Probably old news: another MPEG-2 GPU render bug

johnmeyer wrote on 8/19/2014, 1:27 PM

I just sent this to tech support, but I thought I'd post it here as well. This problem happens with Vegas 13.0, build 310.

I have a simple cuts-only 50 minute wedding video using video from my Sony CX700V camcorder which was recorded using the 1920x1080 60p setting. Here is a link to a very short sample of one clip

Download Sample Clip

When I render this using the standard MPEG-2 DVD Architect Widescreen template, I get corrupt video. I used Womble to cut out a small portion of the 50-minute MPEG-2 file in order to provide an example of the horrendous corruption I am seeing. Here is that clip:

Download Example of MPEG-2 Corruption

I think similar things have been reported before.

In my attempts to find the cause of this bug, I discovered two workarounds. One workaround is to first render everything to Cineform. This works, but obviously it takes more time, and I happened to be low on disk space, so this wasn't a good solution for this project.

I then went to the Options --> Preferences --> Video tab and unchecked "GPU acceleration of video processing."

This solved the problem.

So I think this is yet another GPU bug.

BTW, I did try to reduce this down to a small test case, so I could send the media files to Sony tech support along with a VEG file. However, when I made project smaller by deleting most of the media on the timeline, keeping only the portion of the project where the problem appeared in the rendered video, the resulting twenty second render resulted in good video. Thus, the problem appears be related to what video surrounds the problem area.

Just to be totally complete about my setup and test conditions, I did make one set of modifications to the default DVD Architect MPEG-2 template: I reduced maximum bitrate from 9,500,000 to 8,300,000; I increased average bitrate to 7,000,000; I increased minimum bitrate to 512,000; and I enabled "two-pass" encoding.

My video card is an NVidia GeForce 9800 GT. The driver version is 285.62. I would like to update this, but the NVidia site is very confusing as to what driver I should use.

Other relevant information, in order to satisfy the "system spec police:"

Operating System
Platform: Windows 7 (64-bit)
Version: 6.01.7600

Processor
Class: Intel(R) Core(TM) i7 CPU 965 @ 3.20GHz
Identifier: GenuineIntel
Number of processors: 8
MMX available: Yes
SSE available: Yes
SSE2 available: Yes
SSE3 available: Yes
SSSE3 available: Yes
SSE4.1 available: Yes
SSE4.2 available: Yes

Display
Primary: 1680x1050x32

Memory
Physical memory: 6,135.1 MB
Paging memory available: 12,268.4 MB

Comments

OldSmoke wrote on 8/19/2014, 2:11 PM

I downloaded your short sample but I cant reproduce the problem even with GPU on. I actually cant see what the problem is; there is something wrong with your link. Try and drop the "s" from "https" and also avoid spaces or special characters in your path and file names.
As for Nvidia drivers, I have very good success with 296.10, 334.89 and I am currently on 335.23 which is version 21 of 334.89. All of these drivers support the GeForce 9800 GT as well and I would give them a try starting from the oldest.

Grazie wrote on 8/19/2014, 2:13 PM

John, if you can rummage about in the SCS video card suitability, see where your card sits? Seeing that turning off GPU may just indicate its suitability.

Cheers

Grazie

johnmeyer wrote on 8/19/2014, 3:25 PM

I downloaded your short sample but I cant reproduce the problem even with GPU on. Yes, that is what I said in my initial post: I was unable to reduce the problem by only having one or two files on the timeline. This was true even if I used just those clips that match the portions of the original 50-minute MPEG-2 file that had the problem. So, you cannot reproduce the problem with just one clip.

Then why did I upload this clip? I did so to avoid any questions or any confusion as to exactly what media I was using. I can't begin to count how many posts in this forum the past few years waste time while people ask for system specs and exact media specifications, even when the OP provides sufficient information. So, I was simply trying to avoid repeating that dance.

I actually cant see what the problem is; there is something wrong with your link. Try and drop the "s" from "https" and also avoid spaces or special characters in your path and file names.Done. I removed the "s" and also made the links clickable.

As for Nvidia drivers, I have very good success with 296.10, 334.89 and I am currently on 335.23 which is version 21 of 334.89. All of these drivers support the GeForce 9800 GT as well and I would give them a try starting from the oldest.Thanks. That is very useful. The NVidia support site does a poor job of advising what driver to use. For instance, if I enter the name of my card, it suggests a very, very old driver. However, if I poke around their support forum, many people suggest driver versions similar to what you are recommending. If I then go to the download link for that later version driver, sure enough, the list of supported cards includes my GeForce 9800T. Since the recommendation provided by the NVidia driver download site doesn't mention some of these later drivers, I have been a little reluctant to install them. Based on your input, I'll go ahead and try the drivers you mention, although I think I'll start with the latest one and then work backwards rather than the other way around.

