Faster Simulation Processing Speeds

#1

Hi CGTalk,

I’m doing some nHair simulations and I’ve noticed that the CPU processing level is only around half on each thread. Is it possible to get better simulation performance without risking system instability?

#2

Many features are either single-threaded (1 core only) or hobbled at the CPU level to only use one thread, even if a core supports two (hyperthreading). Legacy coding. The only way to really speed things up is to get a CPU with a faster single-core, single-thread performance. This bottleneck has been around for quite some time, thus multi-core chips emerged. But the code still hasn’t necessarily changed, and in some cases may simply not work on multiple cores/threads.

You see a great shift towards GPU compute processing in this types of things. Anything that can be parallelized should be.

#3

Thanks for the reply InfernalDarkness, it just seemed strange that rendering is on all 8 threads at 100%, but simulation only gets to about 50% on all threads.

#4

It’s simply because the rendering code has been heavily optimized where possible to take advantage of multiple cores, whereas the dynamics code is very different, and can’t always be optimized in the same way. It’s bananas and pomegranates.

In a perfect world, all dynamics would be offloadable to the GPU.