Encoding Test with Hyperthreading - here are the results

[ninja.rogue]

File size: PAL DV AVI 30" - children into swimming pool
System settings: Dual Xeon 2.4 with Supermicro P4DCE+, 1024 MB Rambus Samsung PC800
Standard Bios setting (no overclocking or performance enhancing optimization)
W2k pro SP2 and XP pro - no sp
DC precision 10
Standard matrix for Tmpgenc
vbr setting min 2000 avg 6000 max 9000
Program release: Tmpgenc 2.59 Plus and Procoder 1.035
Both encoders have been set to High priority into task manager
Source file is on separate disk than target file

Encoding Times:

Windows XP PRO - ht enabled
Tmpgeng hq 2pass vbr 3' 25"
Procoder hq 2pass vbr 2' 38"
Procoder mq 2pass vbr 13' 35"

Windows 2k PRO - no ht
Tmpgenc hq 2pass vbr 3' 00"
Procoder hq 2pass vbr 1' 54"
Procoder mq 2pass vbr 8' 20"


Windows XP PRO - ht disabled
Tmpgenc hq 2pass vbr 3' 23"
Procoder hq 2pass vbr 2' 12"
Procoder mq 2pass vbr 12' 55"

Average Q value:
Tmpgenc hq 7.7
Procoder hq 4.68
Procoder mq 4.63

I remain waiting for your comments
Cheers
Paolo

[john@sympatico.ca]

THis is confusing, lower values are better right? so Win 2K performs better than either configuration of Win XP?

[ninja.rogue]

This is confusing for me too.
Indeed, xp is NOT overperforming with reference to W2K in general.
But these results are overwhelmingly in favour of a "classic" dual rig with no hyperthreading or XP at all...

[auenf]

HT making little to no difference doing mpeg-2 encode im not suprised about, altho i see you have a Dual XEON system, and i cant think of why having virtual dual processors on a physical dual processors is a good thing ;)

Enf...

[ninja.rogue]

Auenf,
as long as there are no "real" SMP and SMT enabled applications - i.e. software which can work with both dual and hypertreaded processors (like my rig) - I believe that the only positive result of such a configuration would be to allow you a parallel encoding with two Tmpgenc encodes.
In short, you can split your source file in two, encode both files simultaneously and make a comparison of the total time split-encoding-merge with the encoding of the single source file.
That is what I am going to do ;-)
Will let you know what happens.
Cheers
Paolo

[auenf]

thats fine for tmpgenc, but i wonder if CCE will let you run itself twice properly?

Enf...

[delly01]

Afaik, the P4 is not good at floating point arithmetics, and lives merely on the massive parallelization within the thread. Therefore, if you have a lot of flops in order which can be parallelized within one thread onto the cpus multiple fpus, you will already use the processors ressources to their maximum extent.

If you enable hyperthreading, you enable the processor to distribute these fpus between two threads. If they both make extensive use of the fpus in a way that can be parallelized, these threads will work on the same ressources, thereby blocking each other slightly.

Bye,

Detlev