HC Speed with/without SMP

[jdobbs]

I was just wondering if anyone else has done any speed testing with and without the SMP parameter set for HC.

Based on some reports here on DOOM9 I set the default for HC to SMP disabled when multiple instances are running. But, in my own testing on AMD processors (dual and quad) I find that the encoding runs about 13% faster if I leave it alone and let SMP stay enabled.

I'm considering reversing the default, but I'd like to hear others experience first. It could be just an AMD thing. If I find that to be the case, I may do some testing to determine the processor type and change the default based on that.

In order to enable SMP you would add this to the INI file's "[Options]" area:

HC_NOSMP=0

This disables SMP (also what you would get if nothing was in the INI):

HC_NOSMP=1

[blutach]

I have HC_NOSMP=0 and find it is faster (also using AMD X2). This is especially so on the last segment, where if that segment is longish, you can really pick up speed. On short last segments, I find no meaningful difference.

Regards

[shon3i]

@blutach, did you mean leave only SMP an set one (single core) insance in DVD RB?

[blutach]

Sorry for the misunderstanding. I run with multiple encoder processes on. I did try with MEP off and No SMP, but also found it wasn't as fast as my current configuration.

Regards

[rendez2k]

Has anyone tested this with an Intel quad core? My new machine (with a Q9450 processor) should be arriving today so I'm interested to know the best setting.

[blutach]

Post your results! That's what jdobbs is seeking.

Regards

[rendez2k]

Post your results! That's what jdobbs is seeking.

Regards

Will do! Just wanted to check before I started testing if anyone else was or had. Its going to take me a day or two to get Windows back on anyway!

[jdobbs]

I'm sitting here looking at my own post asking myself "Why did you name the parameter HC_NOSMP?" -- Why didn't I just call it "HC_SMP"... that way the "1" and "0" choice would be a lot easier to follow.

Double negatives... what was I thinking?

[blutach]

@jdobbs - if you recall, in the beta, it was a postive and changed to a double negative. :)

Regards

[rendez2k]

Gave it a try with HC_NOSMP=0 and the processing time went from 46 mins to 45 - so not a big difference. Does it depend on the DVD though?

[blutach]

It can. If the last segment is longish, then I have found giving it the whole processor will have quite an effect. If it's just a couple of minutes, then there will be little gain.

Regards

[zacoz]

Ran 3 tests over last couple of days

SMP Disabled (No entry in the ini) + MEP (2) Enabled = 100 minutes (3 / 84 / 13)
SMP Enabled (HC_NOSMP=0 in the ini) + MEP (2) Enabled = 99 minutes (2 / 83 / 14)
SMP Enabled (HC_NOSMP=0 in the ini) + MEP Disabled = 101 minutes (2 / 87 / 12)

Test Disc was Stargate: The Ark of Truth R4 with some preprocessing (remove crap and some audio & subtitle streams). Steal space from extras set at 33%, HC set to best, and just realised that I still had a filter active for all three tests - RemoveGrain(mode=2). Ooops, also had idle priority set, :eek: although tests were run overnight without any other active foreground programs running and no tasks such as AV scan scheduled. Processor is Core2Duo E6550 running without overclock.

Last segment was 351MB in size shrunk to 181MB.

[zacoz]

I'm considering reversing the default, but I'd like to hear others experience first. It could be just an AMD thing. If I find that to be the case, I may do some testing to determine the processor type and change the default based on that.

@Jdobbs

Did you end up reversing the default in v1.28.2?

I've added HC_NOSMP=0 for my own preference, just interested is all.

[jdobbs]

It currently defaults to NOSMP -- but the setting only applies for multiple processes. So:

If only one instance: SMP is always enabled
if > 1 instance and no setting or HC_NOSMP=1 then SMP is disabled for all instances
if > 1 instance and setting is HC_NOSMP=0 then SMP is enabled for all instances

I really wish I had named it HC_SMP instead. I don't like negative logic. I must have been brain-dead the day I added that parameter. :(

[linx05]

I really wish I had named it HC_SMP instead. I don't like negative logic. I must have been brain-dead the day I added that parameter. :(
Change it in the next release :)

[jdobbs]

Yeah, I can do that. But then I'd have everybody who added the old parameter mad at me when they've done a couple of encodes and realized it isn't working anymore.