A little trick that I discovered using Virtualdub

[mrtristan]

This may or may not be common knowledge, but I decided to share it anyway, in case it's not. If any of you have a dual processor system, you can take advantage of it with Virtualdub in certain situations.

First off, Virtualdub doesn't support multiple processors, so it won't really go any faster in a dual machine that has the same single frequency rating as a single chip machine. For example, a dual 1.67 Gigahertz won't be much faster at encoding video than a single 1.67 Gigahertz machine.

There is a trick that will help in certain situations, though. In my case, I was experimenting eith Divx, since it generally encodes faster than mpeg2. With the right settings, they are nearly identical to me in terms of quality versus bitrate.

Anyway, on to the trick. If you plan on running any filters, then this will work for you.

Open Virtualdub twice.

Use one running copy to run the filters and set it to frameserve.

Open the framserved file in the other copy of Virtualdub and use that one to encode to your final format.

This pertains mostly to Divx, since it's probably the only final product codec that you would use with Virtualdub.

I saw an increase from 15-16 fps to 20fps, after trying this out. Without opening two copies of Virtualdub, the combined processor use would hover around 50%. Using two copies, it ran at 85-90%.

I noticed a while ago that it was faster to de-interlace with Virtualdub and frameserve to Tmpgenc, than it was to run the filter natively in Tmpgenc. It didn't dawn on me that I could do the same thing with Virtualdub, if I only opened up the program twice.

I tested a 1:00 capture originally compressed with huffyuv, at 640x480. The final size was also 640x480, encoded with Divx 5.2, at a bitrate of 2400.

[mrtristan]

I was tinkering around some more and came to this conclusion: Virtualdub's strength is as a frameserver - Whether it's serving to a copy of itself, or another program.

I wanted to test a divx encode against an mpeg encode, but I was impatient, so I ran them all at once.

I opened up Virtualdub three times and used two copies to frameserve.

I encoded to divx with the third copy and also ran tmpgenc.

My source material was 720X480 DV. The only filter that I ran was deinterlace. The final format for both was 320X240.

While both were running, each encoded at about 20 frames per second. If I had run them as two functions at different times, my computer would not have achieved 40 fps with either setup.

I also tried encoding with two versions of Virtualdub and running the filters natively, in the past. Oddly enough, when I started encoding with the second copy, each program bogged down.

I have a request, since I can't test this on my own. If anyone with a single cpu machine could try to open up Virtualdub twice and serve to itself, I would appreciate it. I'm just curious as to how much of a benefit two processors are.

[bunyip]

Just did a quick test (source 352x576 uncompressed AVI) using deinterlace and huffman compression.

Single Vdub doing both achieved an average 18fps.

Two VDubs, the first doing de-interlace and frameserving to the second, with the second doing the compression, achieves an average of 13fps.

Certainly an expected result on a single CPU.