ATI vs. Nvidia in a Video Shootout

ThinSkin writes "ATI and Nvidia are well known for hailing their products as leaders in 3D apps and games, but little is known that both companies are trying to stake their claim in the video market as well. ExtremeTech is featuring an article that tests cards from ATI and Nvidia to determine who takes the cake in video quality and performance. Using CPU utilization scores and visual quality comparisons during video and DVD clips, the author concludes that ATI's latest generation of GPUs have an edge over Nvidia, particularly in DVD playback and with video acceleration."

Comments

New algorithm

[2.7182]

I thought that Nvidia had the edge because they are using the new fast subdivision algorithms of Jean Gallier at Penn CS dept.

Video on Linux

[Anonymous Coward]

What is the state of video on Linux?

I would love to see a comparison of performance and video quality of these same cards on Linux. Do the drivers even support any of this functionality? Is CPU usage similar?

Honestly

[asv108]

I don't even look at ATI anymore when building a system for my own use. Nvidia has had excelent Linux device driver support for a number of years now. The last few personal systems I built were nvidia dualhead systems running linux, and I have never had a driver problem.

My latest system is dualhead dual-dvi pci-express 7800GT system running on Ubuntu. I was expecting the video configuration to be a major pain the ass, but everything worked well.

Until ATI has the same level of Linux support, I will not take their products under consideration.

Not Microsoft's fault

[MobyDisk]

If you want your video to look its best and run as fast as it can, you have to enable all sorts of settings in the advanced properties of your player (or players, plural), and those settings can be different between ATI and Nvidia cards.

Yuck.

In short, Microsoft needs to seriously clean up this mess. Video codecs need to hook into a common framework, one that the graphics cards manufacturers can target for acceleration without needing to work with every individual codec maker on the planet.

This is an interesting statement, because the author just described exactly how DirectVideo works. Each step in the decoding process is a pipeline, and a "codec" can plug-in to this and provide whatever steps in the process that it can do best. For example, if playing a video looks like this:

Read a DVD -> Reading a file -> Decrypting -> Decompressing -> Motion compensation -> YUV2RGB -> Deinterlacing -> Scaling -> Displaying on video device -> ATI X1800

There can be a separate component registered for each step. Or many. And DirectVideo can determine which one is the most appropriate for the given input, output, and hardware configuration. So if you video card supports hardware YUV2RGB scaling, then it will do it. If not, the software can.

The problem is partially that crappy companies get in the way. I downloaded a codec so I could view DV files, and it registered such that all video types were DV. This is a common scenario that requires a purely brain-dead programmer:

boolean IsThisTheProperCodecForThisVideoType?(string videoType)
{
// TODO: Look at type code and see if it is a DV file
return true;
}

ATI Linux

[DaCool42]

For those of you wondering about linux drivers - ATI's fglrx linux driver works fairly well (I use it to play HD .ts files on a Radeon 9800 pro). The only problems are lack of support for xvmc, and some problems with dual head (confusing config, xinerama issues). I don't have any performance issues with full bandwidth 1080i content and 5.1 sound running on a 720p display (video de-interlaced with mplayer's halfpack filter).

A counter point

[lakeland]

While Nvidia's closed-source drivers are clearly better than ATI's, the opposite is true of the open-source drivers. If you are looking to build a system without binary drivers, or are using non-x86 and so cannot use the provided drivers, then you're better off going with ATI.

I imagine this is no coincidence, how many people can be bothered working on the nv driver when the nvidia driver works so well... But it does worry me how easily we have come to accept binary drivers now that they work so reliably for 90% of the users.

Re:VLC versus Elecard for HDTV

[Xesdeeni]

Yeah. Comparing 2-D playback of DVDs in 2006, is like comparing 3-D frame rates using Quake II...passe.

Riddle me this Batman:

1. Can the card accelerate MPEG-2 playback (DxVA, et al)?
1.a. How much CPU is necessary to play back HD content (720@24p, 720@60p, 1080@24p, 1080@30i) without dropping frames?

2. Can the card accelerate MPEG-4 (h.264 part 10) playback?
2.a. How much CPU is necessary to play back HD content (720@24p, 720@60p, 1080@24p, 1080@30i) without dropping frames?

3. Can the card accelerate WMV (VC-1) playback?
3.a. How much CPU is necessary to play back HD content (720@24p, 720@60p, 1080@24p, 1080@30i) without dropping frames?

4. Can the card accelerate MPEG-2 encode?
4.a. How much CPU is required to get real-time encode (i.e. 1 hour of video takes 1 hour to encode)?

5. Can the card accelerate MPEG-4 (h.264 part 10) encode?
5.a. How much CPU is required to get real-time encode (i.e. 1 hour of video takes 1 hour to encode)?

6. Can the card accelerate WMV (VC-1) encode?
6.a. How much CPU is required to get real-time encode (i.e. 1 hour of video takes 1 hour to encode)?

7. Can the card synchronize 1080i video with 1080i display (i.e. the field synchronization between the decoded video and played video don't drift - hint, neither ATI nor nVidia can do this today)?

Xesdeeni

What about Matrox?

[Xabraxas]

2D quality on Matrox cards is outstanding. How come we couldn't get a comparison with on of their cards. I have a Parhelia laying around here somewhere but unfortunately it's not quite working anymore (the screen is a nice shade of pink).