DGAVCDecNV 1.0.13: GPU decoding on Nvidia

[Guest]

At the bottom left of the screen is a line:

Don't have a PayPal account?

At the end of the line below that is a link:

Continue

Hit that link and you'll be able to donate by credit card.

[screw]

Sorry, but when I do open your "Donate" link, for me there is no such a choice as you did describe:
http://img515.imageshack.us/img515/4928/ppnolinkzh2.png

May be there is something wrong with my browser (IE) or firewall.

I shall try to find another solution than.

Screw

[Inventive Software]

Clicking the donate link on the relevant page on neuron2.net brings up this: (UK, FF3)

[Guest]

Can't see upload sites at work. You'll have to wait until tonight for any comment, or else attach the image instead and I can approve it right away.

[DarkZell666]

@Inventive Software & Screw : Paypal shows me the specific link neuron2 mentions, and clicking on it points me the same page you both have; ie: a page to pay with a credit card without an account

It seems depending on what cookies you have in your cache, the website doesn't display the same info

[rebkell]

Originally Posted by neuron2

Can't see upload sites at work. You'll have to wait until tonight for any comment, or else attach the image instead and I can approve it right away.

Here's the image as an attachment:

[screw]

Well, I do guess, it more likely depends from which country I am reaching this PayPal page. Possibly PayPal does not accept direct payments from some countries, like Latvia.

OK, I shall find the another way.

Screw

[DarkZell666]

Originally Posted by screw

Well, I do guess, it more likely depends from which country I am reaching this PayPal page. Possibly PayPal does not accept direct payments from some countries, like Latvia.

OK, I shall find the another way.

Screw

Screw, the page you're looking at actually allows you to pay WITH a credit card and WITHOUT creating an account.

[ReferenceDivx]

It looks like things are progressing quite nicely. Neuron2 i just sent you $15. Its the first time I've paid for software in sometime. Can't wait to see whats next from you. I hope to do an analysis of power consumption with and without your software in the near future.

[livetoflyfish]

Alright, both of my systems:

Q9300, Quadro FX 570 w/ 256mb
E6600, Geforce 8500 GT w/ 256mb

performed in the exact same manor. They both indexed BD AVC material. With a simple avsynth script loading just the indexed file, the full res video played choppy in Media Player Classic. They both encoded material using virtualdubmod and WMV9 VCM. Trying to encode video either through Windows Media Encoder or through WMV encoder scripting crashed the program. Why WMV? Simply because I've been doing it for years and I know it.

I have money burning a hole in my pocket to go get an 8500 GT with 1Gb of memory. I'll probably hold off while to see how this progresses:

BTW, I am working on a solution such that all instances use a single instance of D3D and float the CUDA contexts. If this succeeds it will have the useful consequence of dramatically reducing memory needs for multi-instance.

[ReferenceDivx]

Since megui is not working yet i created my own 2pass x264 launcher program. You can get the program and source code here. http://www.hddvdstrikesback.com/vers/x264launcher.zip

Example command line
x264launcher.exe c:\movie\ 8000 1
x264 must be in C:\Program Files\megui\tools\x264\
The output directory must contain test.avs

The second argument is the desird bitrate
Select 0 for standalone hd-dvd fast and 1 for standalone hd-dvd
Select 2 for standalone Bluray fast and 3 for standalone Bluray
Example of useage below for hd-dvd fast
x264launcher c:\input 8000 0

[saemonz]

Originally Posted by ReferenceDivx

Since megui is not working yet i created my own 2pass x264 launcher program. You can get the program and source code here. http://www.hddvdstrikesback.com/vers/x264launcher.zip

Example command line
x264launcher.exe c:\movie\ 8000 1
x264 must be in C:\Program Files\megui\tools\x264\
The output directory must contain test.avs

The second argument is the desird bitrate
Select 0 for standalone hd-dvd fast and 1 for standalone hd-dvd
Select 2 for standalone Bluray fast and 3 for standalone Bluray
Example of useage below for hd-dvd fast
x264launcher c:\input 8000 0

What type of your NV display card? How fast pass2 when you use standalone Bluray?

[ReferenceDivx]

I have a 8500gt and a q6600 cpu. I can encode the 2nd pass of the Bluray standalone profile at around 6.15fps. FFdshow give me about 5.90fps. The first pass is about 10-13% faster with dgavcdecnv though. I'm still in the process of testing.

These figures are for full 1080p24 input and output, with no filtering.

[saemonz]

We have same hardware, Have you try to disable display.(This option is intended for getting an idea of the performance of the GPU's decoding engine. When this option is enabled, all display operations are disabled except for one: the FPS field of the Info dialog is updated. If you load your stream, select this option, and start a preview, the FPS field will show the raw decoding rate of the GPU decoder as implemented in DGAVCIndexNV. Note that the conversion from YV12 to RGB24 used when display is enabled is very CPU intensive. So if you want to assess the frame rate and CPU utilization of the GPU decoder, it is necessary to disable the display.)



CPU usage 99%？

[ReferenceDivx]

1080p has over twice the number of pixels as 720p. So I imagine that it would be about half the speed as 720 encoding.

[Guest]

My findings so far on the problem with crashing in MEGUI, etc...

It appears that all the managed applications (you know, the .NET crud) crash when trying to allocate a second D3D instance. The crash is within the Microsoft D3D CreateDevice() function. I can't trace it and there is nothing useful on the web about this issue.

Using floating CUDA contexts is not a viable solution because they must all be within one process. If the encoder applications use multiple processes, we're screwed. In any case, it seems like a better system-wide solution to have a HW resource support separate client processes.

So, my plan is to package the GPU decoder as an out-of-process local COM server. That way, only one D3D instance is ever needed. I know that the first instantiation always succeeds with the managed applications; for example, the preview screen works in MEGUI but it crashes when you enqueue the job. A useful side effect will be a reduction in memory requirements. And it will be valuable to have a COM GPU decoder that can be re-used by other applications!

So it may take some time to get this working as COM programming is new for me. I do have a simple server and client working, however, and I am now plugging in the decoder. The frame will be returned to the Avisynth filter in shared memory.

We'll just have to see if this solves the problem, as it is hard to predict what gotchas might arise.

[squid_80]

Just for kicks, try testing with a CoInitialize(0); call directly before CreateDevice and see if that helps at all.

[ultratoto14]

Nice to see that you detected the problem, continue you're good work.

[Guest]

Originally Posted by squid_80

Just for kicks, try testing with a CoInitialize(0); call directly before CreateDevice and see if that helps at all.

It doesn't help.

[crypto]

Originally Posted by neuron2

So, my plan is to package the GPU decoder as an out-of-process local COM server. That way, only one D3D instance is ever needed.

Out-of-process COM servers are factor 10 to 16 times slower than inproc. Where will be the interface level? I suspect that if you plan to transport decoded images over COM, you will lose all speed advantages of hardware decoding.

I am still not sure what the problem with MeGUI is. I can run multiple instances of vdub GPU AVS scripts. What is so special with MeGUI?