Summary: ATI's Director of Marketing for Multimedia Products, Godfrey Cheng, offers to do an interview with FiringSquad. We sent in Alan. If only ATI knew what they were getting into...
FiringSquad: Why don’t you tell our readers about yourself? i.e. What do you do for ATI, how long have you been working there, what was the greatest achievement you’ve seen at ATI, and the biggest mistake at ATI you’ve seen.
Godfrey Cheng: My name is Godfrey Cheng and I am a Libra. I would consider playing poker professionally except that working at ATI rocks. I am the Director of Marketing for Multimedia products (anything with a Wonder suffix) and I am also responsible for driving all Media & Video initiatives. I have been with ATI since 1998. The greatest achievement that I have seen at ATI was when we delivered the Radeon 9700 Pro to the market and put nVidia into a state of shock & awe. The biggest mistake that I have seen at ATI was the Rage Fury Maxx…..not because of the product, only because everybody in Asia pronounced it “Rage Furry Maxx”. We are heading for good times again.. watch…
FiringSquad: It seems like AVIVO can be summarized into these categories:
1. Capture of the source (DTV products, Theater 550 products)
2. Video processing (which consists of compute offload (MPEG-2, H.264) and quality enhancements (deinterlacing, noise reduction))
3. Display (10-bit output, HDMI, HDCP)
4. Software (Catalyst Control Center)
Are we missing anything?
Godfrey Cheng: This is accurate….. All these steps above, represent roughly what we call the Video Pipeline. Before any video is actually seen by a customer, there are a finite series of steps that must occur to process and display the video. Any errors or weak links in any one of these steps results in these errors being propagated further down the chain resulting in poorer video quality. Avivo is a platform of technologies designed to systematically improve each and every step listed above. While we are focusing on primarily HD H.264 decoding, De-interlacing & scaling, complete 10 display processing and 2 Dual Links this time…..we may talk about new TV products or mobile technologies the next time we talk about Avivo technologies. The one constant is that we are going to systematically invest in each and every step of the video pipeline to deliver the best video & display quality on the PC.
FiringSquad: When might we see All-in-Wonder boards in the US with a ATSC or QAM tuner?
Godfrey Cheng: As you know we don’t discussed unannounced products. But dang, such a product would be really cool huh? Specially in a world without Broadcast Flag for ATSC.
[Alan's comment: The Broadcast Flag is a proposed "feature" of HDTV in which content providers can set restrictions on how the content can be used. These include an inability to save a program to the HDD or permanent storage media, inability to make copies-of-copies, or forced reduction of video quality, or an inability to skip over commercials. Originally this law was intended to go into effect on July 1, 2005 but a federal court ruled that the FCC did not have the authority to make this law. Thus, the broadcast flag is NOT in practice at this time. However, the federal court made no ruling on whether the concept of a Broadcast Flag itself is legal. The "broadcast flag" could be made into law if it were introduced in Congress. If this were to happen, a federal court would likely have to decide whether or not the Broadcast Flag that prevented copying fell within accepted practices such as DVD or SCMS copy protection or if it was in violation of the rulings that let us have VCRs.
ATI cannot discuss unannounced products, but I hope they see value in an AIW product with HD tuners. In many smaller HTPC builds, there is no room for a dedicated PCI HDTV Wonder; an AIW would be very useful in this regard.]
FiringSquad: What does ISF certification actually entail in terms of technical measurements? That is, while the ISF is a highly respected organization, it would be helpful to know what they consider to be their minimum standards. That is, even for something as exclusive as THX Ultra2 certification, you have a wide range of quality within the listo f certified products. (Likewise, seeing THX Ultra2 certified receiver tells me that it’s 3.2 ohm stable, and can play at reference levels in a 3000 cubic ft room when paired with THX certified speakers)
Godfrey Cheng: ISF, like THX, is an independent company’s certification of the display quality of a vendor’s hardware. ATI worked closely with Microsoft to bring ISF certification to the PC World earlier this year. What you will find is that the quality that is necessary to achieve ISF certification will change, the bar will get higher with every major product release. As the leader of video & display quality in the PC World, we are confident that our hardware and software will be setting the high bar in the ISF Labs. An overview of what ISF tests for can be found at http://www.imagingscience.com/rl/ISFRL_Certification_Overview.pdf.
