Video Quality Analysis Part 1 with NVIDIA GeForce October 17, 2005

How Deep Does the Rabbit Hole Go?

Page:: ( 1 / 19 )

The question is why. Should we even be doing this?

No, I don’t mean the question of whether or not we should share links to the community. I’m asking: why are we all trying so hard to get to the truth? Children in grade school can give you the quote “ignorance is bliss” but we can actually understand why that’s true. Imagine if you didn’t know what you do about technology. You would go to Best Buy and get the graphics card with the coolest box, and when you discovered that the 32-bit PCI graphics card you just bought only runs Half Life 2 at 10 fps at 640x480, you wouldn’t get pissed, but would just assume that choppiness was the state of PC gaming. You would be beaming after buying “professional” speakers from two guys in a white van, and you would actually be excited about getting a Dell. Imagine if there was no Internet and no review magazines. You’d be happy just because you didn’t know any better.

Like Neo in the original Matrix film, we are all given the choice between taking the red pill or the blue pill -- the choice between learning the truth and facing the consequences of that knowledge or choosing to continue with our lives as we have been, finding comfort in believing whatever we want to believe.

I’m going to make you an offer. If you choose to you continue on, I will show you just how deep the rabbit hole goes. I can’t guarantee that you’ll like what you read. Some of you may not even want to believe me or will even try to downplay its significance. All I can offer you is the truth. What you choose to do with it is up to you.

If you choose to quit now and stop reading, that’s OK. You’re not going to die. I’m not going to reveal any life-altering truths. You’ll be happy and return to the life you had before you encountered this article. You may even end up spending less money. Lao Tzu once wrote: “To know that you do not know is strength. Not knowing that you do not know is sickness. The sage is sick of being sick, therefore he is well.” As long as you know that this article will be here for you when you’re ready, that’s OK with me.

So what your decision?


SIDEBAR: Overly elaborate introductions are my specialty.

The Basics

Page:: ( 2 / 19 )

However, when it comes to video performance and quality, all of a sudden most people suddenly assume that it’s all the same, or that you’d have to be a videophile with a trained eye to see a difference, or that the company that provided the best video quality 10 years ago is still the company with the best video quality today. After all, as long as it’s “ISF certified” it’s good, right? What this article is going to show you is that

1. There are substantial differences in video quality between different products
   
2. You will notice the difference
   
3. You will care about the difference
   
   

Pop Quiz

Page:: ( 3 / 19 )

For each of these questions, make sure you’re looking at the full resolution pictures and not just the thumbnails.

[image]

Pop quiz question 2

Page:: ( 4 / 19 )

[image]

Question 3

Page:: ( 5 / 19 )

[image]

Quiz question 4

Page:: ( 6 / 19 )

[image]

Pop quiz answers

Page:: ( 7 / 19 )

When it comes to DVD movies or 1080i HDTV, the ultimate arbiter of image quality ends up being the quality of the video processor which handles deinterlacing and optionally, noise reduction.

If you do not deinterlace at all, you may run into the artifacts seen in the first question of the quiz. Believe it or not, I've seen the $12,000 Qualia TV run into problems like this with HD content at the Costa Mesa Sony Style (presumably an issue with improper settings). This is actually a cadence test rather than a pure deinterlacing test, but we can go over that later.

The difference between proper deinterlacing and improper deinterlacing can mean the difference between maintaining a full-resolution image and one where half of the data thrown away. If you've heard people say that 1080i looks worse than 720p, it's only because their deinterlacer is throwing away half of the data to produce 540p! (1920x540 vs 1280x720). Movies that are broadcast at 1080i allow full reconstruction of 100% of the data (1080p at 24 fps). This is what is shown by the Antarctica pictures. Although the lower resolution screenshot is still superior to DVD, it's a step back from what is truly available in the source.

The third set of test images shows you that even if you're dealing with a very high-quality progressive scan source such as 720p HDTV, a high-quality video processor is still necessary for removing the random noise (grain) that was introduced either in the original source recording or at the time of MPEG-2 compression.

The differences between video processors for image quality are big. With TVs getting bigger and cheaper, visual problems are only going to be magnified. Unless you're a person who only watches movies on a Sony PSP, you should care about the quality of the video processor.

