Atomic I/O letters column #116

Originally published 2010, in Atomic: Maximum Power Computing
Reprinted here April 11, 2011
Last modified 28-Apr-2012.
 

Laaaaaag!

I've just moved from using my PS3 on an Asus MW221U 22" monitor (HDMI-DVI Adapter, PS3@720P) to a Sony KDL40HX800 40-inch LCD (HDMI, PS3@1080P). On connecting up the shiny new panel something is just not the same when it comes to game play, mostly in MW2. I think what I am trying to overcome is input lag on the Sony.

Following a few of the standard things like enabling the "game mode", which disabled a selection of the unnecessary processing options, and directly connecting the PS3 to the panel rather than through the AV receiver, there was a noticeable difference. Playing between both panels you still get a feeling you are behind the events on screen, though, and the results at the end of the match speak for themselves.

I have done an input lag test by running the 22-inch and 40-inch as clones from my PC from each output of a Radeon 4870, DVI direct to the 22" and with a DVI to HDMI adapter to the 40". Based on the timers, there is 40-60ms delay to the Sony when running in its default game mode and up to 100ms on other settings. I know this is a small difference but I guess being accustomed to the smaller screen is a factor when you are used to how things react. As every display will have some sort of input latency to process the picture and chit chat over content protection I now wonder what the latency is on the Asus, which seemed to be great in hindsight, although I'm over the small screen and it was 16:10.

With such a range of TVs out there and a lot of people using the big screens for HD gaming, is big-screen choice contributing to the potential of some gamers and the frustration of others? Trying to find information on the panels prior to purchase is a nightmare, with most people confusing it with pixel response time. Apart from trying the VGA input on the TV from the PS3 (not sure how that'd work out), and maybe exploring the service menu, is there something else I can do to reduce the time it takes to get the picture on the screen?

Andrew

Earth-to-moon light-speed chart
I'm sure we've all wondered how far your monitor would have to be from your couch for light-speed to create perceptible lag. Click for legible version!
(200,000-kilometre cables taking the signal to the monitor in the first place would more than double these numbers.)

Answer:
First: Congratulations on actually doing a controlled test! It's easy to "feel" lag that isn't really there, just as it's easy to decide sugar pills are a medical breakthrough or that magic pebbles make loudspeakers sound better, if you've an inclination to believe those things. (Note that people have a tendency to "see" not only things that they want to be true, but also things that they fear are true.)

My quick answer to your actual question, though: No, there's nothing else you can do. There are a few things I could suggest, but you've done them all already!

The actual input lag of a modern HDTV with "game mode" turned on and every other image-enhancement feature you can find turned off is, usually, too small to notice.

The problem is that there's also an inescapable baseline lag for any 3D software - on a console, or a PC.

The graphics card can't "rasterise" an image for display until it's rendered it, and it can't render it until the CPU has figured out what needs rendering, and there's a little more delay on top of that because the CPU can't deal with new input if it's still busy with previous input. If a given hardware-and-game combination quantises logic, rendering and rasterising all at the same 50Hz speed starting at the vertical-blank moment, then it's impossible for any input to produce output less than three frames later. That gives a baseline of 3/50ths of a second of lag, or 60 milliseconds.

(Things get more complicated when simulation, rendering and output threads are running at different speeds. Many modern games poll input very fast, and have fast simulation threads, but never output more than 30 distinct frames per second. It's also possible to hand data to the GPU before a whole frame worth of it has been calculated, and monkey with vblank timing.

And then there are games that look laggy, because the rendering thread lags behind the simulation one. So you see your gun fire six frames late, but the bullet actually killed the target only two frames after you pressed the button. (This is the opposite of unhelpful client-side prediction in networked games, where you see yourself shoot someone but a moment later the server informs your game client that actually, he shot you first.)

For almost all 3D games, you can bet on at least three frames of genuine lag. Some games are much worse.

The user generally can't do much about this baseline lag, but it makes display-device lag more important. 60ms of lag is only noticeable if you're a pretty twitchy gamer playing a pretty twitchy game, but 60ms plus another 60ms of output-device lag will be noticeable to even quite dozy Guitar Hero players. (Multiplayer network lag adds its own special flavour to the stew.)

It is indeed difficult, if not impossible, to get HDTV manufacturers to cough up input-lag numbers. This is partly because the numbers vary depending on what the HDTV's displaying, but mainly because lag numbers aren't getting any better, as the manufacturers add fancier and fancier image-enhancement features.

Most HDTV buyers care about a punchy, colourful picture, but seem to be unable to detect sound that's a quarter of a freakin' second out of sync with the video. So HDTV makers have incentive to fiddle more and more with the image, and no incentive to reduce the lag that this fiddling creates.