[Alan's comment: Unfortunately, without specifics, the value of the Imaging Science Foundation certification is limited. There are several problems which prevent ISF certification from being a truly useful measure.
One problem is that ISF certification changes each year. This means that last year's ISF certified product may not be certifiable the following year. This can only add to the confusion. The more important problem is that the criteria required for certification is only provided under NDA. For example, looking at the above PDF we can see that the ISF runs a extensive set of tests ranging from the Y/C delay of the analog inputs to the quality of the outputs, to something as simple (but important) as having an infrared remote work at a 20 foot distance (a requirement of a ISF certified Windows MCE2005 system). However, we are never told how much of a Y/C delay is acceptable. Likewise, what happens if a product A has superior playback DVD quality than product B, but fails at the ability to reproduce the PLUGE or "blacker-than-black" criteria? (ISF calls this foot room preservation). The whole point is that blacker than black is not intended to be seen. So a hardware calibrated system that did not pass the blacker-than-black signal would fail ISF even if it had a perfect reproduction of the intended source.
In a way, we the consumer, are asked to take the ISF's quality measurement with good faith. This can be tough. ISF is in the business of making money by charging companies for testing. Microsoft is in the business of getting Media Center PCs to be accepted by the general public. ATI is in the business of convincing consumers to buy newer hardware. No one in this process is looking out for the consumer without having at least some conflict of interest. While it's true that the ISF has a reputation to uphold, it's hard to know how lenient or stringent the ISF is. Consider the fact that ISF certifies the deinterlacing in Windows Media Center PC's, however these deinterlacers score very poorly on the HQV Benchmark DVD, a test where the criteria and scoring are very well defined.
THX certification is somewhat similar to ISF certification. Most of the criteria required for certification is unknown. That said, there are certain details that are disclosed (the 3.2 ohm stable amps, etc.). Perhaps most importantly is that THX Ultra2 processors have valuable and tangible features such as de-correlation, Cinema Re-equalization, and boundary gain compensation.
When you pay more for THX Ultra2 certification in a home theater receiver, you pay for some unique processing algorithms. When you pay more for ISF certification, you're asked to have faith that someone at ISF is looking out for the consumer.
We would rather that THX and ISF make their criteria public knowledge. Consider the NHTSA car crash tests. It's no secret how those tests are done. The result has been car manufacturers working harder to improve car safety. Imagine if it was "big oil" who worked with an "independent third party" company to test car safety...]
FiringSquad: What does ATI consider to be the limiting factor for dual-HD-tuner products? Not enough consumer interest? Design issues with getting it on a single board?
Godfrey Cheng: While dual TV Tuners are really cool, what we are finding is that OEMs prefer the flexibility of single tuners so that they can mix and match. With MCE 2005 (roll up 2), you can have any combination of 2 analog and 2 digital tuners….. Also, the vast majority of TV enabled PCs only ship with a single tuner today due to cost. Should the market want dual tuners (dual analog, dual digital, 1 analog + 1 digital)
FiringSquad: FM tuning has been a part of a few AIW products. ATI has historically been first-to-market with innovative multimedia functions (ATI-TV back in the Mach64 era, ATI TVO TV out in the 3D Xpression era). Has ATI looked at HD-Radio PC products for the US market? (Is radio even a target market for ATI?)
Godfrey Cheng: Digital Radio is not something that we are considering….. ATI is a graphics company :)
FiringSquad: Will component HD or HDMI capture solutions ever be possible in a reasonably priced AIW-type product? (For hooking a console up to a PC)
Godfrey Cheng: This is something we are looking at but we have no plans either way.
[Alan's comment: Hopefully the PS3 and Xbox 360 will drive this market. The size of televisions has changed substantially and the shift to all-digital broadcast will happen in 2006 if we're lucky, and 2009 if we're not – both well within the product lifespan of the PS3 and Xbox360.]
FiringSquad: What about AIW with digital cable and satellite support?
Godfrey Cheng: This is something ATI is definitely interested in. We are working with Microsoft on some products that we cannot discuss yet. Stay tuned. :)
[Alan's comment: Windows MCE2005 has some preliminary support, but requires the use of an external box.]
FiringSquad: How computationally intensive is H.264 decoding, and how difficult of a challenge was it to integrate that decoding ability into the X1800?
Godfrey Cheng: At a given resolution, H.264 is about 3-4 times more computationally intensive than MPEG2 to decode. Just some numbers…… A 1080i MPEG2 HD stream is about 19.2 Mbps. For H.264, this 1080i stream would be about 8-9 Mbps but it would take 3-4 times the effort to decode the H.264 stream.
