Doom9's Forum - View Single Post

Kedirekin · 5th December 2003, 22:20

I feel compelled to comment, as I've spent a fair amount of time thinking about multiprocessor (MP) systems.

This test isn't just testing CCE encoding speed, it's also testing the speed of AviSynth.

On a single processor system, the script might be using something like 86% of processor cycles, leaving the other 14% for CCE.

On a dual processor system, all of one processor (100%) can be dedicated to the script. All of the left over processor could be dedicated to CCE, but the script is only capable of delivering data so fast, so CCE only ends up using 16%.

The apparent result is a 15% increase in CCE encoding speed, but the test is somewhat biased by a slow script. With a more efficient script, the ratios are completely different, and tend more and more in MP's favor.

Just a point of clarification, CCE is MP enabled. The reason you don't see 100% processor usage is that AviSynth is not MP enabled.

A more fair test of CCE sans frame server might be to use the script to create a Huffy (or raw) avi file, then encode from that. That would really show off the advantages of MP systems. Unfortunately that is misrepresentative of what you're really trying to test (frame server+encoder performance) in the other direction.

BTW: why is that script so slow? I wouldn't think ColorBars or ShowSMPTE would slow the script down so much.

5th December 2003, 22:20	#17 \| Link
Kedirekin Registered User Join Date: Oct 2001 Location: Minnesota Posts: 1,110	I feel compelled to comment, as I've spent a fair amount of time thinking about multiprocessor (MP) systems. This test isn't just testing CCE encoding speed, it's also testing the speed of AviSynth. On a single processor system, the script might be using something like 86% of processor cycles, leaving the other 14% for CCE. On a dual processor system, all of one processor (100%) can be dedicated to the script. All of the left over processor could be dedicated to CCE, but the script is only capable of delivering data so fast, so CCE only ends up using 16%. The apparent result is a 15% increase in CCE encoding speed, but the test is somewhat biased by a slow script. With a more efficient script, the ratios are completely different, and tend more and more in MP's favor. Just a point of clarification, CCE is MP enabled. The reason you don't see 100% processor usage is that AviSynth is not MP enabled. A more fair test of CCE sans frame server might be to use the script to create a Huffy (or raw) avi file, then encode from that. That would really show off the advantages of MP systems. Unfortunately that is misrepresentative of what you're really trying to test (frame server+encoder performance) in the other direction. BTW: why is that script so slow? I wouldn't think ColorBars or ShowSMPTE would slow the script down so much. Last edited by Kedirekin; 5th December 2003 at 22:26.