View Full Version : Processor utilization in Modeler

10-13-2006, 02:48 AM

Recently i upgraded from a AMD Athlon64 single core processor workstation to a system with 2 AMD opteron Dual core processors. I achieved radical improvements in render speed but in modeler nothig much seemed to improve. Its specially when I turning, moving and zoomin hi poly models (50.000 polys an higher) the movement slows down. The reason for this seems to be that th computer uses only 1 thread out of 4 available ( 25% of the resources as i can see in the task manager window). When I'm using LW internal renderer I had a similar problem but solved this by bumping up the Multithreading value in Render Globals. In Fprime it worked instantly.

Are there any way to tweak Modeler (or windows) to speed the dam* thing up ?

My workstation :
Tyan thunder mainbord
2 x amd opteron 270 dual core
4 gig ram
2 x WD raptor 150 gig in raid0 config
1 Tb storage
Nvidia Quadro FX 3450 graphics card

Windows XP 32-bit
Lighwave 9.0


10-13-2006, 03:37 AM
The slow down in modeler is mostly OpenGL related, so you just have to wait until the new version is released. It should have better OpenGL speeds.

10-13-2006, 03:51 AM
Nope, often you will even get slower results with non multi threaded software because most cores on multi core processors run on slower frequencies.

10-13-2006, 03:55 AM
Have the same problem, Modeler only using one core ... but I also dose have that problem in Layout but not when rendering.

10-13-2006, 06:12 AM
Modeler is single threaded, and rely on cpu more than graphics card.

10-14-2006, 10:59 PM
Open two modelers (one on each cpu core) and switch back and forth using the hub to communicate between the two. Do two operations at once, everything will double in speed. SLI modeler, piece of cake.

(just kidding :))

10-15-2006, 03:51 PM
Multithreading is actually available to Modeler and Modeler plugins, but it's very unlikely to be used... As i speak, i'm not aware of any Modelers that have multithreaded tools. Rendering has been multithreaded for many decades...

It's just too hard to make a Parrallel algorithms for most number crunching operations.

Several plugins could be made to be multithreaded i'm thinking Booleans, Poly Reduction etc.... But 80% of Modeler will likely always be single threaded.
So having 50 cores isn't about to make anything faster other than maybe rendering and the odd dynamics engine go faster.

I notice that Dual Processer machines often do cause a hiccup or slight pause sometimes, this can be fixed by changing the processer affinity to one CPU only. Especially in Modelers.

Just because machines are coming out with multiple cores, doesn't mean any software will ever be designed to use them. Parrallelization is completly complicated.

10-16-2006, 05:36 AM
Well to be specific modeler uses cpu to accelerate viewports as far as I understand -- rotate something and you will see processor use go up to 50% with two cores, 100% with one. It actually uses more than one core (at least on amd), but it doesn't use more than fifty percent of each. It also uses the graphic card, but not very efficiently. The next 9 point release will probably include a new modeler opengl implementation that'll rely more on graphic card and be loads faster.

It is actually highly recommended to set affinity, it can be done automatically with a small program called xpcu.

It would be sweet though if Lscripts could be multithreaded automatically ...