Comments

apit34356 wrote on 4/6/2008, 11:19 AM
Task manager measuring time is somewhat less that 80-100% accurate, In general terms, the system has a list of tasks that is running, where it a number of counters associated with each task. As the system timer goes off (100millisec) it generates an interrupt, that jumps to a system task scheduler, based on many things like propriety status, number of consecutive time slices permitted, etc, the scheduler decides if it returns to the current task or goes to the next task---- usually during this time a few system checks are done also. During the time slice alloted to the task, if an IO call is make or "other system services", the "services" manager will update the task scheduler data that this task is idling and schedule the "services" with pointers into a special services task list to permit execution. This is a very simplify example, but the point is the system only "guesses" at best if system is 100% or 50% with a good margin of error like 3-13%.
rmack350 wrote on 4/6/2008, 12:24 PM
Render time is, of course, the key thing you want to measure. Usually the CPU usage is related but not always and you are right to suspect that if the CPU is running at 30% then you could get more speed out of it, but sometimes you can't.

Imagine in the car analogy that you're trying to measure fuel consumption instead of speed, fuel consumption being a little more like CPU usage. It varies depending on whether you're going uphill or downhill, or if you're in free fall. You don't need more fuel when you're going as fast as you can possibly go.

Rob Mack
apit34356 wrote on 4/6/2008, 1:42 PM
Another "example" of multi-tasking, assume a ham is a task job, now its sliced into four big pieces and a piece is given to four individuals (cpus) at the table, each one eats the ham at different speeds depending on quality of the meat. now, instead of ham, think of the task being a noodle machine- an unique sequence that depends on the previous data frame, now four individuals are waiting noodle dinner to process, but the noodle flows out a in single long string. Only one cpu gets the "noodle" ;-)
clearvu wrote on 4/7/2008, 4:17 AM
I tried the changes you suggested and I only went from 2:08 to 2:07.

Not what I would call an impressive difference.
rmack350 wrote on 4/7/2008, 10:41 AM
Ahh. I like it! Lot's of potential in the food prep analogy.

Rob
jb1203 wrote on 6/24/2008, 9:53 AM
I have a dual dual cor 4 threads AMD... If I set my render thread to 2 instead of 4 will it render faster? I will have to give it try. Let me know if anyone has tried this and what the results were.

Thanks