Test resuts: MSPro single vs. Premiere dualie

[Dr Mordrid]

http://www.hyperactivemusic.com/MS%20Pro%206.0.htm

Check out;

Mike L; MSPro 6.5 using a single AMD 1700+

Elie: Premiere 6.01 using dual AMD 1700+

Sheesh....

Dr. Mordrid

[Sciascia]

Wow!! I really like Adobe products, but am I ever glad that I went with ulead in this case.

[Jerry Jones]

I wish all of those Premiere-addicts on the Canopus forum could see the results.

Of course, the forum moderators would probably censor such a post anyway.

But - yeah - [MediaStudio + SSE2 + 3DNow! optimized code] beats [Premiere 6.0 + dualies].

Interesting stuff, but it still won't convert the Premiere diehards.

Jerry Jones

[Patrick]

Waytago Celeron!

Interesting to see that my lowly Celeron 900@1116 is 4.85% faster in rendering a DV file with Ulead MSP6.0 than a Dual Athlon XP 1700+ running Adobe Premiere. Amazing!

[BrianP]

Premiere has never fully supported dual processors, so why was this test configuration even included?

Do the tests processor-to-processor and leave it at that. This apples to oranges stuff reeks of the kind of benchmarking that Apple resorts to with their Pentium comparisons.

[Patrick]

Comparing apples and oranges...

Interesting to see that my lowly Celeron 900@1116 is 32.7% faster in rendering a DV file with Ulead MSP6.0 than a Single Athlon XP 1700+ running Adobe Premiere. Amazing!

[Dr Mordrid]

Welcome to the wonderful world of truth in advertising: SMP gives you nowhere near double the performance of a single CPU.

First the program has to be set up for it across the board. Premiere isn't, regardless of the whinings of its more reactionary proponents.

Secondly no matter how well it's coded you have losses, so 2x performance in dualies is a dream and not reality. Most times 1.5-1.7x is the most you can expect.

OTOH MSPro 6.5 supports SSE/SSE2 and 3DNow!+ throughout the program.

BrianP;

Why?

I think it's VERY IMPORTANT to know that you don't have to invest in an expensive dualie system to get fast rendering times in your editor. Sensibilities to the contrary be damned.

Also, I have it on good authority that those tests were made as generic as possible, even leaning towards Premiere in some respects.

One example is where a basic effect had to be modified in order to match it in both editors the MSPro one was changed and Premiere allowed to use its default setup.

If anything this gives Premiere an advantage that it ends up not being able to capitalize on.

Oh well.....

Dr. Mordrid

[BrianP]

My mistake. By the headline of the topic I thought it was solely a comparison between a dual-processor and single processor, not a wide-scale, varied comparison.

[IM_Riktar]

Secondly no matter how well it's coded you have losses, so 2x performance in dualies is a dream and not reality. Most times 1.5-1.7x is the most you can expect.

Not quite right. TMPGEnc will give you close to double (about 95% from my testing). As for the rest, see my post in this thread:

http://forums.murc.ws/showthread.php?s=&threadid=31308

[Dr Mordrid]

Of course that varies WIDELY among MP supporting programs. Premiere is the perfect example of one on the opposite side of the pendulum swing. As for the rest;

http://forums.murc.ws/showthread.php?s=&postid=237658#post237658

Dr. Mordrid

[IM_Riktar]

"Ditto Doc"

[KuroNeko]

I was planning to get a dual Athlon XP, now that stable boards of different manufacturers are available.

Until I read the test of the Asus and MSI boards at Tom's. The boards were OK, that's not the problem. The problem were the testresults of software like 3D studio and Lightwave.

3D studio didn't make use of the dual power at all (where I know it should - unsure if it's a test issue at Tom's or an incompatibility with dual Athlon recognition), and the most disappointing result of all, Lightwave did use the dual power very efficiently, but a single P4 still beat it easily! Reason: Lightwave is optimised for P4 (SSE2).