Deinterlacing 101

When you look at a still image on a PC monitor, plasma or LCD panel, or DLP/LCD projection TV, you are looking at a “progressive scan” image. This means that the display updates all of its pixels simultaneously. The problem is that all DVDs and 1080i HDTV are "interlaced," so they transmit only half of the image, alternating between the odd and even lines. We call each "half image" a field. So for a 720x480 image, the DVD stores it as two blocks of 720x240 pixels.

The challenge is reconstructing the original image as faithfully as possible. It's not an easy task. You can always throw faster hardware at H.264 decoding or MPEG-2 decoding, but hardware is only part of the problem when it comes to deinterlacing. Yes, deinterlacing requires significant computational horsepower, but it is also a challenge of algorithms. Figuring out how to reconstruct the image in the best possible manner is substantially more difficult than making sure you have enough hardware to back it up.

Testing the video processors

Page:: ( 8 / 19 )

HQV (Hollywood Quality Video) is a marketing brand-name for Silicon Optix's video deinterlacing technology. Some of you may have remembered me talking about the company in our Ultimate Home Theater PC system build a few months ago. Their technology is based upon Lockheed-Martin's military imaging technology and really represents some exotic stuff. The "HQV Benchmark DVD" was engineered to make testing of video processors easier by having all of the necessary test patterns on a single handy disc.

Anytime you have a manufacturer-developed benchmark, it's important to evaluate the test for bias. The HQV Benchmark DVD was designed to showcase the strengths of the Silicon Optix's products. So, if you were able to develop a video processor that was better than Silicon Optix Realta, the DVD would not be able to show you that difference: they'd both get perfect scores even though one may be better than the other in real-life. That's a clear limitation of the benchmark.

That said, it turns out that because HQV reflects the top-of-the-line, no-compromise video processor, the tests used to showcase HQV technology end up being an enormous challenge to "normal" video processors such as the kind you find in DVD players that don't cost $3000, or video cards from ATI and NVIDIA. So for these types of products, the HQV Benchmark DVD is a great and honest test that brings objectivity to video quality. That's why the benchmark has been adopted by a wide range of testers, ranging from the techies at AVSForum.com and Audioholics.com to the reviewers at Widescreen Review and the Perfect Vision. I know that NVIDIA's PureVideo team has used the benchmark in-house, and I'm sure ATI's team has as well.

Recently NVIDIA commissioned an independent 3rd party to use the HQV Benchmark DVD to compare NVIDIA's performance against ATI's. (Click for PDF document here). There's nothing wrong with a company commissioning a study. If NVIDIA believes in the superiority of their technology over the competition, why not try to get the word out?

The problem is that the HQV Benchmark DVD is only good when the reviewer is scoring the tests properly. Although the company has a distinguished record when it comes to evaluating traditional PC hardware such as CPUs, Principled Technologies has made some significant errors in this commissioned study. I'm not sure how much Principled Technologies understands video technology since they inexplicably talk about "Dolby Noise Reduction" when it comes to video (Dolby produces no such technology; they meant Digital Noise Reduction). They also tested GPU video processing quality by outputting 1080i DVI into a Samsung LCD-TV with a native resolution of 1280x720, which means that their DVI results are confounded by the video processor in the Samsung TV. Worse, Principled Technologies occasionally came up with what can be called creative interpretation of how a few image quality tests are scored.

So for the second half of this article, I'm going to go over the HQV Benchmark DVD and explain how it should be used. Instead of just focusing on the hard details of the benchmark, I'll also be going over the basis for each test which will help you understand video processing technology and why the test is needed. When appropriate, I will point out where the NVIDIA commissioned document from Principled Technologies has erred.

Why do I understand the HQV Benchmark better than Principled Technologies? Well, for one, I've been hired by Silicon Optix to help with version 2.0 of their benchmark – so I know the strengths and weaknesses of the current version. I have also been providing feedback to the company based upon their prototypes, so I have a HQV video processor myself. With my digital photography and imaging background, this means that I know what technology Silicon Optix uses to achieve a perfect score as well as what 3rd-party software tools can achieve the same look through different algorithms. In contrast, Principled Technologies arbitrarily came up with their own criteria based upon "discussion amongst testers" rather than obtaining the readily-available Denon DVD-5910.

