Measurements of the Intel 3.06GHz Pentium 4HT/TMPGEnc2.59

[King Dick]

Having noticed that a number of forum members are "just waiting to get their hands on the new Pentium 4HT processor", I thought it might be a good time to let those members know just what they can expect when they go out to spend their "Christmas Money".

This new processor has been advertised as a "Virtual Dual Processor" which uses "hyper-threading" technology (not to be confused with muti-threading which is the basis of true dual processor systems).

To test these claims a series of measurements was started on the Pinnacle forum where I did some dual processor comparison measurements. I was later able to test for myself the Pentium 4HT and was surprised to see how good the agreement was with the work of others, especially as my tests used Windows 2000 Pro/SP2 whilst others were using Windows XP Pro.

The measurments were made using a very high grade source material of a disco featuring high speed motion and strobe lights (excellent for testing just how good an encoders motion detection algorithms are). The source is 1231 seconds long captured from a tripod mounted Sony VX2000 camera, using a Pinnacle DV500+ PCI card, in 16:9 true anamorphic DV format (PAL, 720x576).

I have used the conventions of ST for single threading, MT for multi-threading and HT for hyper-threading, TMPGEnc v2.58 and v2.59 are both enabled for all. In my own measurements I used v2.59, and a template was constructed for 2 pass VBR 16:9 PAL mode with 2000min, 6000avg, 9000max, with standard motion search.

ST, HT off, 58.1%RT per pass
ST, HT on, 58.9%RT per pass, speed improvement 1.37%.

MT, HT off, 59.8%RT per pass
MT, HT on, 84.9%RT per pass, speed improvement 42%.

That is the good news, however, there is also bad news:-

MT, HT on, 17% increase in processor power consumption.
DivX, Avisynth, and VirtualDub all showed 10% reductions in speed with HT enabled (obtained from another reliable source).

As a comparison CCE 2.66, where MT is on by default, gave:-

HT off, 74.2%RT per pass
HT on, 80.5%RT per pass, speed improvement 8.49%.

There is also more bad news for those of you who are thinking "we will just turn off HT when we need to use the programmes that run slower":-

The enablement of HT is at this time through a setting in the BIOS, so to disable it you must stop the computer, set the BIOS, and restart!

So one can conclude that programmes written for HT will benefit, whilst those that are not will remain unaffected or show a decrease in speed. It is not to be expected that all programmes will be rewritten in the near future.

One other point that was observed during these measurements, and has been mentioned many times on the forum, whether you are using CCE or TMPGEnc, having the source file on one hard-disk and writing the output to another on a different IDE channel yields a 9% increase in speed.


A Merry Christmas to one and all, Dick.

[Mirror_Image]

And of course you aren't biased against anything that says "Intel" on it. Nice job giving the links for all those "studies". Oh, wait, you didn't...