Hi all !
I have no idea how to optimise renders with threads and subdiv caches and would like to understand this better. I have a new MacPro with 12 cores and 64GB of RAM and think I could get a lot out of it...if I knew how renders work !!!
Terragen actually seems to see 24 cores and that is just confusing me even more. Now what would be an optimal thread number and cache setting ???
I would like some help in understanding what all these settings mean and how to try to profit from this monster computer even more. Can someone point me to some documentation on this so that I may understand a bit what I am doing ???
Appreciate any help !
Paul
The 24 cores comes because your CPU has the "Hyper-thread" function switched on. Meaning for each processor core that is physically present, the operating system addresses two virtual or logical cores, and shares the workload between them when possible. So each physical cpu you got actually appear to the OS as two processors, thus the OS can schedule two processes at once. This is a good thing. I have a dual quad set-up and TG sees 16 processing cores! Yippee!
I allocated 100Mb cache for each proc cores. Below that amount TG was complaining :D
And the optimal thread number is: as much u got! :D :D
Use of all 24 threads is fine, but you will start to see losses in efficiency. It may be possible to get similar performance with say 20 threads, and also leave a little resources available for other applications. There is overhead in processing that is required to manage each thread basically, and when you have tons of threads, the overhead is higher relatively speaking, thus it reduces efficiency. It's worth testing 20 vs. 24 threads. But in general there's no need to mess with it.
Regarding caches, do not change it unless Terragen gives a warning and asks you to. 100MB per thread is good, as archonforest mentions, and Terragen *should* automatically adjust this as it detects more cores and allocates more threads. Larger caches do *not* inherently result in better performance. 100MB/thread is fairly optimal in most cases, especially with high thread counts.
- Oshyan