(And yes, Copy-Control Crap like HDCP may add some lag of its own. Hurrah.)

Once you've got the HDTV in your house you can measure lag by, as you did, setting up side-by-side screens (an old CRT TV or monitor has zero inherent lag), or by pointing a video camera at the HDTV and your controller and counting frames between button-press and on-screen action. (Infinity Ward got Ben Heckendorn his own bad self to make them an LED-equipped button-press indicator!)

The video-camera method lets you figure out whether a given setting change really made a difference, but if it tells you your HDTV is lagtacular no matter what you do, your only options are to put up with the lag, or make friends at the electronics shop by returning the TV.

If you haven't bought your HDTV yet, you can calculate lag by taking your PC or console to the shop, which will conveniently probably also have a video camera you can use. The chance of this being doable in any randomly-chosen shop is low, but a bit of phoning around will probably find you somewhere with agreeable gamers on staff.

It's also possible to short-cut the whole palaver if someone's already tested a TV you're considering, and posted the results online. A few years ago the Web-forum pickings in this department were thin, but the situation's getting better as HDTVs get cheaper and cheaper, and more and more people learn about the problem. (If only by wondering why Guitar Hero and Rock Band have those calibration options now.)

 

Measure with micrometer, mark with chalk, cut with axe

How is it possible for several system monitoring programs to report different sets of CPU clocks and temperatures (pic attached)?

Disagreeing system-monitor programs

I've disabled SpeedStep in the BIOS but Core Temp still says that the BIOS is underclocking my CPU - why is this?

I have an i7 920, P6T mobo, and standard cooler.

Ben

Answer:
A temperature sensor - actually, the hardware-monitoring chip to which the analogue sensors are connected - does not actually report a temperature. It reports a number, which different programs (including the BIOS setup program) interpret in different ways.

So even if a given program reckons it recognises whatever hardware-monitoring chip your motherboard (and/or graphics card) uses, it may give wildly wrong numbers if your mobo/CPU/probably-BIOS-version-too combination has changed the relationship between the reported number and the actual temperature.

(You can also get subtly wrong numbers if your temperature sensors read a bit high or low. This latter problem - more common at lower temperatures than higher, I think - is seldom worth worrying about. Just remember that one computer that seems to consistently run three degrees warmer than another may actually not be different at all. An idle-temperature difference may be entirely illusory, even if it's quite large.)

It's sort of like car speedometers, which don't actually directly monitor speed - they monitor how rapidly the wheels are rotating. Put smaller wheels on your car and your speedo will read high, put bigger wheels on and it'll read low. (And your rollin' on 20s will probably soon be interrupted by red and blue lights. For this reason, by the way, most speedos are deliberately set up to read a little high with the car's standard wheels, so just putting higher-profile tyres on won't make the speedo read hazardously low.)

The multiplier works the same way. There's no single standard way for a CPU and motherboard to tell a monitoring program the CPU bus speed, multiplier and/or core speed, so once again the monitoring software has to explicitly recognise the hardware it's running on.

I think Real Temp gets the numbers right for most current systems, including Core i7s. The docs page talks about calibrating the software to get really accurate numbers.

 

Plugging electron-leaks

Normally I'd research this myself, but I'm rather busy atm and thought I'd hopefully do both of us a favor (I get the answer with little effort and you get something to write about).

I remember how happy I was with the first set of NiMH AA batteries that I bought. My digital camera worked for more then a few lousy shots. I soon discovered their limitation, in the form of self-discharge, so I couldn't have them charged and sitting on my shelf waiting for use in a few months when I needed them. Around 2008 I came across Duracell "Active-Charge" low-self-discharge NiMH and bought a set, I thought I'd give them a try and buy another set if they went well after 6-12 months. Surprise surprise, they worked very well, but when I went back to buy some, none of the batteries carried the "Active-Charge" branding.

Do these type of batteries still exist (from Duracell or other common battery manufactures available in retail outlets)? Are they just the standard NiMH that are available everywhere today and not carrying the extra branding?

A.

Answer:
If the package doesn't specifically say "pre-charged" or "ready-to-use" or something of that sort (you probably won't see "low self-discharge" actually written on the pack), the batteries will be the standard high-self-discharge type. I can identify no clear relationship between battery brand and discharge rate; as a general rule, NiMH cells with higher capacity will have higher self-discharge, but that's not a guaranteed relationship, and most, if not all, battery brands inflate the capacity numbers anyway.

Standard NiMH cells are fine for high-discharge, short-standby applications like radio-controlled toys and powering frequently-used photo flashes. They're acceptable for cordless phones and even cordless PC input devices, too, if you're able to put the device on charge when you're not using it.

