my personal guide for svp usage (Page 1 of 2)

cool I learned something new.
Is it as good as CUDA?

When I put LAV Decoder as preferred ffdshow disapears and SVP don't use the smoothness. Any tips on it?

Ok if I do that I get the same GPU ussage than If I leave all external filters in blank and I use normal ffdshow (comparing it when SVP is not working), so is really DXVA from LAV Filter working? I'm using for testing a Radeon HD 6670 12.2 drivers. Or maybe it should not increase GPU usage at all like this and still using DXVA?

I'm trying it on media player classic, cause on Pot player I have many crashes, this system has 2 Xeons based on i7 a total of 16 threads, maybe it's using too many memory, I need to disable HT in order to have less threads and fix svp to a fewer threads or it increases memory ussage too much.

Umberlee
PotPlayer and LAV decoder are both very like "to eat" memory more than other players and decoders.
You can check it in Performance graph from SVP.

And of course more threads takes more memory.

About GPU load. I recommend to use GPU-Z with 3 indexes of different graphic blocks load (GPU, Memory controller, Video engine).

extraeye
LAV codec is the best in many nominations. Best codec used much memory too.

So is Coreavc prefered?
Yes. If you have NVidia card. I use CoreAVC when I need more free memory to my SVP-script.

Umberlee
You're doing something wrong 

My results with 1-core SandyBridge with HT off (avg CPU load in %)

ffdshow QuickSync: 40%
ffdshow software: >100% and freezing
LAV DXVA + ffdshow raw: 40%
LAV QickSync + ffdshow raw: 50%
LAV software + ffdshow raw: >100 and freezing
PotPlayer DXVA + ffdshow raw: 20%

And when it comes to all 2+2 HT cores on i5-2430M:

ffdshow software: 70%, up to 95% sometimes
Any HW decoder including LAV DXVA + ffdshow raw: 10-15%

Umberlee
The bad thing is that when I use SVP I got little stops with this configuration with LAV DXVA, a thing I don't have when I remove all the external filters and I use exclusevely ffdshow. Then SVP runs smooth like always.

Yep, that's the problem with weak ATI/AMD cards. I've got the same on both i5-2430 + 6450 and Phenom II X6 + 5670.
BUT this's only for full 1080p. Cropped BD-rips 1920*800 are played smoothly with DXVA too.
I don't know what about low-end NV cards 

I use Pot + LAV... but I want to get the max of my computer (Im decoding with CPU not GPU)

What videoplayer and decoder do you recommend?

Umberlee
when I use SVP I got little stops with this configuration with LAV DXVA
Yes. It is framerate's hell.
I use LAV CUVID or CoreAVC with None acceleration. Only these two decoders work perfect to me. With smoothness.
I check by EVR Custom Statistic by Ctrl-J. The grapic must be parallel to other lines. See picture from this post.

Umberlee
maybe on weak VGA cards it's not that good to use SVP + LAV DXVA
Yes. I have Quadro 600 on my work computer with i5-2300 processor. And I can play video smoothly only with software decoding if this video is FullHD. It is hard to this videocard to do hardware FullHD-decoding and SVP's GPU-acceleration together.

travolter
Fast measurements results:
LAV DXVA2 + MPC-HC = 865 MB
LAV CUVID + MPC-HC = 836 MB
LAV None + MPC-HC = 778 MB
ffdShow + MPC-HC = 771 MB
CoreAVC CUDA + MPC-HC = 734 MB
CoreAVC None + MPC-HC = 726 MB
CoreAVC None, Low latency + MPC-HC = 709 MB

Total savings: 865 - 709 = 155 MB

You can do it for another decoders and players.

@MAG79

Thanks for the stats . Ill be using coreavc and ffdshow now.. meanwhile Ill test different players. I like Pot because it contains great built in splitters but I have to check what amount of memory I  can save Vs MPC-HC

I use x86 AVX: http://www.mediafire.com/?ceodtrhjc20qb1d this build of MPHC,LAV,DXVA2 COPY on FX-4100,ATI 6570 1MB
and works good.  Thread is at Doom 9 MPHC tester build.