Okay guys. I apologize in advance because I'm going to be ranting about this one. You should know my usual disclaimer that I used to work for Amptron, which is a competitor of Viewsonic (both sell LCD monitors in the United States). I'm going to be bashing on Viewsonic a lot here. Put on your fire gear if you haven't already. It's probably a better idea for me to "sleep on it" and decide whether or not I should post this later, but I ain't that good with ideas.
***** Preamble (skip this part if you just want to read about why I think Viewsonic's numbers are false)
I graduated with a degree in Mechanical Engineering. As an engineer, the general philosophy is to approach things carefully -- always err on the side of caution, always check your work, don't try to overstate your results, but always hedge your work and state its limitations because you can get people killed otherwise. We don't want no bridge collapsing because some engineer said it can support 100 tons but it really only supports 50 tons. We don't even want an engineer saying it can support 100 tons if it really can support 100 tons, because we're taught to put safety factors in our work -- so a bridge that is designed to support 100 tons is only going to be rated for 70 tons or so (safety factor is usually about 1.5 or so).
This is the opposite of a marketer. You'll notice that I often speak out against marketers when I post. Why? Because it seems like a marketer's maxim is that "anything that can't be proven false is true" in terms of what they say. A corollary is that they can define things however they want. This is why I say that a company's marketing team says whatever they think the legal team can defend in court based on the engineering team's results. If they can back it up in court (and of course, by using a lot of "when we said this, what we really meant was...") then they'll say it to make money. The unfortunate thing is that it works -- it does bring in more money than otherwise, simply because most consumers don't have enough technical knowledge to really understand the terms, and thus tend to gravitate toward cool-sounding words ("Amplified Impulse") or easy-to-grasp numbers ("4 ms").
My interest in LCDs was sparked when I started working for Amptron. It's a (relatively) tiny company located in Southern California, and I do sort of feel sorry for the people there, because it's a small company because it doesn't have the budget to become a bigger company because it doesn't have a marketing department to fool everyone into buying its products over that of other companies and thus get more money rolling in. It's had a zero (whole) dead pixel policy since 2000, when it first started selling LCD monitors, because its dead pixel policy is on a subpixel basis, and any three subpixels within 1 cm of each other gets replaced due to its adjacency criteria. A whole dead pixel, of course, is three subpixels within 1 cm of each other. By comparison, HP came out with a similar thing in 2003, and Viewsonic announced that they would be offering a zero whole dead pixel policy as well in 2004. Formac offers a "zero dead pixel warranty" for $99 which if you read the fine print, defines a pixel as an RGB element (all three have to be dead for it to count as a dead pixel), defines "dead" as dark-only (bright pixels don't count), and due to "rough shipping conditions" they won't replace a monitor unless it has three dead pixels or more -- this for an extra $99 "zero dead pixel warranty". Oh, and theirs expires after two weeks. With other companies doing that, you would think that a company offering such a policy since 2000 would be making waves. Nope. No marketing department to get the word out -- and honestly, I think the head of the RMA department never really realized it when he wrote the policy (when I asked him when I saw it, he was kinda like "Well...I guess we would huh"). In fact I can show that it's better than about 90-95% of the policies out there (there's a few that's better, but none within the same price range; all the rest are worse of course, including pretty much all the big name-brand ones). But they never thought to market that, relying on direct sales via stores instead, and has thus stayed small.
Dell was founded in 1984. Amptron was founded in 1986. Guess who had (and has) the bigger marketing department.
Anyway, somewhere along the line, I got curious about how Amptron's products compare with everyone else's. On this one, I can be brutally honest (now that I'm no longer working there). The CMV and Polyview monitors that they sell aren't exactly top of the line, but more like, well, the Toyotas of the computer market -- more average and humdrum, not the kind of thing you're going to impress any girl over. Definitely not a sports car. But hey, I drive a Toyota and I'm not basing my self-esteem on my monitor, and I just needed something that worked fine and dandy for office applications and Starcraft, and my Amptron monitor obliged by magically making all ghosting disappear when I play Starcraft despite being a 25 ms panel. Maybe I'm too busy trying to not get killed by zerglings or something. When I compared it with other monitors, though, I was somewhat shocked. Almost all monitors offer some special technology, whether it be opticool or magicawesome or colorfabulous or whatever. I tried to read up on whatever they had on their websites to try to figure out how the technology worked -- I'm an engineer, after all. Part of the reason why I was somewhat shocked, though, was to find out that almost all the numbers given are bogus. Contrast ratio? Measured in complete darkness, and the contrast ratio is actually around 80:1 to 100:1 in regular room lighting. Viewing angle? Defined to be at a point when the image is absurdly crappy. But hey, use a realistic angle and you're left in the dust due to everyone else offering better numbers from more lax standards. So crappy image it is.