For applications with significant standby time, though, more expensive "LSD" cells with lower nominal capacity will work much better.

There are rather more brands of LSD NiMH cell than there are actual different kinds of cell under the shrinkwrap. The current "Duracell Pre-Charged" LSDs, for instance, are actually Sanyo cells - just different packaging for the same cells sold under Sanyo's "Eneloop" brand.

I think they still are Sanyos under the skin.

You can save a few bucks if you buy LSD cells from eBay dealers. Lesser-known brands like Uniross's "Hybrio" or Gold Peak's "ReCyko" cost a bit less, but work the same. You'll probably not get scammed!

The price difference isn't large any more, though. So unless you're really pinching pennies or buying a lot of batteries, you might as well buy LSD cells locally.

 

This is a reprint of a column originally published in Atomic: Maximum Power Computing magazine here in Australia. The e-mail address for the I/O column is io@atomicmpc.com.au, but I can't answer all of the letters I get.

If you're not an Atomic reader, then sending mail to their letters address is somewhat perverse. Use this alternative instead.

Other letters columns

I/O #1
I/O #2
I/O #3
I/O #4
I/O #5
I/O #6
I/O #7
I/O #8
I/O #9
I/O #10
I/O #11
I/O #12
I/O #13
I/O #14
I/O #15
I/O #16
I/O #17
I/O #18
I/O #19
I/O #20
I/O #21
I/O #22
I/O #23
I/O #24
I/O #25
I/O #26
I/O #27
I/O #28
I/O #29
I/O #30
I/O #31
I/O #32
I/O #33
I/O #34
I/O #35
I/O #36
I/O #37
I/O #38
I/O #39
I/O #40
I/O #41
I/O #42
I/O #43
I/O #44
I/O #45
I/O #46
I/O #47
I/O #48
I/O #49
I/O #50
I/O #51
I/O #52
I/O #53
I/O #54
I/O #55
I/O #56
I/O #57
I/O #58
I/O #59
I/O #60
I/O #61
I/O #62
I/O #63
I/O #64
I/O #65
I/O #66
I/O #67
I/O #68
I/O #69
I/O #70
I/O #71
I/O #72
I/O #73
I/O #74
I/O #75
I/O #76
I/O #77
I/O #78
I/O #79
I/O #80
I/O #81
I/O #82
I/O #83
I/O #84
I/O #85
I/O #86
I/O #87
I/O #88
I/O #89
I/O #90
I/O #91
I/O #92
I/O #93
I/O #94
I/O #95
I/O #96
I/O #97
I/O #98
I/O #99
I/O #100
I/O #101
I/O #102
I/O #103
I/O #104
I/O #105
I/O #106
I/O #107
I/O #108
I/O #109
I/O #109
I/O #110
I/O #111
I/O #112
I/O #113
I/O #114
I/O #115
I/O #116
I/O #117
I/O #118
I/O #119
I/O #120
I/O #121
I/O #122
I/O #123
I/O #124
I/O #125
I/O #126
I/O #127
I/O #128
I/O #129
I/O #130
I/O #131
I/O #132
I/O #133
I/O #134
I/O #135
I/O #136
I/O #137
I/O #138
I/O #139
I/O #140
I/O #141
I/O #142
I/O #143
I/O #144
I/O #145
I/O #146
I/O #147
I/O #148
I/O #149
I/O #150
I/O #151
I/O #152
I/O #153
I/O #154
I/O #155
Dan's Data letters #1
Dan's Data letters #2
Dan's Data letters #3
Dan's Data letters #4
Dan's Data letters #5
Dan's Data letters #6
Dan's Data letters #7
Dan's Data letters #8
Dan's Data letters #9
Dan's Data letters #10
Dan's Data letters #11
Dan's Data letters #12
Dan's Data letters #13
Dan's Data letters #14
Dan's Data letters #15
Dan's Data letters #16
Dan's Data letters #17
Dan's Data letters #18
Dan's Data letters #19
Dan's Data letters #20
Dan's Data letters #21
Dan's Data letters #22
Dan's Data letters #23
Dan's Data letters #24
Dan's Data letters #25
Dan's Data letters #26
Dan's Data letters #27
Dan's Data letters #28
Dan's Data letters #29
Dan's Data letters #30
Dan's Data letters #31
Dan's Data letters #32
Dan's Data letters #33
Dan's Data letters #34
Dan's Data letters #35
Dan's Data letters #36
Dan's Data letters #37
Dan's Data letters #38
Dan's Data letters #39
Dan's Data letters #40
Dan's Data letters #41
Dan's Data letters #42
Dan's Data letters #43
Dan's Data letters #44
Dan's Data letters #45
Dan's Data letters #46
Dan's Data letters #47
Dan's Data letters #48
Dan's Data letters #49
Dan's Data letters #50
Dan's Data letters #51
Dan's Data letters #52
Dan's Data letters #53
Dan's Data letters #54
