• Welcome to OS2World OLD-STATIC-BACKUP Forum.
 

News:

This is an old OS2World backup forum for reference only. IT IS READ ONLY!!!

If you need help with OS/2 - eComStation visit http://www.os2world.com/forum

Main Menu

SysBench 0.9.5c Has Been Released

Started by Saijin_Naib, 2008.05.22, 05:03:48

Previous topic - Next topic

Saijin_Naib

I benched my desktop, but the author's site is no longer accepting benchmark results, so I guess I'll post them here.

For all you HT non-believers out there, there is a marked improvement in system performance with HT enabled under OS/2, verified by Sysbench. So far, about 13% improvement in Integer tests and 14% improvement in floating point tests. I'd say this is pretty significant.

Paul Smedley

Quote from: Saijin_Naib on 2008.05.22, 05:03:48
I benched my desktop, but the author's site is no longer accepting benchmark results, so I guess I'll post them here.

For all you HT non-believers out there, there is a marked improvement in system performance with HT enabled under OS/2, verified by Sysbench. So far, about 13% improvement in Integer tests and 14% improvement in floating point tests. I'd say this is pretty significant.

My result is at http://smedley.info/result.html

David McKenna

Saijin....

  I have also attached my results (using 1 processor). My experience with the HT turned on is that it seems slightly sluggish (this is with ACPI 3.09 in APIC mode). Running Sysbench with HT on, the CPU results were as you say - about 15% better. On the other hand PM Graphics were about 20% slower and memory was very much slower - 6990 vs. 4565, which would probably account for the 'sluggish' perception. I'm curious what your memory benches with HT on and off?

Dave McKenna

Saijin_Naib

I honestly have not compared memory nor PM/DIVE tests across the two situations as I was really only looking at what effect HT would have on CPU loading. I will have to get back to you in a few days, my computer is all packed away from moving home from college. I do know that the latest panorama benched a little bit better on PM and DIVE tests versus the version that shipped with RC4, so that made me happy.

Shai

#4
would be interesting to know whether the tests use threads because that has a major influence on how the benchmark performs with HT enabled. the system can run certain processes on different processor (cores) but I guess your CPU load isn't that high when your system is idle, so a non-threaded-sysbench (non threaded tests) should mainly run on a single processor with minimal speed gains which in turn means that any additional speed gain you already have has to be a product of thread-utilisation. you should watch your CPU cores' load during the tests. maybe there should be an explicit thread-efficiency test which implements a heavily thread-based algorithm (while all the other CPU tests should avoid using threads) so the influence of multiple cores can be measured clearly and has a larger influence on the overall CPU test result.
btw. why do you think HT/multiple processors should have any significant influence on the DIVE performance? at first I think the graphics chip has to bear the main computational load and wouldn't this also mean the CPU-based DIVE/PM code runs slower with faster threads than without!??

Pete

Hi All

I had a look at the results from Paul http://smedley.info/result.html and a look through the results.txt from Saijin_Naib and compared them to my own results (attached).

As regards threading: I suspect from comparing these results that threading must be in use. This is based simply on the results from Pauls 4 cpu system, my 2 cpu system and Saijin_Naib's single core system. However, I could be wrong...

Regards

Pete