You can take my article with a grain of salt given my own involvement with Silicon Optix. That said, I will let the screenshots speak for themselves and you can run your own tests with the HQV Benchmark DVD. Likewise, you could say that FiringSquad is technically a competitor to Principle Technologies and that reflects an ulterior motive. That's not an ulterior motive: I'd just like to keep companies honest whenever I have time to write.

So let's dissect the HQV Benchmark DVD.


SIDEBAR: Windows XP Media Center Edition 2005 systems allow you to use two HDTV tuners to record two shows concurrently.

The HQV Benchmark DVD

Page:: ( 9 / 19 )

[image]

Color Bar / Vertical Detail

The first test on the DVD is the "Color Bar / Vertical Detail" test. It turns out that this does not actually evaluate detail. That test comes later. Instead this tests how well a video processor can distinguish between moving images and still images. Things in motion should be deinterlaced in a different manner from things that are not moving.

Recall that interlaced video consists of alternating odd and even lines. These fields are recorded 1/60th of a second apart. When things aren't moving, you can simply display both fields and everything will be OK. If you did this when objects are moving, you'd bring up the problem of feathering because half of the lines represent a snapshot of the object at a different point in time. Feathering is the artifact you say in the first quiz question. Surprisingly, figuring out whether things are in motion or not is very difficult. A 3-year-old can do it with ease, but coming up with software code to do it reliably is not.


A standardized still-image is used for the first test of the HQV Benchmark DVD. Since nothing is moving, the right way to deinterlace it is to show the full detail of the image, even with fine details that are only one pixel in height. If a video deinterlacer assumes that this picture is a moving picture, the processor will assume that what's recorded is a block that's flashing between white and black at a rate of 1/60 seconds. This produces flicker.

If the image is sharp and you can see both the black/ white and green/red stripes that's good. If there's any flicker at all, then the deinterlacer is doing a poor job of determining what is and is not in motion. Flicker is equivalent to ignoring half of the detail available in the picture. In a worst-case scenario, the boxes wouldn't flicker but would be solid in color. This would mean that the video processor is throwing away all of the even or all of the odd fields.

[image]

Counterpoint

Well, what happens if the director was recording flicker and was filming an object that alternated between white and black at 60 Hz? What would happen? It turns out that the data on the DVD would be identical. Playing devil's advocate, you could argue that HQV and everyone else with a sharp image are doing the wrong thing. Flicker is good. This is an ambiguous test in that both sides would have technically accurate statements. In practice, this is a bogus claim made only by companies that fail the test.

It's true that you cannot say 100% whether the DVD is supposed to be striped or if it's supposed to be a flashing box. On the other hand, every modern DVD has images with details that are only one pixel in height. The choice is between preserving detail in virtually all modern DVDs, or throwing away half of the resolution in virtually all modern DVDs on the off-chance that one day, some filmmaker somewhere will decide to record an object that flickers at 60 Hz.

Appropriate Scoring

This test evaluates whether a de-interlacer can properly determine whether things are in motion or things are static. This is important because video processors need to apply different algorithms depending on the source.

0 = The boxes are solid black, white, green, or red. The processor is throwing half of the resolution away.
5 = Any flicker is seen anywhere on the screen. The video processor is throwing half of the resolution away, but is alternating between throwing away the even and odd data.
10 = Stable, striped black/white and striped green/red rectangles are seen.

Jaggies Pattern 1 and 2 and Waving Flag

Page:: ( 10 / 19 )

The next three tests of the HQV Benchmark DVD evaluate how well the video processor does with moving objects. The first "jaggies pattern" is a simple spinning bar, the second "jaggies pattern" is a set of three bars with a waving motion (like waving your hand), and the third test is a real-world picture of the American flag in the wind. These all represent scenes that any respectable deinterlacer should recognize as being in motion.

If a deinterlacer passed the first test only because it always assumed that objects were static, these test patterns would reveal feathering artifacts and result in a failing score at any degree of rotation.

