View Full Version : Can someone try this?

03-04-2003, 03:37 PM
Dual Processers isn't twice as fast as one single.. Or? Is that because the CPUs are working on different part of the image?

But if you use ScreamerNet and and render 2 frames.. CPU1 render one frame and CPU2 render the other frame..
Now we know that both processor doing the same work.. Maybe it will render 2 frames a bit faster than it will when we render 2 frames in Lightwave?

Can someone try this?

03-04-2003, 04:36 PM
kinda new to LWSN, but have tried this(J. baker/screamernet controller pointed this out)
but in the test i ran here the difference was nominal3-4 sec.perframe,(faster...although my first thought was that maybe its slightly quicker due to less Application "overhead"but i really dont have a clue?
also i have read that there are some math functions which cannot be "split" for dual processing so you have to wonder if that comes into play, is less time wasted swapping data between the processors?)
although with hundreds of frames ...over the long haul i am sure it would make a difference...
"the only way to beat cubic inches, is cubic feet"

03-04-2003, 07:40 PM
LWSN is designed as a single-processor application, and runs most effectively when there are two copies running on a dual-processor machine. There is no significant drawback (in render time) between the two LWSN nodes. They may not have exactly 100% utilization in each processor, but the difference has never been significant in my experience, unless the frames are significantly different.

03-04-2003, 09:03 PM
I'm not sure I understand. Would I be better-off buying a dual processor and setting up screamernet to see each processor as a seperate node, or buying another single processor and rendering across the two computers, or does it make no difference.

Julian Johnson
03-05-2003, 03:06 AM
I haven't had a chance to dig around this one much, but LWSN was turned into a multithread renderer a while back so that it behaves in the same way as Layout when it comes to threads i.e. it uses the same number of threads as Layout and it will utilise the dual processors in the same way (responding to the thread value setting in the Prefs file). An easy way to check is to set up a Mode 3 render and save your Layout prefs with a single thread set and then compare the CPU Monitor results by doing the same process with 8 threads - with a single thread processor usage is limited to half of the available CPU headroom whilst with 8 threads both are used to their max.

mlinde and Paintboy have it totally right AFAIK that using both CPUs as nodes is better but I'd be interested to see what kinds of differences there are if the thread value in the Layout 3 Prefs file is set at 1, 2, 4 or 8. I imagine if you have a value of 8 *and* you're using both processors as nodes (so that you effectively have 16 threads going on) performance might suffer...