Continue to Site

Eng-Tips is the largest engineering community on the Internet

Intelligent Work Forums for Engineering Professionals

  • Congratulations SSS148 on being selected by the Eng-Tips community for having the most helpful posts in the forums last week. Way to Go!

Hyperthreading

Status
Not open for further replies.

rfus

Mechanical
Mar 12, 2004
435
It has always been said that hyperthreading should be disabled. I installed SW07SP4 on a Pentium 4, 3.4 GHz machine and downloaded Anna Wood's Punch Holder to benchmark it. With hyperthreading disabled the feature statistics total was 306 sec. I then enabled hyperthreading and was blown away when the feature statistics came in at 274 secs, and I watched the CPU ride at 100% as it blew through LPattern5.
hpthlt9.jpg

It would appear that this SP4 release (and I want to run the test on 0.0) is more 'multithreaded' than previous versions. Can anyone else confirm on an earlier SP of 07 if enabling hyperthreading improves performance. I suspect this is a result of SW begining to tailor the app to take advantage of multicore processors. It seems that disabling hyperthreading is no longer the rule.

RFUS
 
Are you using single or dual core cpu?

[cheers]
 
Single core;

Hyperthreading disabled 110.24 secs
Hyperthreading enabled 125.8 secs plus a lonnnnnng time for the Ctrl Q.

I'll stick with hyperthreading off for now.

[cheers]
 
rfus,

If rfus had a dual core Pentium 4 you would see four CPU graphs on the screen if hyper-threading was also turned on.

There are portions of SolidWorks that are multi-threaded. Cosmos and PhotoWorks, but also things like drawing view re-generation and I belive by test results I have seen that pattern re-generation in models is also multi-threaded.

As each new version of SolidWorks is released, more portions of it become multi-threaded. So the adage from several years ago to turn off Hyper-Threading is not valid today.

We run all of stuff with hyper threading on for the older systems and with dual core processors and see improvements in performance with the type of modeling we do.

rfus, be sure to send me your results when you get done for the punch holder re-gen so I can add it to my google spreadsheet.... :)

Regards.

Anna Wood
SW 2007 SP4.0, WinXP
Dell Precision 380, Pentium D940, 4 Gigs RAM, FX3450
 
CBL, yes, that was a single core. Were your results with on 07 SP4 or SP0? I do agree that it seems to hang a little harder after the rebuild but the feature stats report shorter with hyperthreading enabled and it is appearent that things like a linear pattern are no longer single threaded.

Anna, I was pretty busy when the benchmarking post was going on here, but took a look at it last night and do want to get in on the mix. I will get you the results on this P4 3.4Ghz, and I wanted to run Scott's benchmark too. I run on the AMD FX-60 too [correction on that AMD vs Intel post, its a A2 - 940 pin socket, not 939] (as does Theo), and saw jeff's results, but thought they should be better, so I want to take a close look at it. Steffan's overclocked E6600 results were impressive and I'm trying to get work to upgrade me to one of those.

RFUS
 
Also try the SSE enabled kernel (I think it is still included in the x86 SW). Last time I tried (SW2006) it was good for about 5% performance improvement.

Stefan Hamminga
EngIT Solutions
CSWP/Mechanical designer
Searching Eng-Tips forums
 
Last results were with SW07-SP4 and with the geometry pattern enabled for LPattern5.

[cheers]
 
The Linear Pattern does seem to be multi-threaded. During the rebuild it was pegged at around 65-70% usage and would spike close to 100% from time to time.

Jason

SolidWorks 2007 SP3.1 on WinXP SP2

 
Here's some more numbers a month too late.

Hyperthreading off: 309.30
Hyperthreading on: 270.16

For some reason my machine seems to dig it. And these stats are for my work machine.

XP Pro SP2, P4 3Ghz, 4GB, Quadro FX 450, SW2007 SP4.0

Tony

HP NW9440
2 GB RAM
Quadro FX 1500M (84.88)
Solidworks 2006 SP5.0
Solidworks 2007 SP4.0
 
I'm with Jason on that--I saw several points where the processors went to 100% during the rebuild--to my pleasant surprise. I see this regularly when updating parts and then returning to the assembly document, or doing a CTRL-Q rebuild in parts--so some items in modeling are definitely taking advantage of more than one thread.

For rendering, it's amazing. For rendered animations, perhaps the best possible investment you can make. Right now it still appears that Intel is riding the crest of the wave with their chips (over AMD).



Jeff Mowry
Reason trumps all. And awe transcends reason.
 
Status
Not open for further replies.

Part and Inventory Search

Sponsor