Encoding Performance with/without Hyperthreading - negative results

[ninja.rogue]

Hello my friends,
I have installed a dual-boot configuration of my dual Xeon 2.4.
On partition C: I have w2k-pro (which obviously does not have hyperthreading enabled), on partition D: I have xp-pro, which sees 4 processors.
I have tried to test with the same DV AVI file (obtained from Matrox RT2000) of 11' 30" of very complex video. Swimming pool child play.
I have used Procoder 1.035 2-pass vbr 2000-6000-9000 with highest settings, and Tmpgenc 2.59 same settings.
Encoding times are surprising: both Tmpgenc and Procoder are equal or even faster in W2K with 2 processors less...
Procoder is indeed much faster: I have encoding times of 4:1 in W2K and of 5:1 in XP (where it should use 2 processors more...). Using master quality gets 20:1 in W2K and 25:1 in XP.
Tmpgenc has 5:1 speed in W2K and 5:1 speed in XP - here it is quite the same.

Any explanation?
Original file is on a NTFS partition of disk 1 and final file is on a NTFS partition of disk 2.
Cheers people!
Paolo

[destemido]

i think there's a trick in windows XP. HT (hyper-threading supported, but is not activated, so i've heard. Isn't WinXP limited to 2 CPU's??? meaning it can detect 4 virtual processors but it may not use them all. try searching the web for patches. also... and this one is important. not all aplications benefit from hyper-threading. it's the allmighty INTEL that says that. so performance can vary between -25% to +75% (or something like that!!)
windows XP also uses a feature (not sure of the name) but what it does is to automatically manage the system resources, that is why why hardware sites what to test the efficiency of a specific chipset they use win2k to avoid this feature.
i'm sorry for not being a great help... but the information world is a very complicated and complex one.

[destemido]

oh... you could always try win2k server as it supports more CPU's ..
4 i think,

[Pauly]

From what I understand about hyper-threading it does not give you the equivalent of two CPUs in one. What it does is allow you to use two different areas of the CPU at once so that you can execute two dissimilar threads at the same time. Say Web browsing while video encoding. This is a gross simplification but you get the idea.

I currently run XP with two non-Hyper-Threading Xeons but it still displays 4 cpus in device manager. I believe XP always does this regardless of hyper-threading.

There is usually an option in your motherboard bios to enable and disable hyper-threading, it would be worth checking this if you have not already.

I hope to get my own dual 2.4GHz HT Xeons this week so I will let you know if I find anything interesting.

Cheers
Paul

[ninja.rogue]

Hi Pauly,
I have already enabled hyperthreading in my bios.
There is no significant improvement in cpu-intensive activities, like video encoding.
My tests have shown that indeed there is a performance decrease when using hyperthreading.
I will soon post the official results.
Tmpgenc for example is long way faster in non-hyperthreaded environment - simply because it still sees just the two phisical processors. It uses the other ones, may be, but total cpu utilization never gets to 100%. It still remains between 50 and 70%.
Will let you know
Cheers
Paolo

[destemido]

you're confirming what i just said. not all aplications benefit from HT. but in order to benefit from it, they should be coded so that they can take advantage of it. if not, then only the phisical CPU's will be used.

[destemido]

you're confirming what i just said. not all aplications benefit from HT. but in order to benefit from it, they should be coded so that they can take advantage of it. if not, then only the phisical CPU's will be used.

[Pauly]

I eventually got round to installing my 2.4GHz Xeon HTs last night and ran a quick test using CCE 2.66 on XP SP1 with and without Hyper-Threading.

Feeding CCE with avs file though avisynth, DVD2AVI from PAL VOBs

Without HT I get around 2.8 time realtime encoding
With HT I get around 3.2 times realtime encoding

As you can see I get a significant increase in performance with Hyper-Threading.

This was just a quick test, I will try doing more comprehensive tests latter if I can find the time, but on first impressions I see a very worthwhile improvement with HT.

Cheers
Paul

[destemido]

yep to say the least.

[destemido]

by the way where did you get CCE 2.66?? the trial version i have doesn't allow me to save anything.