Seeing that turning off GPU may just indicate its suitability.I don't understand what you are trying to say. I think you are suggesting that, since I solved my problem by turning off the GPU, that this must indicate that the GPU is not suitable. However, since the GPU is supported by Vegas (it reports that "CUDA" support is available), Vegas thinks it is suitable. On the other hand if all you are saying is that since GPU rendering doesn't work, then it must be that Vegas and my GPU don't correctly work with each other, that is that basic point of my original post, so I don't see the point of your comment.

Grazie wrote on 8/19/2014, 3:35 PM

Is your card one of the cards that gets support for GPU acceleration? Wouldn't that you turning off GPU indicates that it isn't?

Grazie

OldSmoke wrote on 8/19/2014, 3:48 PM

The 9800 GT is supported, no issue there. I didn't have any issue at Nvidia.com to find which driver will suite your card. Go to http://www.geforce.com/drivers and do a manual search for Geforce, 9 Series, Win 7 64bit.

It could very well be a driver issue but if you get it working, don't expect too much from it. Your CPU is quite powerful, 4 core 8 threads and you will need at least a GXT560Ti to get something going. I used a GTX470 with a 4 core 8 threads i7-2600K before and the gain was in the area of 20% with GPU on. The combination of CPU and GPU is a bit like a putting two horses in front of the carriage; if they are not sort of equal the result will be not so desirable.

As for you problem, I do have projects with 1080 60p MTS files but they are from a Canon HF G30. I have no issue with those and I doubt they are very different from the Sony files. Maybe you can provide more short clips? I still get nothing when I click on the bug file.

Grazie wrote on 8/19/2014, 3:53 PM

Ah clarity! Thanks OldSmoke. Yup, I got the GTX560ti, it's in my system specs along with the driver. This need to get the correct combo has driven many of us up the wall.

Grazie

johnmeyer wrote on 8/19/2014, 4:04 PM

Is your card one of the cards that gets support for GPU acceleration? Wouldn't that you turning off GPU indicates that it isn't?I just answered that. No point posting any further ...

The 9800 GT is supported, no issue there.Tell that to Grazie.

It could very well be a driver issue but if you get it working, don't expect too much from it. Your CPU is quite powerful, 4 core 8 threads and you will need at least a GXT560Ti to get something going. Yes, I totally agree. I made this point many times about 6-8 years ago, before Sony started providing their initial GPU support in 8.0d (the first 64-bit incarnation). My point was that CPUs were not only getting more powerful because of multi-core, but that many of the instructions added to the CPU were very similar to those initially only found, back then, in DSPs, a term which preceded GPU. Nowadays, with a good Intel multi-core processor, its not clear how big a performance gain one can get by using the graphic card's GPU. I made these posts because I was arguing for Sony NOT to provide GPU support and instead use that programming resource to add new editing features. Other people, when responding, predicted (correctly, as it turned out), that GPU code would add all sorts of support issues.

I've seen the various benchmarks done in this forum which show rendering performance gains with GPU enabled, but many of these benchmarks use generated media, in order to make it simpler to download the project. For real-world rendering, when using actual video files, I'm not sure how much can be gained in a system like mine, when enabling the GPU.

So, I'll definitely download the newer NVidia drivers, and then do some tests, but if I don't see some huge improvement in performance, I'll just live the GPU off and get on with life.

I wish I had a nickle for every post in these forums the past six years where the problem was eventually traced to bugs in the interaction between Vegas and various GPU/driver combinations. The only reasons I had it enabled in the first place was that was the way Vegas' installation program initially configured itself, and also I didn't think that something as well-established as MPEG-2 rendering could possibly have such fundamental, awful bugs in it after all these years.

The one thing I know for certain: any increase in speed that comes at the expense of any sort of instability or quality is a bogus "improvement," and must be avoided.

I still get nothing when I click on the bug file. I don't know what else to do. I just tested the link by first signing out of my Dropbox account, so that I "look" just like you (an anonymous user). I clicked on the link in my revised first post above, and was able to download.

OldSmoke wrote on 8/19/2014, 4:26 PM

There are great gains with GPU acceleration and I wouldn't be able to handle my 4K files without it. GPU acceleration is not only applied to rendering or generated media, there is a great deal it does for everyday work too. However, and I said it before, it is important to understand that only a good combination of CPU and GPU will work to your advantage. If you are in the US, a $150.00 can get you a GTX580 off eBay and you will see the difference it makes; provided you have a sufficient big power supply for it.
I cant blame SCS for going this direction as others have done so too; you just have to keep up with the competition. GPU acceleration, like it works for me, is a low cost upgrade compared to a multi CPU system with 2x 6 core Xeon as a minimum.

Edit:
I used my iMac to have a look at the link and it worked; it could well be an IE issue. Anyway, the problem starts with the first transition and those are GPU accelerated. I would try and update the video driver or even try and reinstall the driver. Is this the first time you are using a VP version that fully supports GPU acceleration? I also doubt that VP8 had GPU acceleration already. As far as I know, it started with VP10 and was only supported for rendering via Sony AVC...but I could be wrong too.

Grazie wrote on 8/19/2014, 4:37 PM

Excellent points OldSmoke.

