setting affinities while benchmarking.

I've also made the do-nothing threads optional with the default being off.
Prime95 is offline   Reply With Quote
Old 2017-03-01, 14:15   #5
VictordeHolland
 
VictordeHolland's Avatar
 
"Victor de Hollander"
Aug 2011
the Netherlands

23×3×72 Posts
Default

Intel CPUs with the Turbo function ON will most likely clock higher if only 1 core is loaded.

For instance on my i5 2500k
Base 3.3GHz

Turbo:
1 core: 3.7GHz
2 core: 3.6GHz
3 core: 3.5GHz
4 core: 3.4GHz

Turbo frequencies per generation/CPU are listed in Intel's specification pdf's (I attached 2nd gen as an example).

If the CPU is overclocked, it's usually best to disable the Turbo function.

Speedstep lowers the frequency and voltage of the CPU to save power (for instance when the CPU is idle). IMHO it's only a useful feature in laptops. I'd happily let the CPU consume a few extra watts for the extra responsiveness on a desktop.
Attached Files
File Type: pdf 2nd-gen-core-desktop-specification-update.pdf (432.6 KB, 378 views)
VictordeHolland is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Looking for AMD Llano to run benchmark tests mrolle Hardware 5 2013-02-16 12:25
Trial Factoring backwards James Heinrich Factoring 16 2011-11-27 11:28
Wrong CPU speed reported in mprime benchmark James Heinrich Software 5 2009-06-13 12:56
Going backwards to go forwards? jasong Hardware 7 2006-03-04 10:46
Benchmark using linux mprime client? nngs Software 2 2005-03-08 19:01

All times are UTC. The time now is 05:57.

Wed Apr 14 05:57:06 UTC 2021 up 6 days, 37 mins, 0 users, load averages: 3.07, 2.84, 2.54

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.