The only method for dealing with motion is to throw away some of the pixels that would cause feathering (at least when it comes to products costing less than $60,000). So, the difference between good and bad video processors is how selective it is at throwing away data. If you only throw away the pixels that would cause feathering, you maximize as much detail as possible. Remember, this is the "treatment" phase of the problem – it's only as good as the accuracy of detecting which pixels have the artifacts.

When you throw away data, you can replace it by averaging the pixels above and below the image. This loss of resolution causes jagged edges to form, most prominently on diagonal lines. Higher quality deinterlacers try to address this. This can be done with diagonal filtering or some sort of more intelligent reconstruction method that's more involved than just averaging the pixels above and below. Minimizing the amount jaggies gets increasingly difficult as the angle becomes more oblique. Scoring of the jaggies tests takes this into account.

[image]

1. Does the video processor only throw away data when it needs to?
   
2. Does the video processor incorporate advanced jaggy-removal technology?
   

Counterpoint

These tests are conceptually unambiguous in that there is no reasonable argument to suggest that these are flawed. Of note, the speed of the rotation and the amplitude of the waving in the HQV Benchmark DVD are designed to maximize the amount jaggies. This makes it easier to identify what video processors are doing. There are some manufacturers who fail this test, but have developed their own video clip that looks like the HQV Benchmark, but in fact has been carefully crafted so that it's easier to pass the test by reducing the speed of the bar or changing the size of the yellow zone. Beware of any similar looking tests that do not have the HQV logo.

Some readers may ask if technologies such as diagonal filtering may be effective for test patterns, but cause problems in real-world use. This is a great question, but fortunately the answer has already been determined. Faroudja pioneered diagonal filtering about a decade ago and has won an Emmy award for this achievement. In that decade, diagonal filtering has proven to be technology grounded in good science that's effective in real-world tests.

Appropriate Scoring

This test evaluates what a video processor does once it determines that things are in motion. Any time things get blurry, or you get jaggies, the video processor is throwing away data. Sharp lines mean that the video processor incorporates diagonal filtering (Faroudja, Gennum, HQV) or some sort of advanced jaggy-prevention technology (NVIDIA PureVideo). It's worth noting that although NVIDIA scores 3 points in this test, they fail at the very last frame of the yellow zone. An optimist would say that they're only one frame away from getting 5 points.

Jaggies 1:
0 – Logo is jaggy or jaggies are only seen when the bar is <45 degrees.
3 – The logo is sharp, and jaggies are only seen when the bar <20 degrees
5 – The logo is sharp and jaggies are only seen when the bar is < 10 degrees

Jaggies 2:
0 – Jaggies are present in all three bars
1 – Jaggies are present in the bottom two bars
3 – Jaggies are present only in the bottom-most bar
5 – All three bars have smooth edges at all times

[image]

Picture Detail

Page:: ( 11 / 19 )

In this test, a video clip of a cityscape is used to evaluate for picture quality. The video provides several areas to look for detail such as in stairs, a grass field, or a bridge.

Basic video processors do not incorporate any sharpening tools at all. These deserve a score of zero. This is what is meant by a "soft image." Most video processors including those from NVIDIA and ATI fall into this category. In this regard, NVIDIA and ATI should only receive 0 points on this test rather than the 5 that were awarded by Principled Technologies. Importantly, a score of zero does not mean that the product is blurring the image. It means that the playback device is outputting exactly what is on the DVD.

[image]

Counterpoint

HQV has decided that bad sharpening (5 points) is better than no sharpening (0 points). There are arguments for and against this distinction. I don't agree with this entirely. Sharpening can be performed at any stage of the video processing chain. When testing a "system" with HQV (i.e. monitor + DVD player), this scoring system makes a lot of sense. You need some sharpening. That said, sharpening on the TV is usually better than sharpening on the DVD player because it can be optimized for the size of the screen (Note that I'm talking about $200 DVD players – once you get to the DVD players that are $1k or more, things change).

If I was buying a DVD player, I would rather buy one with no sharpening than buy one that forced sharpening to be on and produced artifacts. This is because I would later add sharpening through the TV or monitor. A better way to approach sharpening would be a system of 0 if there's sharpening artifacts even at the lowest user setting, 3 points is there's no sharpening, 5 points if there's user-adjustable sharpening that has artifacts only when turned too high, and 10 points for sharpening that has no artifacts that can also be user adjusted. This is a case where the single numerical score does not tell you as much as a picture. That scoring system is only provided so that you can get a sense of what's important and not when it comes to video.