FiringSquad: What about the argument that by the time H.264 is popular, most CPUs will be fast enough to do software decoding? After all, HD content isn’t intended for 20” SDTVs – the early adopters of HD content are also people with killer PCs.
Godfrey Cheng: Your argument, unfortunately, is a common misperception. While CPUs do many things well, there are just some things that the GPU does better. You do not want a CPU running at full power all the time….remember Notebooks are a big portion of the market so fan noise and power consumption are as important as the decode performance itself! The early adopters for H.264 are the HTPC guys…no one there wants a CPU with a helicopter rotor cooling it. We know in many instances that even the fastest CPUs today cannot play back . Our estimates show that this will remain the case for a considerable amount of time (the same thing happened when MPEG2 first arrived on the scene).
[Alan's comment: Godfrey brings up a very good point about notebooks. Even if you have a fast enough CPU, offloading the video processing to the GPU almost always results in significant power savings. I agree 100%.
On the HTPC front though, I'm less inclined to agree with ATI. At Computex, ATI showed a Pentium 4 3.6GHz with Hyperthreading running 30% CPU usage on 25MBps HD H.264. The P4 560 is roughly on par with an Athlon 64 running 2.4GHz. Thus, AMD's fastest dual-core CPU from 5 months ago should easily be able to do it at ~45% CPU utilization. With a nice Zalman based cooler, the CPU fan noise is going to be less of a bother than the noise from the GPU.
Of course, that's just my prediction. I predict that CPUs are going to be fast enough by the time H.264 is important. For ATI to be right when it comes to desktops, H.264 adoption has to occur so rapidly that people with slower CPUs will want H.264 decoding on their GPU, but be happy with their CPU.
As a point of reference, when MPEG-2 first arrived, ATI and S3 were among the first companies to provide a high quality solution. ATi was first to market with hardware acceleration and S3 was first to market with alpha-blended subpictures. Looking back, we were looking at Pentium II 233 and 266's and the ATI 3D Rage Pro for DVD playback on the PC to become a selling point (ATI had motion compensation an earlier 3D Rage II). By the time Matrox launched its G400 in early 1999 (about 1.5-2 years after the 3D Rage Pro), hardware acceleration for DVD playback was not that important.]
Godfrey Cheng: The Radeon X1800 can easily decode a single HD MPEG-2 – we have not really mapped out beyond a single stream because there is no application in the PC space. The Xilleons are the monsters of the CE world…..last time I checked they can decode 104 HD streams and 877 SD streams simultaneously….. (just kidding but it can do multiple 2 HD streams and our latest Xilleon is the 240).
[Alan's comment: The PC and CE world are on a collision course. Eventually it'll matter. NVIDIA doesn't do multiple video streams either. It'd be nice to have in the short-term for media and video editing, but once dual HD tuners becomes more prominent, picture-in-picture HD is going to be a nce feature.
For reference, the Xilleon 240 was recently announced in August. The Xilleon 220H is the previous-gen product from ATI found in many shipping products. It also does multiple HD streams.]
FiringSquad: How much of the Radeon X1800’s compute abilities are due to dedicated blocks on the GPU, and how much is achieved through the 3D pipeline allowing future upgrades?
Godfrey Cheng: There are parts of video processing are better done with dedicated blocks of HW and other things are better done with our programmable shaders - we have both. Ultimately this doesn’t matter since the output is the only thing that is important but I will comment that it is critical to consider power consumption & fan noise when we make these architecture decisions.
[Alan comment: This also has implications on the level of trickle-down. Dedicated blocks allow for "dedicated HTPC" chips (imagine an ATI motherboard with high-end integrated video) while processing through the 3D pipeline means that next year's mid-end gaming GPUs can benefit also. NVIDIA also does a combination of dedicated blocks and 3D GPU]
FiringSquad: What additional detail can you provide about the new vector based deinterlacing? It *detects* edges in a few different directions, but what does it actually do differently from older algorithms once it detects the edges?
Godfrey Cheng: I cannot add anything more than what is in our whitepapers. You do realize that when we discuss algorithms, our legal department calls me up for a “meeting” as we are really talking about ATI’s intellectual property.
