View Full Version : HT & screamernet

07-08-2004, 04:06 AM
I'm just about to set up screamernet.

With multi processors it needs different nodes for each processor.

Does that mean that a P4 processor with hyperthreading should be treated as a dual processor and have two nodes?

Tom Wood
07-08-2004, 06:18 AM
I have dual Xeons that are hyperthreaded so I end up with four nodes on a single machine. I don't know about P4s, but if it shows two processors in Task Manager, then each should be able to be used as a node.


07-08-2004, 07:20 AM
Screamernet is just a program - so technically you can run as many copies as you want; until of course you run out of RAM and/or everything slows down to such a point it becomes a pointless exercise.

True multi-processor machines should be treated as separate nodes - bear in mind though; they're both using the same RAM resource.

Multi-threaded single processors will obviously be better at dividing the workload more efficiently than their non-multi-threaded counterparts, so if you want to run more than one node on a single MT processor you can - but again RAM becomes the issue.

Whether you have more than 1 node running but with 'multithreading' set to 1 - or whether you have 1 node running with 'multithreading' set to 2,4, or 8 is up to you. Best thing to do is set up a test to see which gives the best performance.

One thing I'm not too sure about is whether the 'multithreading' setting in 'Render Properties' is picked up by the 'LWSN.exe' program and utilised.

I'm assuming it is - anyone?

Tom Wood
07-08-2004, 08:21 AM
One thing I'm not too sure about is whether the 'multithreading' setting in 'Render Properties' is picked up by the 'LWSN.exe' program and utilised.

Hey Matt,

I think it might be because it affects my render time when changed. My scene files are typically only about 200KB and the object files only about 600KB. (I'm animating a talk show using characters like my avatar over there.) I discovered that a single MULTI thread setting in Render Properties is faster than any higher setting. The consensus on the forums was that for simple scenes like mine, LW takes more time than it is worth to get ready to use a MULTI thread.

Note, this has nothing to do with HYPER threading which is set in the BIOS. And, I only run one instance of LWSN because it finds all four nodes on my hyperthreaded dual Xeons by itself. (Well, with the help of your excellent tutorial. :p )

For a machine with four nodes, is the theoretical limit on RAM for each node about one-fourth the total available RAM? And is it the sum total of the scene and object files being rendered that affect that limit?



07-08-2004, 09:55 AM
Thanks guys,

I've got my own little dual xeon arriving tomorrow to help pick up the pace of getting this job through:)

I'll let you know how I go.