So processor optimisations are much more powerful than dual CPUs.

I need a dedicated MPEG2 rendering station, only thing this machine will be doing is render MPEG2's using either Tmpeg or CCE SP. Without dedicated testresults I won't risk the considerable extra sum for a dual, instead I'll stick to a single XP or P4.

On that note, has anyone any information on how much speed improvement you get from the supposed P4 optimisations of Tmpeg 2.51?

Neko

[IM_Riktar]

For what this is worth,,,,,,,

Ok, I decided to try and run my own tests to see what the "Plus" version of TMPGEnc 2.51 would do. Here is what I did.

Made a DV file from the Media Studio Pro Benchmark Page testfile just so everyone could run this on their machines with the same file through TMPGEnc.

I loaded the rendered file (550 megs worth) into TMPGEnc and used the wizard to create a DVD NTSC template. I decided to stick with the wizard defaults for the sake of uniform results. This is what I got in time testing:

1 CPU - 8:48
2 CPU - 5:49

Speed increase figures to about 51% utilizing both CPU's. HUH?!?!?!?

WAIT JUST A DAMN MINUTE HERE! I know I wasn't drinking that many beers when I first made tests on TMPGEnc that gave me results in the 95% range of speed. Let me look for my notes here,,,,, ( cut to wide angle, trips over beer cans strewn about office ) OW!!!! I gotta get this place cleaned up,,, anyway,,, Yah right here. I did see that right, what gives with this? Ok Lets try this again and run task manager to check CPU utilization.

Hmmmm, CPU usage is running between 58 - 79% with only 1 CPU enabled? How is that possible? ( Checks graph settings ) Why yes, I do have both CPU's checked. Somethings not right here. Let's go back to my first tests. I mean after all, it was a older version of TMPGEnc I used right?

So I load up a 93 meg DV clip from a recent project and set up TMPGEnc 2.51 Plus for "my" setting of SVCD. Here are those results:

1 CPU - 2:51
2 CPU - 1:28

Speed increase figures to 94% increase (roughly). And now the task manager is reporting what I would expect to see. 47 - 50% with one CPU and 97 - 100% with both running. WTF is up with this? ( Scratches head carefully to avoid painful lump )

While the testfile is a good benchmark across different Single CPU types, I am left thinking that it isn't the best to judge realworld SMP performance. I do realize of course that never was the intention of the file/test.

To be fair, one would have to compare results using the same file I did in their own machine. And since we are talking about a 93 meg file, I don't think sending it by email would be the best idea. Not all of us have those fancy-schmancy broadband connections.

I am still left with questions after the answers.

Any ideas?

[Dr Mordrid]

Those test files were intentionally made to stress codecs because every single frame is different from its predecessor. Not only does the title marquee but the background color is constantly changing. Then there are the transitions & moving paths.

This is particularly the case with temporally compressed codecs like MPEG. This kind of test gives MPEG a real workout because it can't do what it usually does; duplicate large portions of the background and synthesize the rest when encoding P and B frames.

As such the test files also give the dualies a real workout because the codec stressed the most (MPEG-anything) can't skate by just copying data from one frame to another.

"Normal" video footage where there aren't so many changes may well be easier to encode, but then you don't know how the multi's work when the software is stressed. This is very useful information when you're adding effects, overlays, pips etc.etc.

Now you do....

Dr. Mordrid

[IM_Riktar]

Fair enough Doc. I am still left with the feeling that for actual video footage the duallies I am running are giving me their best effort for conversion with TMPGEnc. Of course I now realize that the titles, graphics I create, and transitions in the timeline won't see the full benefit.

But considering the actual footage I shoot is the majority of the final product, I am comfortable with my test results.

Anyone wanna try a test on their P4 with my video file? I would be willing to email it to someone with a broadband connection. Or maybe write it to a CD and mail it out. I will provide the settings I used in TMPGEnc 2.51 Plus for a equal comparison.