[Alan's comments: Hopefully, I'll be able to get a better sense of what they're doing once I spend some more time. NVIDIA is just as ambiguous when it comes to their "modified bob" algorithms. I've got a pretty good idea of what NVIDIA is doing based upon some custom test patterns of mine, so it'll be interesting to see if ATI is doing it the same way. Both companies have confirmed that it's not "diagonal filtering."]
FiringSquad: We’re seeing things like dedicated PPUs. Do you envision a day where there’s a dedicated VPU for the PC?
Godfrey Cheng: The GPU is the last step before a customer sees pixels…… I envision that we will add more video features in the GPU over time to always improve video quality so I do not see the need for a dedicated VPU.
FiringSquad: Does ATI implement different algorithms for deinterlacing HD MPEG-2 versus DVD resolution MPEG-2? For x800? For the x1800? For the Xilleon 220H?
Godfrey Cheng: The big benefit that ATI has over our competitors is that we have a consumer group that has a ton of knowledge and know-how…..the Xilleons are powering most of the HDTVs sold in North America….so we un-bashfully steal their algorithms and put them in the Radeons for the PC market
FiringSquad: Do you have any comments on the HQV Benchmark DVD?
Godfrey Cheng: It is a good benchmark for Standard Definition DVD Quality. It is one of several benchmarks that we are looking at. We are also looking at HD benchmarks. We will announce our video benchmarking direction in the near future.
[Alan's comments: I'm working on some HD benchmarks too]
FiringSquad: ATI has noise reduction for analog TV, but what about DVD/HDTV/H.264
Godfrey Cheng: Noise reduction is not necessary for digital sources, as any noise removal would be removed at the authoring or encoding level…..but we do have a whole host of video processing tasks that we do for digital streams to make them look better like color correction and de-interlacing. In the case of encoded analog streams, it is best to remove the noise during the encode process using parts like the Theater 550 Pro. As an example, the HQV benchmark, tests noise reduction by playing back an MPEG2 video that has noise in it. The noise in this video was generated during the analog encode process. As such, if the noise was removed during the analog TV encode there is less reason to de-noise on the playback.
[Alan's comments: " Noise reduction is not necessary for digital sources" – I think you meant "Noise reduction should not be necessary for digital sources." I've provided some good examples in our HQV benchmark analysis and you can ask anyone at AVS Forum or Widescreen review about the value of product like the Algolith Mosquito at removing noise. Oh well, those who do not have noise reduction downplay its significance…]
FiringSquad: How does the X1800’s deinterlacing and scaling compare to the Xilleon 220H?
Godfrey Cheng: The Radeon X1K family is, by far, the best video processor for the PC Market. The Xilleon is the best CE TV processor in the market. Spiderman never fought against Batman…….Radeons & Xilleons are both rock stars.
[Alan's comments: Well, that's because Spiderman belonged to Marvel and Batman belonged to DC. We do have Alien vs. Predator and Marvel vs Capcom though. Was hoping for something more than a marketing answer, but we'll leave that for another day :)]
FiringSquad: Do you have any opinion on Philips Trimension? Good? Bad?
Godfrey Cheng: I have no opinion on this.
[Alan's comments: Trimension is Philip's custom technology which "adds" extra frames. It's like those games engines which interpolate the animation for smoother motion. Intervideo WinDVD has this in software. Philip's TVs with Pixel Plus 2 technology have the same feature. Bang and Olufsen uses it in their Plasma panels as well. It's interesting because it's a completely artificial look but often works very well.]
FiringSquad: Other than the Apple Cinema 30”, are there other dual-link DVI monitors currently available at retail?
Godfrey Cheng: There a few displays that have dual-link DVI input on the market today and many more to come. Along with Apple, IBM, Viewsonic, EIZO Nanao, and Iiyama, just to name a few, all make such displays. Here are some examples: Iiyama AQ5311DTBK (http://www.iiyama.us/default.asp?SID=&NAV=236&PROD=2713), National Display Nova (http://www.nationaldisplay.com/buzz2/nova.htm)
[Alan comment's: The iiyama 3MP panel (and Viewsonic's and IBM's) all use an Lenovo/IBM panel. IBM had developed *and* shipped a 9 megapixel monitor, but Lenovo has decided to drop the 9MP model.