***** Why I think Viewsonic's 4 ms spec is bogus
The biggest bogeyman in the "misleading specs war" is response time. As you no doubt know, response time is the time it takes for a pixel to change from a given shade to another one. These things are electronic shutters, after all, kind of like a camera's aperture, and it takes time for them to move from close to open and to anywhere in between. With 256^2 = 65536 possible transitions to choose from, however, it's hard to pick one to report. The previous standard was defined by the ISO to be the sum of the time it takes to go from black to white and from white to black. Sounds good, right? After all, that's the two farthest transitions possible, in terms of how much the crystals have to rotate, so it sounds like it'd be the biggest numbers. Unfortunately, the closer the initial and destination states, the slower the rotation speed, and gray-to-gray actually takes longer than black-to-white or white-to-black -- often longer than them combined.
This is the first thing I have a beef with. Previously, the spec was actually black-to-white-to-black, two transitions. This makes more sense than you might think. They knew that the transitions weren't equal -- one was longer than the other, so both were required. This is not that uncommon when measuring, actually; for example, land car speed records are required by Guinness to be the average of both directions, to cancel out effects due to wind speed, slope of terrain, etc. Nowadays though, with overdrive and other technologies on the rise, it's become advantageous for companies to use gray-to-gray instead. Partly because of user complaints, but no doubt because with overdrive, gray-to-gray is faster than black-to-white. Also, gray-to-gray...is only one direction! Because no one likes to say "gray to gray to gray", companies have managed to cut their newest numbers in half, by only reporting one transition rather than the sum of two transitions. Now consumers have to multiply any gray-to-gray measurement by two to mentally compare it with previous numbers -- but of course, the average consumer isn't going to know to do that.
Anyway, I watched with some derision when Viewsonic launched their 4 ms media blitz. They even registered (or at least use) fastresponsetime.com to promote their product. Already I didn't like it when they called their marketing ad a "white paper", since those are supposed to be academic and highly technical and analytical (compare what Viewsonic calls a white paper on that website and an original one on the same subject, which came out in 2001: http://www.mitsubishielectric.co.jp/service/tft_tech/new/img/sid_2001_29_03.pdf ). But no doubt some people who see that will think "wow this has gotta be the truth because it's a white paper" and put their advertisement-shields down. But hey, near the end, they do give the graphs to back it up, and even though quite a few numbers are closer to 5 ms than 4 ms, there's enough that's low that they can justify their marketing.
Now though, Tom's Hardware has gotten one of those monitors, and tested it out. And Viewsonic's numbers turn out to be completely bogus.
The reason is one of the most elementary parts of control dynamics: when making a dynamic measurement, the end time is when the value stays within an envelope around the desired value, not when it first reaches that value (the envelope is necessary because of data noise and because it's hard for the actual value to be at exactly the desired value). It's hard to emphasize enough how big of a difference this is. Take cruise control. Say you're going 30 mph and you want to go 60 mph, and you got a good cruise control system so you put in 60 mph and let it run. It does you no good if the cruise control floors the pedal and zooms up to 85 mph, then realizes it's going too fast and so slams on the brakes, then at 45 mph it realizes it's going too slow and so floors the pedal, etc., until it gradually stays around 60 mph. It does no good to other drivers nearby either. So as an engineer designing a cruise control system, you want to make it get to the desired value fast, but you also understand that you'll sacrifice some of that convergence speed in order to ensure that you don't oscillate around the intended velocity too much. It really becomes a tradeoff issue, not unlike a damped mass-spring system -- underdamped and the mass keeps oscillating around the "intended" value, overdamped and it takes too long to get there. I can't take credit for this example, by the way -- it was on my very first homework assignment in my control dynamics class, what values for gain and stuff to use for a cruise control system for a car to go from 0 to 60 mph in the minimum time needed to end up staying within 5% of the intended value (60 mph). That's how elementary it is.