I seem to appear to have hit a number of low-cost sweet-spots. I have TWIXTOR now GPU-ed, and I'm able to get really good Previewing.

Grazie

videoITguy wrote on 8/19/2014, 6:48 PM

johnmeyer, Mpeg-2 encoding bugs have appeared in past versions ( particularly V12Pro) because of errors made by SCS programmers in the templates for render - and this should never be ascertained as symptomatic with having GPU on. GPU issues are certainly more related to processing for which GPU is designed to assist.
Mpeg-2 is not!

If you are getting cross effects from your card driver, memory handling, the physical bus that the card is mounted in etc...this becomes a whole different issue for you to pursue. leave GPU off in your workflow.

NCARalph wrote on 8/19/2014, 7:01 PM

You might try setting your Preview Dynamic Ram setting in Preferences to 0. That's the only way I've been able to render using my NVidia 570 on Vegas Pro 13.

In my case, short clips would render be longer ones didn't and failed at different points during the rendering so your results aren't unique.

johnmeyer wrote on 8/19/2014, 8:55 PM

SCS is already responding to my bug report. I am therefore trying to come up with a way to reproduce the problem by using the same media, over and over, on the timeline.

The problem seems to happen during those moments where I'm pretty sure that the encoding bitrate spikes way up. I assume there is a tool that would give me a readout of instantaneous bitrate in an MPEG-2 file, but I don't have one. However, most of the problems happen when there were multiple strobe flashes from the still photographer, as well as during crossfade transitions. Both of these generally demand more bits from a VBR encoder.

If you are in the US, a $150.00 can get you a GTX580 off eBay and you will see the difference it makes; provided you have a sufficient big power supply for it.I wouldn't mind spending some real money to upgrade. If money were not a huge object, is the GTX580 the one to get? I have a 700W power supply, and with all cores pegged during a render, I'm using less than half of that (based on an actual measurement of power draw from the mains plug). It is spec'd at 240W vs. my 9800 GT at 105W. NVidia recommends at least a 600W supply.

Is this the first time you are using a VP version that fully supports GPU acceleration? I also doubt that VP8 had GPU acceleration already. As far as I know, it started with VP10 and was only supported for rendering via Sony AVC...but I could be wrong too.No, I have VP10 & VP11, so I've rendered (and previewed) using GPU before.

I think you are right about V8.1 not having GPU. I just looked back at some of my old posts, and the reason I thought V8.1 had GPU is that it was the first 64-bit version of Vegas, and because of this, promised significant reductions in rendering times because of that new architecture, much the same way that GPU later promised to make renders and timeline preview faster.

Mpeg-2 encoding bugs have appeared in past versions ( particularly V12Pro) because of errors made by SCS programmers in the templates for render - and this should never be ascertained as symptomatic with having GPU on. Are you saying that there are specific settings in the MPEG-2 render templates that may result in bad renders? The only thing I have spotted is that SCS made a huge change in the MPEG-2 maximum bitrate when rendering using VBR. The old maximum was 8,500,000 bps, and they increased that to 9,500,000. I don't see how that can work when you combine that with any sort of decent audio, so I always reduce this to 8,300,000.

OldSmoke wrote on 8/19/2014, 9:10 PM

[I]I wouldn't mind spending some real money to upgrade. If money were not a huge object, is the GTX580 the one to get? I have a 700W power supply, and with all cores pegged during a render, I'm using less than half of that (based on an actual measurement of power draw from the mains plug).[/I]

For me it was. I have used GXT470, GTX560, 2x GTX570 and now 2x GTX580. I am also particular about brands, ASUS or EVGA is what I use. My 2x GTX580 plus the i7-3930K running on all "cylinders" are drawing around 550W so you should be fine with a single GTX580. The difference between 470 and 560 was marginal, 10% improvement. From 560 to 570 a good 15% and then another 15% with the GTX580; this is all with single cards. Today, the difference between CPU only and GPU on is like day and night.

Aside from Nvidia there are also great cards from ATI/AMD like the HD6970 which are actually better suited for OpenCL support like timeline preview requires.

Nevertheless, you first should try and get your system running with the 9800 GT; it easier to upgrade from there.

In case you missed the "old" article that came with the SCS Benchmark project, here is a link: http://dl.dropboxusercontent.com/u/39278380/VP11_Benchmark.pdf

johnmeyer wrote on 9/2/2014, 8:40 PM

SCS told me to upgrade V13 to the latest build, and to also upgrade my nVidia driver. Based on OldSmoke's advice, both here and in other recent threads, I didn't upgrade to the newest nVidia driver but instead downloaded and installed version 334.89.

The timeline performance still stutters when the camera pans, but is otherwise smooth (this is when viewing Sony CX700V 60p 1920x1080 AVC-encoded video). However, when I rendered the project to MPEG-2, using the same template as before, I got good video. Prior to do this render, I re-enabled the GPU in the Vegas preferences.

So, I have my fingers crossed. I'll still be doing lots of checking after doing renders, until I'm convinced I've solved the problem.