In retrospect, my question wasn't precise enough. Radiologists certainly use high-resolution displays (EIZO, Planar, National Display, etc.) that require dual link DVI. Virtually all medical images are captured digitally now – even X-Rays. ATI actually has a potential market here. The high-res PACS (picture archive and communication systems) from Agfa, G.E. Centricity, and Philips/Stentor are all pretty laggy. A fast video card would be beneficial, although in these types of applications, you can't simply swap out video cards. We talked about ISF certification early, but what you'll need is DICOM (Digital Imaging and Communications in Medicine) certification for medical devices. Likewise, most of these PACS systems are leased.
Of course, none of these are actually available in a retail store where you can see it live. Now do you see why media and video guys get so excited about Apple's?]
FiringSquad: Are there any currently available products with 10-bit HDMI inputs?
Godfrey Cheng: HDMI does not support 10-bit; however DVI does over dual-link DVI, and that’s the other advantage of having dual-link DVI. One can use it to drive higher resolutions, such as Apple Cinema’s 2560x1600, or to drive higher color depth. We can do both with our Avivo display engine. I can read your goat like mind: Are there any displays doing this today? The answer is yes; NEC and EIZO Nanao have 10-bit displays today, and we think there will be more vendors offering 10-bit in the future.
FiringSquad: If I am running a multi-monitor setup in Windows XP, will ATI AVIVO products give me a mechanism for using a different color LUT for each monitor for color correction?
Godfrey Cheng: What is YES Alex?
[Alan's comments: I'm going to test this…:)]
Godfrey Cheng: Legal alert…..no comment…..nVidia can do their own research
[Alan's comments: For historical purposes, S3 disclosed that they use Ulichney Void and Cluster algorithm in their 16-bit 3D dithering approach (NVIDIA and ATI were using ordered dithering at the time, with the Rage128 apparently having a lower-order filter). Meh, all these secrecy is no fun. S3 essentially went under (Via/S3 isn't quite the same thing), so maybe giving out the technical details is a bad thing.]
FiringSquad: Since 6-bit LCD panels already do their own temporal dithering is it possible to configure the X1800 not to do a temporal dither?
Godfrey Cheng: You can configure X1000 series to do either temporal or spatial dithering, or no dithering for that matter; the hardware is flexible enough to allow you to pick and choose; that’s one of the beauties of Avivo! You can dither yourself or your friends!
[Alan's comments: That's awesome! I love it when companies give users fine control of their hardware.]
FiringSquad: Multimedia Center is a set of distinct applications bundled together. Does ATI have plans to clean up the suite to remove some of the bloat and re-write it into something leaner and more integrated like what is offered by MCE2005, SageTV, etc.? Have you ever considered a Linux version of Multimedia center (where a dedicated HTPC can be built around EazyLook)
FiringSquad: The HDTV Wonder captures in the native ATI-VCR format which then requires a lengthy process to transcode into MPEG-2. Will future versions of Multimedia Center offer an ability to directly record the MPEG-2 transport stream?
Godfrey Cheng: Our future software strategy will encompass recording the MPEG2 stream from digital TV directly.
[Alan's comments: That's great!]
FiringSquad: ATI offers MPEG-4 encoding of analog TV. Will future versions of Media Center allow someone to take MPEG-2 HDTV recording and convert it into an AVI with an MPEG-4 video compression and AC-3 audio (i.e. decoded by ffdshow).
Godfrey Cheng: Transcoding? Isn’t that part of Avivo? We obviously see value in converting one format to another but we are not ready to discuss our product plans yet.
[Alan's comments: All our questions today are company secrets…]
FiringSquad: Some HTPC owners prefer to use external video processors. Is it possible for future versions of Multimedia Center and AVIVO to provide automatic switching of native-rate outputs (480p in -> 480p out, 1080i in -> 1080i out with no user intervention)?
Godfrey Cheng: As per your question above, we are looking at component, DVI, HDMI capture but we do not have any plans to bring any of these products to market. We have to see the market and the need. Additionally, the copy protection schemes involved with both these standards limit our ability to input and record this video.
FiringSquad: Since you’ve had to deal with some tough questions, here are a few easier ones. In the Hexus review (http://www.hexus.net/content/item.php?item=1033&page=2), you mention your 1989 944 Porsche Turbo S. Wasn’t the “Turbo S” versus “Turbo” distinction only for 1988 models? In 1989, didn’t all 944 Turbos have the features of the previous year’s Turbo S, but lacked the “S” moniker?
|© Copyright 2003 FS Media, Inc.|