Slashdot Mirror
Input Lag, Or Why Faster Isn't Always Better
mr_sifter writes "LCD monitor manufacturers have constantly pushed panel response times down with a technique called 'overdrive,' which increases the voltage to force the liquid crystals to change color states faster. Sadly, there are some side effects such as input lag and inverse ghosting associated with this — although the manufacturers themselves are very quiet about the subject. This feature (with video) looks at the problem in detail. The upshot is, you may want to test drive very carefully any display boasting low integer millisecond pixel response times."
I really thought this was common knowledge.
When I bought my Eizo LCD last summer, the first thing I did was read around. These issues came up immediately.
Long story short: Prad was my friend.
Rotaluclac
No, plasmas have near instantaneous response times that are pretty much identical to what you get on a CRT. The issues you get with a plasma is called "phosphor lag" which has to do with the three colors not quite lining up perfectly and it gives you a trailing image of the colors. It's especially noticeable on high contrast edges or if things are moving really quickly. It can be especially noticeable in gaming, but at least IMO it's much less annoying an artifact than the ghosting, smearing, and horrible motion resolution you get with LCDs (and yes they are present even on 120hz LCDs before someone brings that up).
Even humans who are finely in-tune with this sort of thing can't detect changes under about 10ms.
I knew somebody would make some gross misstatement like "The human eye only sees at 25 fps anyway"
And for that, here is the obligatory link to 100fps.com
In short, the shortest flash a human eye can see depends on a lot of things. These factors are explained thoroughly on that web site. The tl;dr version is this: The human eye can discern A LOT MORE than 25 fps.
The refresh rate needed to avoid flicker with an impulse light characteristic display is unrelated to the frame rate needed for perfectly realistic motion quality. Note however that non-flicking sample and hold displays such as LCDs will produce lower motion quality than impulse response displays of the same refresh rate because of the temporal smearing. (see http://www.microsoft.com/whdc/archive/TempRate.mspx for explanation).
CD-ROMs don't. They use "Zone CAV". It's much cheaper and easier to make a drive spin at a constant angular velocity. Unfortunately that results in higher data rates at the outer edges of the disc, so what drives do is they split the disc up into zones. The disc is spun faster for a zone closer to the center of the disc.
Older CD-ROM drives used straight constant-angular-velocity, and would advertise the fastest data rate (which was at the outer edge of the disc).
The only time a modern CD drive will spin with constant linear velocity is when it's playing back audio in real-time. And even then, many players buffer now, so they use the Zone CAV method anyway.
Actually, the issue here is probably more due to the fact that movies are shot at 24 frames per second. 24 doesn't fit into 60 properly, so there will be times where the scene repeats more in one set of refreshes than another. See wiki entry on Telecine, notably telecine judder: http://en.wikipedia.org/wiki/Telecine
With a 120hz refresh, 24 can go into 120 evenly, so you won't see any choppiness.
Karnal
1. You seem to assume that there actually is some kind of pro gamer gear. All the pro LCDs are actually as in graphics artist pro, and usually actually have the slowest response times of them all. It's "pro" as in "it'll look like that when printed too" (and maybe we'll throw calibration hardware and software in too, 10bit per colour component instead of 8 if it's a several thousand dollar model, led backlight, etc), not as in "it'll display the image in 1ms". It's mostly static images that'll get displayed on those.
The very panel that goes into one already works against you. The fastest ones are TN+Film, but those tend to be in 6 bit per component and dithering instead of 8, have shitty viewing angles (often to the extent that you can see a slight difference between the centre and the corners just because the line from the pixel to your eye falls differently), and at least according to behardware.com the "+Film" part creates more non-homogenity too. The most accurate ones are VA ones (as in, MVA or PVA), but those are also the slowest by far. Guess which goes into a "pro" level display for graphics professionals? Right.
2. If you have that fast reflexes and actually live or die by shooting 1ms earlier, most TFT's have an extra problem: most first buffer the whole image, then scale/display it, because it's the easiest way to deal with scaling an image of a different resolution. Unfortunately they do it even when you use their native resolution.
I.e., what you see on the screen is actually what they received 1 to 3 frames in the past. At, say, 60 fps, on some models you can actualy see the image as it was received 50ms ago. I.e., the difference between 1ms and 5ms latency of the panel is entirely the wrong bottleneck to optimize there.
(Since you mentioned Fatal1ty, last I've heard he used a CRT, btw.)
Better models in this aspect are starting to appear, but it took a while and they're still few and far in between. Mostly because it's not one of the numbers dangled in front of the fashion victims, so there was very little incentive to do anything at all about it.
3. The numbers you get told are by and large... well, not lies, but the standard was written by the vendors for their benefit not yours. E.g., a 5ms display if it's measured black-to-white-to-black can be actually faster than a 1ms grey-to-grey with massive overdrive, and produce less ghosting.
The short and skinny was that the black-to-white-to-black standard was already a lie by itself, and only used because it was the smallest number you can measure without overdrive. The standard as defined by the vendors lets them ignore the first and last 10% of the moving from colour A to colour B. Even that ought to give you cause for thought: that number didn't say "it will reach colour B in time X" but merely "it will get within 10% of colour B within time X". A 10% error is piss-poor on the logarithmic scale of the eye. And it lets them ignore the long asymptotic rest of the curve. But in a transition from black to white or back they can ignore more of the long tail than in a grey to grey transition, according to their own bogus standard, so that's why everyone quoted that.
This all changed when someone invented overdrive. The idea here basically is that you can accelerate faster and overshoot the finishing line if you want to. The measured time still is "in how much time you can get within 10% of the finishing line." It doesn't matter that then you overshoot by 50% and spend even more time coming back asymptotically from the _other_ side. But you can't do much overdrive o
A polar bear is a cartesian bear after a coordinate transform.