Well, it seems like in order for Viewsonic to justify 4 ms, they overshot their mark. As Tom's Hardware shows through oscilloscope graphs, the value first zooms up way high, then gradually settles down to the desired value. The example shown was of going from 0 to 175: the pixel first goes up to 210, then settles down to 175 afterwards over the next several frames. That's a 20% overshoot, folks. If you wanted to go 65 mph, you're now travelling at 78 mph. Hope no cops are around. So while the pixel does reach the requested value in 4.5 ms (needing 4 ms to be within 10% of the requested value), it doesn't actually stay within 10% of the requested value until 31 ms afterwards. That's a long time to be more than 10% off from the color you wanted. But Viewsonic used the first time the pixel's shade reached that value in their reported measurements. Apparently, then, it's easy to have a very quick response time calculation now -- under this logic, when testing for gray-to-gray, you should really just have the monitor alternate between black and white; during the transitions, it'll go from the initial to the desired values at some point, and this minimizes the number that you have to report. No doubt some legal team somewhere is already writing a brief defending this.
There are recent reports of a "sparkle" effect when watching movies using Viewsonic monitors with overdrive. After reading the Tom's Hardware article, I have no doubt that this is because of the overshoot. Which is really a shame, since now basically people will associate overshoot with bad picture quality and extra sparkles, and it's completely unnecessary -- if the liquid crystals are quick enough to overshoot, Viewsonic can also easily just lower the intermediate voltage until the actual value is at the desired value by the end of the frame (rather than over), in order to remove the sparkle effect. But after the advent of 16 ms monitors, then 12 ms, then 8 ms, they had to find a way to justify 4 ms. Unfortunately, now all the other manufacturers who are coming out with overdrive will have people think their monitors will be sparkly, and possibly avoid them as a result.
It's getting annoying. Benoit (the author of the article) said it right when he that "The ISO latency measurement protocol isn't perfect - far from it - but at least it is relatively well defined, with effective protections against exaggeration. With the GTG system, we're getting announced values that have nothing to do with reality most of the time." Because gray-to-gray is not any sort of regulated standard or uniform code, manufacturers are free to come up with whatever system they want to report their numbers under its guise. And those numbers very rarely favor the consumer.
Tom's Hardware Guide (the important page):
http://graphics.tomshardware.com/display/20050602/viewsonic-05.html (and onwards from there)
/rant
Chuck Hsiao
Formerly of Amptron
***** Preamble (skip this part if you just want to read about why I think Viewsonic's numbers are false)
I graduated with a degree in Mechanical Engineering. As an engineer, the general philosophy is to approach things carefully -- always err on the side of caution, always check your work, don't try to overstate your results, but always hedge your work and state its limitations because you can get people killed otherwise. We don't want no bridge collapsing because some engineer said it can support 100 tons but it really only supports 50 tons. We don't even want an engineer saying it can support 100 tons if it really can support 100 tons, because we're taught to put safety factors in our work -- so a bridge that is designed to support 100 tons is only going to be rated for 70 tons or so (safety factor is usually about 1.5 or so).
This is the opposite of a marketer. You'll notice that I often speak out against marketers when I post. Why? Because it seems like a marketer's maxim is that "anything that can't be proven false is true" in terms of what they say. A corollary is that they can define things however they want. This is why I say that a company's marketing team says whatever they think the legal team can defend in court based on the engineering team's results. If they can back it up in court (and of course, by using a lot of "when we said this, what we really meant was...") then they'll say it to make money. The unfortunate thing is that it works -- it does bring in more money than otherwise, simply because most consumers don't have enough technical knowledge to really understand the terms, and thus tend to gravitate toward cool-sounding words ("Amplified Impulse") or easy-to-grasp numbers ("4 ms").
