View Full Version : Segment Memory Limit

11-20-2008, 10:41 AM
I have two Boxx render slaves, each has 8 cores, 4GB of physical memory, and 4GB of virtual memory (page file). I'm using 8 nodes per machine, one per core. I'm using LW95, and using Butterfly Net Render as my GUI for ScreamerNet.

I'm getting Low Virtual Memory errors on my slave machines after a few hours of rendering heavy scenes. By heavy I mean lots of geomery, radiosity, raytracing.

I'd like to try a different Segment Memory Limit in my render globals to see if that resolves the problem. The default is 32. One of the gentleman at Newtek's tech support dept suggested I try changing this value. He described it as how much memory LW allocates to each core at any given point in time.

To anyone who has experience with this, I'm looking for recommendations on what I should set this value at.


11-21-2008, 01:08 AM
8 cores with 4gb between them doesn't seem like much, 500mb ram for each core, at the most.

But as far as segment memory limit, it's easy; set it lower and the line below it will say how many segments the render will get split into, and that depends on the scene. So try smaller numbers, like 30, 26, etc., until it says "2 segments". Try submitting that to your farm to see if you get the same error.

11-21-2008, 02:28 AM
Personally, I'd go higher. I agree that 4gb isn't ideal with 8 cores but I managed ok for a few months until I upgraded to 8gb.

I've got mine set to 2000 currently. I've not had a problem with rendering on my workstation or on the render nodes (quad-core + 4gb) so if you start at half that and go from there... I'd certainly go with at least 512mb, I had it set to that on my old computer with 2gb RAM.

You want to get the render segments down to one if possible, it will speed up the rendering. Ideally you want it at just enough to get a single segment with a little headroom but a larger render will require more memory so go for a figure that covers most of your renders. If you go too high it can start to slow down the renders but usually 1/4 of your total RAM works well.

11-21-2008, 02:31 AM
BTW, I'd reduce the amount of nodes being used to 1 per processor rather than 1 per core. That should help a lot with the memory usage and probably speed things up.

12-02-2008, 03:32 PM
Thanks for writing, guys. I took it down from 16 nodes to 8, set my segment memory limit to 14, and set my multi-threading to 8 threads. Those values seem to have made the entire farm much more stable.