Dan's Data letters #55
Dan's Data letters #56
Dan's Data letters #57
Dan's Data letters #58
Dan's Data letters #59
Dan's Data letters #60
Dan's Data letters #61
Dan's Data letters #62
Dan's Data letters #63
Dan's Data letters #64
Dan's Data letters #65
Dan's Data letters #66
Dan's Data letters #67
Dan's Data letters #68
Dan's Data letters #69
Dan's Data letters #70
Dan's Data letters #71
Dan's Data letters #72
Dan's Data letters #73
Dan's Data letters #74
Dan's Data letters #75
Dan's Data letters #76
Dan's Data letters #77
Dan's Data letters #78
Dan's Data letters #79
Dan's Data letters #80
Dan's Data letters #81
Dan's Data letters #82
Dan's Data letters #83
Dan's Data letters #84
Dan's Data letters #85
Dan's Data letters #86
Dan's Data letters #87
Dan's Data letters #88
Dan's Data letters #89
Dan's Data letters #90
Dan's Data letters #91
Dan's Data letters #92
Dan's Data letters #93
Dan's Data letters #94
Dan's Data letters #95
Dan's Data letters #96
Dan's Data letters #97
Dan's Data letters #98
Dan's Data letters #99
Dan's Data letters #100
Dan's Data letters #101
Dan's Data letters #102
Dan's Data letters #103
Dan's Data letters #104
Dan's Data letters #105
Dan's Data letters #106
Dan's Data letters #107
Dan's Data letters #108
Dan's Data letters #109
Dan's Data letters #110
Dan's Data letters #111
Dan's Data letters #112
Dan's Data letters #113
Dan's Data letters #114
Dan's Data letters #115
Dan's Data letters #116
Dan's Data letters #117
Dan's Data letters #118
Dan's Data letters #119
Dan's Data letters #120
Dan's Data letters #121
Dan's Data letters #122
Dan's Data letters #123
Dan's Data letters #124
Dan's Data letters #125
Dan's Data letters #126
Dan's Data letters #127
Dan's Data letters #128
Dan's Data letters #129
Dan's Data letters #130
Dan's Data letters #131
Dan's Data letters #132
Dan's Data letters #133
Dan's Data letters #134
Dan's Data letters #135
Dan's Data letters #136
Dan's Data letters #137
Dan's Data letters #138
Dan's Data letters #139
Dan's Data letters #140
Dan's Data letters #141
Dan's Data letters #142
Dan's Data letters #143
Dan's Data letters #144
Dan's Data letters #145
Dan's Data letters #146
Dan's Data letters #147
Dan's Data letters #148
Dan's Data letters #149
Dan's Data letters #150
Dan's Data letters #151
Dan's Data letters #152
Dan's Data letters #153
Dan's Data letters #154
Dan's Data letters #155
Dan's Data letters #156
Dan's Data letters #157
Dan's Data letters #158
Dan's Data letters #159
Dan's Data letters #160
Dan's Data letters #161
Dan's Data letters #162
Dan's Data letters #163
Dan's Data letters #164
Dan's Data letters #165
Dan's Data letters #166
Dan's Data letters #167
Dan's Data letters #168
Dan's Data letters #169
Dan's Data letters #170
Dan's Data letters #171
Dan's Data letters #172
Dan's Data letters #173
Dan's Data letters #174
Dan's Data letters #175
Dan's Data letters #176
Dan's Data letters #177
Dan's Data letters #178
Dan's Data letters #179
Dan's Data letters #180
Dan's Data letters #181
Dan's Data letters #182
Dan's Data letters #183
Dan's Data letters #184
Dan's Data letters #185
Dan's Data letters #186
Dan's Data letters #187
Dan's Data letters #188
Dan's Data letters #189
Dan's Data letters #190
Dan's Data letters #191
Dan's Data letters #192
Dan's Data letters #193
Dan's Data letters #194
Dan's Data letters #195
Dan's Data letters #196
Dan's Data letters #197
Dan's Data letters #198
Dan's Data letters #199
Dan's Data letters #200
Dan's Data letters #201
Dan's Data letters #202
Dan's Data letters #203
Dan's Data letters #204
Dan's Data letters #205
Dan's Data letters #206
Dan's Data letters #207
Dan's Data letters #208
Dan's Data letters #209
Dan's Data letters #210



Give Dan some money!
(and no-one gets hurt)