My interest in LCDs was sparked when I started working for Amptron. It's a (relatively) tiny company located in Southern California, and I do sort of feel sorry for the people there, because it's a small company because it doesn't have the budget to become a bigger company because it doesn't have a marketing department to fool everyone into buying its products over that of other companies and thus get more money rolling in. It's had a zero (whole) dead pixel policy since 2000, when it first started selling LCD monitors, because its dead pixel policy is on a subpixel basis, and any three subpixels within 1 cm of each other gets replaced due to its adjacency criteria. A whole dead pixel, of course, is three subpixels within 1 cm of each other. By comparison, HP came out with a similar thing in 2003, and Viewsonic announced that they would be offering a zero whole dead pixel policy as well in 2004. Formac offers a "zero dead pixel warranty" for $99 which if you read the fine print, defines a pixel as an RGB element (all three have to be dead for it to count as a dead pixel), defines "dead" as dark-only (bright pixels don't count), and due to "rough shipping conditions" they won't replace a monitor unless it has three dead pixels or more -- this for an extra $99 "zero dead pixel warranty". Oh, and theirs expires after two weeks. With other companies doing that, you would think that a company offering such a policy since 2000 would be making waves. Nope. No marketing department to get the word out -- and honestly, I think the head of the RMA department never really realized it when he wrote the policy (when I asked him when I saw it, he was kinda like "Well...I guess we would huh"). In fact I can show that it's better than about 90-95% of the policies out there (there's a few that's better, but none within the same price range; all the rest are worse of course, including pretty much all the big name-brand ones). But they never thought to market that, relying on direct sales via stores instead, and has thus stayed small.
Dell was founded in 1984. Amptron was founded in 1986. Guess who had (and has) the bigger marketing department.
Anyway, somewhere along the line, I got curious about how Amptron's products compare with everyone else's. On this one, I can be brutally honest (now that I'm no longer working there). The CMV and Polyview monitors that they sell aren't exactly top of the line, but more like, well, the Toyotas of the computer market -- more average and humdrum, not the kind of thing you're going to impress any girl over. Definitely not a sports car. But hey, I drive a Toyota and I'm not basing my self-esteem on my monitor, and I just needed something that worked fine and dandy for office applications and Starcraft, and my Amptron monitor obliged by magically making all ghosting disappear when I play Starcraft despite being a 25 ms panel. Maybe I'm too busy trying to not get killed by zerglings or something. When I compared it with other monitors, though, I was somewhat shocked. Almost all monitors offer some special technology, whether it be opticool or magicawesome or colorfabulous or whatever. I tried to read up on whatever they had on their websites to try to figure out how the technology worked -- I'm an engineer, after all. Part of the reason why I was somewhat shocked, though, was to find out that almost all the numbers given are bogus. Contrast ratio? Measured in complete darkness, and the contrast ratio is actually around 80:1 to 100:1 in regular room lighting. Viewing angle? Defined to be at a point when the image is absurdly crappy. But hey, use a realistic angle and you're left in the dust due to everyone else offering better numbers from more lax standards. So crappy image it is.
***** Why I think Viewsonic's 4 ms spec is bogus
The biggest bogeyman in the "misleading specs war" is response time. As you no doubt know, response time is the time it takes for a pixel to change from a given shade to another one. These things are electronic shutters, after all, kind of like a camera's aperture, and it takes time for them to move from close to open and to anywhere in between. With 256^2 = 65536 possible transitions to choose from, however, it's hard to pick one to report. The previous standard was defined by the ISO to be the sum of the time it takes to go from black to white and from white to black. Sounds good, right? After all, that's the two farthest transitions possible, in terms of how much the crystals have to rotate, so it sounds like it'd be the biggest numbers. Unfortunately, the closer the initial and destination states, the slower the rotation speed, and gray-to-gray actually takes longer than black-to-white or white-to-black -- often longer than them combined.
This is the first thing I have a beef with. Previously, the spec was actually black-to-white-to-black, two transitions. This makes more sense than you might think. They knew that the transitions weren't equal -- one was longer than the other, so both were required. This is not that uncommon when measuring, actually; for example, land car speed records are required by Guinness to be the average of both directions, to cancel out effects due to wind speed, slope of terrain, etc. Nowadays though, with overdrive and other technologies on the rise, it's become advantageous for companies to use gray-to-gray instead. Partly because of user complaints, but no doubt because with overdrive, gray-to-gray is faster than black-to-white. Also, gray-to-gray...is only one direction! Because no one likes to say "gray to gray to gray", companies have managed to cut their newest numbers in half, by only reporting one transition rather than the sum of two transitions. Now consumers have to multiply any gray-to-gray measurement by two to mentally compare it with previous numbers -- but of course, the average consumer isn't going to know to do that.
