Need advice how to speed up LW 8.3 on 64bit Windows with 4 Cores

[mi5]

I have got 2x Opteron 270 (4 Cores) with WinXP Professional x64.
Should I render in LW with 1 Thread or 8 Threads or should I use ScreamNet?
How many threads should I initiate?
Since my FPrime1.5/2.0 did not run on that system (keyerror), I do not use any plugins.

Does anyone have an advice?

Thanks!

[Aegis]

Multi-threaded rendering performance in LightWave is largely scene dependent - small scenes with little geometry and or FX (reflection/refraction/radiosity/HyperVoxels etc.) may well render faster with a single thread. Try loading one of the benchmark scenes (the "Classic Content" Raytrace.lws is a good one) and trying different numbers of threads - you may find that even though you only (:D) have 4 cores that 8 threads renders fastest - that's certainly the case on my dual (32bit) Xeons with Intel's HyperThreading.

[mi5]

I discovered something funny.
If I use the host-application (Lightwave 8.3) with 8 threads, sometimes the taskmanager shows only 100% usage for only one cpu/core. That means 25% from 4 cores. Unbelievable.
That also happens on screamnet. That means to me, that sometimes there are no further threads, even if my preferences are 8 threads.
But sometimes there is a 100% usage for all 4 cores.
So I made up my mind and decided to use screamnet, because there I can definetly start as many threads, as cpu s/cores are available.
And to balance usage of cpu s/cores, I use 4 nodes each with 2 threads. So allways cpu usage is 100% without making too much overhead.
(Tests were made with 20% reflection, 20% refraction, 20% transparency, 20% specularity, 2 spotlights and 2 volumetric spotlights on text-logos with all options on raytrace, reflection, ..., no antialiasing, but 16 recursions) It took about 1 hour per picture.

I would like FPrime running on that system, but it's not running on 64bit with 4 cores (key error). Only on 32bit Win XP it's running.