Appropriate Scoring

For the purposes of consistency across all reviewers, the appropriate way to score is
10 = Image is equivalent to the reference HQV image or better.
5 = Image has sharpening artifacts
0 = Image has no sharpening

Reviewers using the HQV DVD Benchmark should take into account the source / display interaction and describe whether the DVD player has a) adjustable sharpness including the ability to turn it off and b) whether the sharpening algorithm causes artifacts.

Noise Reduction and Motion Adaptive Noise Reduction

Page:: ( 12 / 19 )

To understand the basis of these tests, you need to understand how noise reduction works. You can either filter out noise spatially (i.e. blur the image slightly) or filter it out temporally by recognizing the fact that noise is random, and removing noise by averaging out the pixels that have changed from frame to frame. Temporal filters are better because they theoretically preserve more detail than simple spatial filters.

The first "Noise Reduction" test evaluates noise by using video containing slow camera pans. These images are recorded using analog broadcast tape (i.e. what TV studios may have used previously) which provide a large amount of baseline noise.

[image]

Motion Adaptive Noise Reduction

Page:: ( 13 / 19 )

In this test, you have video of a roller coaster. A temporal filter that is blindly applied will not be able to distinguish the movement of the roller coaster from random noise – it will see both as "changing pixels." This results in a ghosting or "echo" of the moving roller coaster because it's trying to blend out some of those pixels.

A zero score is given when noise is present (a product which has no filtering) or when temporal filtering is blindly applied resulting in motion trails behind the rollercoaster. None of the video cards tested by Principled Technologies provide any noise reduction and they should all have received zero points in this category. It is unclear what Principled Technologies thought they were seeing when comparing the Intel 945G against NVIDIA products.

A score of 5 is given to video processors that reduce noise at the expense of significant detail. This includes traditional spatial noise filters. In the case of moving objects, a dumb spatial noise filter does a better job than a dumb temporal noise filter. This scoring system allows a video processor that lacks the horsepower for motion-adaptive temporal noise reduction to still score well on the HQV Benchmark DVD as long as it switches between spatial and non-adaptive temporal filters based upon the content.

A score of 10 is given to noise filters which are able preserve detail even when the scene contains moving objects. This includes motion-adaptive temporal filters. Motion-adaptive temporal noise filters distinguish between random noise and moving pixels. When this roller coaster clip is played back through an HQV processor, you get the benefits of temporal filtering (minimum noise and maximum detail in the sky and track) without ghosting or motion trails behind the roller coaster. Again, some sort of advanced wavelet-based spatial filter would also be able to earn a score of 10.

Counterpoint

Principled Technologies claims that these noise reduction tests "have little bearing on evaluating the abilities of a graphics adapter or DVD player" and that noise only reflects copies of video tapes or analog broadcast interference. They are wrong. Many remastered DVDs incorporate some sort of noise reduction but they still have noise that is introduced in the MPEG-2 compression stage. I think noise reduction is one of the most critical elements of a high-end video processor. It makes a substantial difference even with 720p HD broadcasts. Noise reduction is particularly important for big screen TVs as well.

I’ll let you decide for yourself with several screenshots from modern DVDs and HDTV clips. In each case I have taken the unprocessed images and used advanced noise reduction software (Noise Ninja, Neat Image) to show what removal of noise can do for image quality. I've tried to come up with an eclectic mix of content.

Noise reduction images

Page:: ( 14 / 19 )

[image]

Noise reduction images (cont’d)

Page:: ( 15 / 19 )

[image]

Appropriate Scoring

Noise Reduction
0 – No noise reduction
5 – Noise reduction is present, but causes blurring of detail
10 – Noise reduction is present and does not cause loss of detail

Motion-adaptive noise reduction

0 – No noise reduction is applied, or the noise reduction produces artifacts
5 – Noise reduction is present, but there is loss of detail
10 – Noise reduction is present and there is no loss of detail.