Anyway, I watched with some derision when Viewsonic launched their 4 ms media blitz. They even registered (or at least use) fastresponsetime.com to promote their product. Already I didn't like it when they called their marketing ad a "white paper", since those are supposed to be academic and highly technical and analytical (compare what Viewsonic calls a white paper on that website and an original one on the same subject, which came out in 2001: http://www.mitsubishielectric.co.jp/service/tft_tech/new/img/sid_2001_29_03.pdf ). But no doubt some people who see that will think "wow this has gotta be the truth because it's a white paper" and put their advertisement-shields down. But hey, near the end, they do give the graphs to back it up, and even though quite a few numbers are closer to 5 ms than 4 ms, there's enough that's low that they can justify their marketing.
Now though, Tom's Hardware has gotten one of those monitors, and tested it out. And Viewsonic's numbers turn out to be completely bogus.
The reason is one of the most elementary parts of control dynamics: when making a dynamic measurement, the end time is when the value stays within an envelope around the desired value, not when it first reaches that value (the envelope is necessary because of data noise and because it's hard for the actual value to be at exactly the desired value). It's hard to emphasize enough how big of a difference this is. Take cruise control. Say you're going 30 mph and you want to go 60 mph, and you got a good cruise control system so you put in 60 mph and let it run. It does you no good if the cruise control floors the pedal and zooms up to 85 mph, then realizes it's going too fast and so slams on the brakes, then at 45 mph it realizes it's going too slow and so floors the pedal, etc., until it gradually stays around 60 mph. It does no good to other drivers nearby either. So as an engineer designing a cruise control system, you want to make it get to the desired value fast, but you also understand that you'll sacrifice some of that convergence speed in order to ensure that you don't oscillate around the intended velocity too much. It really becomes a tradeoff issue, not unlike a damped mass-spring system -- underdamped and the mass keeps oscillating around the "intended" value, overdamped and it takes too long to get there. I can't take credit for this example, by the way -- it was on my very first homework assignment in my control dynamics class, what values for gain and stuff to use for a cruise control system for a car to go from 0 to 60 mph in the minimum time needed to end up staying within 5% of the intended value (60 mph). That's how elementary it is.
Well, it seems like in order for Viewsonic to justify 4 ms, they overshot their mark. As Tom's Hardware shows through oscilloscope graphs, the value first zooms up way high, then gradually settles down to the desired value. The example shown was of going from 0 to 175: the pixel first goes up to 210, then settles down to 175 afterwards over the next several frames. That's a 20% overshoot, folks. If you wanted to go 65 mph, you're now travelling at 78 mph. Hope no cops are around. So while the pixel does reach the requested value in 4.5 ms (needing 4 ms to be within 10% of the requested value), it doesn't actually stay within 10% of the requested value until 31 ms afterwards. That's a long time to be more than 10% off from the color you wanted. But Viewsonic used the first time the pixel's shade reached that value in their reported measurements. Apparently, then, it's easy to have a very quick response time calculation now -- under this logic, when testing for gray-to-gray, you should really just have the monitor alternate between black and white; during the transitions, it'll go from the initial to the desired values at some point, and this minimizes the number that you have to report. No doubt some legal team somewhere is already writing a brief defending this.
There are recent reports of a "sparkle" effect when watching movies using Viewsonic monitors with overdrive. After reading the Tom's Hardware article, I have no doubt that this is because of the overshoot. Which is really a shame, since now basically people will associate overshoot with bad picture quality and extra sparkles, and it's completely unnecessary -- if the liquid crystals are quick enough to overshoot, Viewsonic can also easily just lower the intermediate voltage until the actual value is at the desired value by the end of the frame (rather than over), in order to remove the sparkle effect. But after the advent of 16 ms monitors, then 12 ms, then 8 ms, they had to find a way to justify 4 ms. Unfortunately, now all the other manufacturers who are coming out with overdrive will have people think their monitors will be sparkly, and possibly avoid them as a result.
It's getting annoying. Benoit (the author of the article) said it right when he that "The ISO latency measurement protocol isn't perfect - far from it - but at least it is relatively well defined, with effective protections against exaggeration. With the GTG system, we're getting announced values that have nothing to do with reality most of the time." Because gray-to-gray is not any sort of regulated standard or uniform code, manufacturers are free to come up with whatever system they want to report their numbers under its guise. And those numbers very rarely favor the consumer.
Tom's Hardware Guide (the important page):
http://graphics.tomshardware.com/display/20050602/viewsonic-05.html (and onwards from there)
/rant
Chuck Hsiao
Formerly of Amptron
Facebook
Twitter