Running A New Comp But Having Trouble

[Mister Hatt]

Both MT and 64bit completely break on my machine, but that could be because I'm running scripts that sometimes hit 5000+ lines. The existing way of MT in AVS isn't really multithreaded though, and is detrimental to your picture. 64-bit is kind of a hack and requires way too much bullshit, I wouldn't bother. GPU offloading is also kinda dumb too given the amount of VRAM and the bandwidth of the interface, but eh.

[Pwolf]

Maybe VRAM but i would be surprised if the pcie bus would cause a significant bottleneck.

[HalOfBorg]

This is all SO far above me that I suspect you are all just talking gibberish and laughing at us, the stupid ones.

[Mister Hatt]

It does in some situations, although most people here wouldn't ONLY be doing math on their GPU so would require VRAM which bottlenecks at a much lower threshold. I've tried some motion estimation algos for large numbers of small data sets on GPU though and the bus got pretty tanked surprisingly. VRAM is the main issue here though, especially for 99% of AVS and encoding use cases.

[mirkosp]

Bro, this isn't even about 64bit and MT anymore, you're killing the source with that chain...

[EvaFan]

I was just doing a speed comparison with high settings. There's no way I'd do all that to that source, season 2 is very good looking already.

[mirkosp]

Ok, then that's reassuring, I was actually worried there.
Still wonder if you actually use spline36resize for resizing needs though. Considering there now is a gamma aware resizer for avisynth (ResampleHQ), I'd suggest switching to that. But I'm not sure if it's available for 64bit avs.
Another thing I'd suggest is switching to the dither package for debanding, since it doesn't have the gradfun2db bug and is actually a lot better quality wise and offers a lot more control (plus it's internally 16bit, and can actually produce 16bit output, which you can use in tandem with a higher bitdepth x264). I fear this one, too, is a 32bit only plugin for the time being, however there might be 64bit builds around for this one.
In terms of quality, I'd also suggest using QTGMC for bob/deinterlace and smoothadjust for tweaking needs (you should never tweak though, unless it's meant as an effect within the amv). These (well, smoothadjust and most actual plugins qtgmc uses) do have 64bit builds too (QTGMC's creator also tells how to use it for MultiThreading, though he also mentions it's not stable).
TL;DR: At the end of the day, it's your choice. I still do have reasons to stick with 32bit avisynth and single thread (and I can just run multiple encodes at the same time to cut down on total time, without the fear of stability and quality repercussions).

[Mister Hatt]

Dunno, I do scene based "threading" and it works plenty fast compared to MT/64bit. Then again, it's also kinda insane multi-system RPC (my attempt at making an MPP pipeline in AVS) so go figure it's a bit hacky still. I think threading has a place in AVS but there are smart and there are hacky ways to do it. Unfortunately, the smart ways are kinda difficult to get going.

[mirkosp]

http://mod16.org/hurfdurf/?p=234

This link is my new best friend.

[Mister Hatt]

Aha, yeah.