3:2 detection and Cadence Tests

Page:: ( 16 / 19 )

Remember when I talked about DVDs and 1080i storing interlaced video and how each field was recorded 1/60 seconds apart? There's are exceptions to that, big ones at that. For Hollywood movies recorded at 24 fps on silver halide film, or 30 fps video recorded on high-end progressive scan HD-DVCAM camcorders, or animation, the full detail can be recorded. The difference is that these sources are recorded as progressive images and then interlaced for distribution on DVD or 1080i HDTV rather than being interlaced recordings. You should always be able to encode any source into 60 fields per second while preserving all of the detail as long as the source was a) recorded as progressive and b) had a frame rate below 30, right?


Yes. That's exactly right.

Let's look at 24 frames per second film. If you divided them up, you could create 48 fields in a second where each pair of fields is recorded at the exact same time! Now, if you needed to take those 48 fields and spread them across 60 fields, you could do it with a 2:3 ratio. If your frames were A,B,C,D then you could both halves of A and both halves of B in the first 2 frames (4 fields). In your third pair of fields, you could have half of B, and half of C, and in the fourth pair of fields, the remaining half of C, and first half of D. In the fourth pair of fields, you could repeat both halves of D. This represents a 2:3 cadence because 2 fields of one frame are shown and then 3 fields of the next.


The ability to reconstruct images from video encoded in this format is called 3:2 pulldown. Believe it or not, the ability to detect this cadence is not trivial especially if a scene starts somewhere in the middle instead of the "A field." The ability to deal with tougher sequences is sometimes called bad-edit detection.

Any time you fail to recognize the cadence, you end up throwing away up to half of the resolution. Suffice it to say, cadence detection is important for maximum detail.

[image]

Cadence Tests

Page:: ( 17 / 19 )

While 3:2 is the most popular cadence used, it's not the only cadence. Documentaries shot on high-speed film run at 30 fps resulting in 2:2. DVCAM or HDV camcorders may use 2:2:2:4 or 2:3:3:2. Movies broadcast on TV may drop every 13th field in order to leave more time for commercials resulting in 3:2:3:2:2. Animation ranges anywhere from 8 to 12 fps.

The next sequence from HQV tests all of the 8 major cadence sequences, awarding 5 points for correctly passing the test. Any time you fail to detect the appropriate cadence, you end up losing half of the resolution so scoring is based upon whether or not the video processor throws away of the resolution or whether the full image is displayed.

[image]

Counterpoint

HQV was engineered as a no-compromise video processing solution, and one element is their cadence detection algorithm which detects "anything and everything." It's no surprise that they would assign 30% of the total score to this feature.

If I was spending $3000 on a DVD player or a dedicated video processor, I really would want a product that would work with anything I threw at it. In that respect, it's a fair test for comparing different $3000 DVD players or video processors. Since DVD discs are not always encoded with the correct "flag" which identifies the 3:2 cadence, this test challenges video processors to actually interpret the video rather than rely on data that is unreliable.


For the rest of us who don't have that kind of budget, it's reasonable to give up some of the less important cadences. The question of course is which ones you can give up.

Of all the cadences, 3:2 is the most critical. That's your Hollywood blockbuster film. That's the bulk of what you'll end up watching. If your system doesn’t reliably detect the 3:2 cadence, your video processor may be discarding half of the resolution available in Hollywood movies.

2:2 cadence can refer to putting 24fps film onto 50 fps PAL. However, it also is important for 30 frames/sec progressive content on DVD or 1080i. This is the second most important cadence for most of you because you'll find it being used in the documentaries and many TV shows that are put on DVD. If you watch TV shows on DVD or have a Media Center PC, you'll want good 2:2 detection.

I think most of us have at least some animation in our collection and die-hard anime fans who watch more anime than Hollywood films might even place more value on the 5:5, 6:4, and 8:7 cadence patterns than the others. 8 fps reflects old school anime, 12 fps is found in higher quality anime and most US animation.

Vari-Speed Broadcast (3:2:3:2:2) is important if you enjoy watching movies on broadcast TV.

Lastly, the 2:2:2:4 and 2:3:3:2 cadences are the least popular formats used. If you're involved with indie-films, make your own videos with software like Final Cut Pro, or need to review digital dailies, it’s important. Otherwise, it's a nice to have on a checklist but is the least essential for typical viewers.

