DGNV HD footage VEL & fps Q's

[Midnyte]

I am not sure if this should be in the build thread or not, so I'm erroring on not.


I am wondering what the decode frame rates people are getting when using 1080p h.264 Bluray source that maxs the video engine's load ~98%.
Testing with 2021x64, I haven't had the incline to test this before.
I'm seeing ~37fps on 1 gpu of a GTX295. Is this typical?



This part probably belongs in the build thread, but is connected to this in my thought process:

If ~37fps is about the max that can be pulled out of the top 200 series gpu's VE for this source.
Would it be possible to leverage multiple gpu's VE's in an alternate frame serving configuration to increase served frame performance?

One of my many uses for DGNV is to create proxy edit footage from HD source, in this use x264 typically uses about 12-15%cpu as its is waiting for the data most of the time.
Once I fired up GPU-z and noticed that the VE was pegged I understood the low cpu utilization of x264.

I know this will probably be a low priority, and possible complex to add some kind of alternate frame serving feature, but it will be nice to use 2 or more of the gpu's VE's in my Quad-SLI setup to push frames.

[Guest]

Is that 37 fps serving through your script or 37 fps playing in DGIndexNV? If the latter, set the Disable Display option to get a true measurement.

And what is the frame size? That would not even be real time for 720P, which is absurd.

[Groucho2004]

I just tested a x264 1080p sample (about 8Mbps) with this simple script:

LoadPlugin("DGDecodeNV.dll")
DGSource("test.dgi")

I get about 100 fps with my GT 240 GDDR5.

I used a little cli utility I wrote myself to measure the framerate. If you are interested you can get it here (http://www.iol.ie/~schubert/AVSInfo.zip).

[Blue_MiSfit]

If your source really is a BluRay, then it's probably 1080p24, and uses CABAC, deblocking, and B-Frames, and probably has an average bitrate around 25mbps.

Remember, bitrate and decoding speed are inversely proportional :)

I think 8mbps giving 100fps on a GT 240 (really a G92 chip IIRC) is pretty reasonable. 37 fps on a GT200 GPU seems pretty slow IMO... though 37 is about 1/3 of 100, and 25mbps is about 3x more than 8, so I GUESS it makes sense if all my assumptions are reasonably close..

What happens if you run avs2avi on an avisynth script that's ONLY a DGMultiSource? (run avs2avi in null output mode to benchmark avisynth performance)

Derek

[Midnyte]

Yes, this was frames being served. Blue_MiSfit makes the speed make sence to me. I believe Index reported this m2ts at ~30mbps, if I remember correctly.
I will try avs2avi in null output to check for a difference, and get back to you.

[Guest]

I asked the frame size and you didn't answer.

[Midnyte]

Sorry about the frame size, source frame size 1920x1080p24, being resized to 640x360.

I am unable to run avs2avi with 2021x32.
It gives the nice "this program has stopped working" with a nice reference to DGAVCDecode.dll or nvcuda.dll. I find the DGAVCDecode.dll reference perplexing as I have never had DGAVCDec.

test.avs
LoadPlugin("D:\videotools\dgindexnv_32\DGDecodeNV.dll")
dgsource("G:\Source\Master\Movie\DTM.dgi", resize_w=640, resize_h=360)

avs2avi test.avs test.avi -c null -o n

Problem Event Name: APPCRASH
Application Name: avs2avi.exe
Application Version: 0.0.0.0
Application Timestamp: 40eb0522
Fault Module Name: nvcuda.dll
Fault Module Version: 8.17.12.5721
Fault Module Timestamp: 4c0d6d75
Exception Code: c0000005
Exception Offset: 00310b8a
OS Version: 6.1.7600.2.0.0.256.4
Locale ID: 1033
Additional Information 1: 0a9e
Additional Information 2: 0a9e372d3b4ad19135b953a78882e789
Additional Information 3: 0a9e
Additional Information 4: 0a9e372d3b4ad19135b953a78882e789

Problem Event Name: APPCRASH
Application Name: avs2avi.exe
Application Version: 0.0.0.0
Application Timestamp: 40eb0522
Fault Module Name: DGAVCDecode.dll_unloaded
Fault Module Version: 0.0.0.0
Fault Module Timestamp: 49b70852
Exception Code: c0000005
Exception Offset: 02c00b8a
OS Version: 6.1.7600.2.0.0.256.4
Locale ID: 1033
Additional Information 1: 0a9e
Additional Information 2: 0a9e372d3b4ad19135b953a78882e789
Additional Information 3: 0a9e
Additional Information 4: 0a9e372d3b4ad19135b953a78882e789

[Guest]

You're going way off topic for this thread. And don't post it in the DGNV thread either, as it is clearly a problem in avs2avi.

[Groucho2004]

Just use the tool I provided in my reply. It uses the AVISynth API properly and doesn't crash when Cuda is invloved.

[Midnyte]

You're going way off topic for this thread. And don't post it in the DGNV thread either, as it is clearly a problem in avs2avi.

I don't use avs2avi, I was asked to use that utility to get more information. I did not view that as off topic.
You do great work, I respect that. Why does it always feel like your out to start a fight? I know your busy, if you don't feel like responding you don't have to. I have assumed from the beginning that the performace is a direct limitation of the individual VE hardware.
I simply stated results I am seeing and asked if they were normal. From others responses I believe they are normal, it seems logical to me that a 30mbps stream would get 1/3 the frame rate of an 8mpbs stream. And hell its doing better then that, and a damn far cry from the 12-18fps the i7 was yeilding. Even with the current feature set, wich I know will expand, I am very happy with DGNV, you should be proud of it. It is a very excellent product.
Please accept my low priority feature request at least on the feasability study level.

@Groucho2004: thankyou your util runs great, and reporting about 37.5fps with the affore mentioned test.avs script. with 1 VE pegged @ 98%

[Blue_MiSfit]

Just as a sanity check, make a 720p60 stream, say 8mbps with CABAC, b-frames, and deblocking all enabled (defaults in x264) and see how your decoding performance is with that.

Derek

[Guest]

Why does it always feel like your out to start a fight? Wow, with 4 total posts and a month of membership, you're suddenly qualified to speak on what I "always" do. On the other hand if you were talking only about your feelings, well, maybe you just have a thin skin.

After your clarification, I see now that you brought up avs2avi because you were pointed to it to analyze performance. Had you said that politely, I would have said "OK, sorry, go ahead".

Regarding your feature request, I don't see any prospect for dynamically assigning picture decodes to multiple GPUs, obviously because that is not supported by CUVID. But no solution for it springs to mind either, given the need to access previous reference frames. How could they be available to multiple GPUs?