Appropriate Scoring

5 – Flickering, jaggies, or loss of resolution absent
0 – Flickering, jaggies, or loss of resolution present

Reviewers should recognize the varying importance of cadences and the type of content people view.

Mixed 3:2 film with Added Video Titles

Page:: ( 18 / 19 )

In order to pass this these successfully, a video processor must selectively apply different deinterlacing algorithms depending on the region of interest. This tests the "adaptive" nature of the de-interlacer.

10 points is given if the text is sharp and the background is clean.

5 points is given if the video deinterlacer defaults to "video" meaning that text is clear but the rest of the screen loses detail

0 points is given if the video deinterlacer remains in "film" meaning that the movie is clear, but the scrolling text produces feathering or combing artifacts.

[image]

Counterpoint

Again this test was engineered to showcase the "we can handle anything" motto of HQV products. In practice, you'll never run into this problem if you're watching DVD movies, and may only run into this problem if you watch movies on TV on selected channels that have 24/7 news tickers, or during some sort of major news event. It's nice to have and it's great for showing off, but ultimately much less important that all of the other tests.

Conclusion

Page:: ( 19 / 19 )

In that regard, it's somewhat ironic to read through Principled Technologies' commissioned study. They've made several errors in evaluating the video playback performance of PC GPUs resulting in 10 points of artificial inflated scores for NVIDIA and an artificial 5 point penalty for the i945G. In my testing, NVIDIA PureVideo solutions playing a DVD via a native 480p window using a 1280x1024 DVI connection resulted in only 51 out of 130 points. This is identical to the results obtained at Anandtech.com but clearly different from the 86 points that Hot Hardware gave or the 68 point that NVIDIA’s commissioned article gave.

The irony is that even though NVIDIA's scores were inflated by Principled Technologies, their "key findings" are ones I still agree with completely. Namely, that NVIDIA PureVideo is superior to ATI's Catalyst 5.8, that PureVideo is superior to basic DVD players, and that PureVideo minimizes some of the common artifacts seen with de-interlacing.

In comparison to higher-end DVD players such as the Denon DVD-2910 ($700 retail), NVIDIA still has a ways to go. Likewise, PC manufacturers in general are behind the consumer electronic manufacturers when it comes to capabilities such noise reduction or detail enhancement. I hope that companies like ATI and NVIDIA who are serious about image quality will consider adding high quality noise reduction and detail enhancement technology in their next generation products.

The state of PC video quality is also going to change. ATI's Catalyst 5.9 drivers feature a brand-new de-interlacing technology which may address some of the limitations of older ATI products. ATI's AVIVO supports noise reduction with 3rd party TV tuners but it's not yet clear if ATI will be able to implement noise reduction for DVD sources, or better yet HDTV sources. I have not yet had the opportunity to test Catalyst 5.9 myself, so I can't make any comment on how it compares to PureVideo but it warrants a closer look. We at FiringSquad hope to take on that challenge. As you can see, this is just part 1 of our series. Our next step will not only be to test the best from ATI and NVIDIA against the best standalone consumer DVD players when it comes to DVD content, but also to challenge the GeForce 7800GT and Radeon X1800 in tests of high-definition deinterlacing, where the processing demands can be as much as 9 times higher.

A Special Treat

The HQV Benchmark DVD has been distributed to various journalists, and consumers can buy the disc for about $30 ($20 with a coupon). While it'd be better if the benchmark were free or advertiser-supported, someone has to pay for the licensing costs associated with the video clips in the Benchmark DVD.

I'm going to give 15 of these DVDs to give away. To enter, all I ask is that you share this article with someone else (who would care) on a message board other than the FiringSquad's. Email alan@firingsquad.com with the subject "HQV Raffle" and include a link to your post before October 31. We want to make this easy, so 12 winners will be selected randomly. 3 winners will be selected based upon the popularity of their thread. So if you're known as the DVD or HDTV guru in your circle, this is an easy way to win. Anyone who SPAMs a messageboard (as determined by me) will have their entry disqualified. International winners will need to pay for shipping.