(This piece has been updated! Scroll down for additional clarification of our measurements.)
Bold claims have been made for PlayStation Move in terms of its latency, or so-called controller lag. At both the GDC press and developer events, Sony itself pegged the lag of the new motion controller at "under one frame".
Previous information from the guys who created the PC drivers for the PlayStation Eye camera also indicated that the mechanism is extremely low latency. In that case, a lag of one frame was measured, and the chances are that in this case it was probably impossible to measure if it was actually any faster.
At the nuts and bolts level, then, it's difficult to argue with the case being put forward when both official and independent sources corroborate each other so closely. In use, however, on actual PlayStation Move software, it's clear that there is some level of latency.
Regular readers of Digital Foundry will know that lag is basically inevitable. The creation and flipping of a framebuffer within the console causes lag, and in measurements (not just ours), the very fastest response has come from the PS3 XMB, which has been pegged at three frames, or 50ms.
On top of that, of course, the proliferation of flatscreen panels means that we have display equipment where latency can change radically even between screens made by the same manufacturer. Anything from two to five frames' additional delay (33ms to 84ms) is fairly common.
Muddying the waters still further is that the PlayStation Move is such a precise implement that game-makers may feel the need to smooth the input, reducing natural player jitter, but again introducing further latency.
So, with all that in mind, I set out to get some idea of the lag in Sony's motion controller while at GDC. The chance to get hands-on with Sony's augmented reality demo (used by the firm itself in demonstrating the controller's tech credentials) was probably the best possible test, and presumably the screens Sony used would be the very fastest consumer-level panels it has bearing in mind just how much "precision" was mentioned at both of the company's GDC events.
A cheap and cheerful Kodak Zi6 handycam was used to film the proceedings. It's an unremarkable camera, but it can film at 60 frames per second, making it very useful for measurements like this where temporal resolution is far more important than the number of pixels.
The augmented reality demo is a worthwhile test for a number of reasons. Firstly, actual game-level processing is at a minimum. The demo is simply displaying the camera feed (in itself used for some of the motion control calculations), and overlaying a couple of fairly simply 3D shapes. Make no mistake, this isn't some Killzone 2 level workout of the PS3 hardware.
While exact latency measurements aren't possible in these conditions, a ballpark idea of the level of response isn't a problem at all. The methodology is remarkably straightforward. Keep your hand as steady as possible, then make fast motions with the controller. Count the frames between your hand moving, and the motion being carried out on-screen.
Equally illuminating is to stop your movement suddenly, then count the frames necessary for your on-screen counterpart to catch up. While not 100 per cent accurate, repeat the process enough times and the frame difference becomes fairly evident.
Bearing all of that in mind, and recognising that we don't know how much latency the display itself is adding, I'd say that a ballpark figure of around 133ms of controller lag (give or take a frame) seems reasonable, certainly not the ultra-fast crispness of response we see from games like Burnout Paradise or Modern Warfare, but fine for most of the applications you would want from such a controller.
Also worth remembering is that the motion control itself is only a part of the mechanics within the Move. Much of the internal mechanics, along with the button presses, will all be sent to the PS3 via the Bluetooth connection: lightning-fast and perfectly equivalent to the response levels of the DualShock 3. In terms of a more definitive test, a 60FPS game with comprehensive support for both Move and the traditional controller could be interesting...
Update: Some interesting feedback to this piece, so I thought an update with clarification on the measurements would be in order.
First up, I chose to use this demo as a best-possible example of Move operating in ideal conditions - no game logic running and basically in "tech demo mode", so as close as conditions allowed to what we would hope to be gameplay at optimum conditions. This is what the video itself is showing - the time taken between input by the player and it being recognised, processed and displayed by Move.
This is the generally accepted definition of controller lag (certainly what's been used in our previous pieces on the subject, and that includes the 200ms pre-production Natal measurement), although we do like to filter out the display lag in the measurement where possible - and in this case, we can't. So hopefully this element of the piece is now clear.
So, to put it plainly, based on frame-counting of the motions shown in the video, and others, the measure of 133ms overall is an indication of the gameplay experience as the camera captured it. So, this includes controller and camera input, processing, display of the completed frame plus the additional latency from the LCD (probably in the region of one to two frames on a set like this).
Interestingly, this measurement puts us in the ballpark area of the PlayStation Eye test I did a couple of weeks ago. As mentioned at the beginning of the piece, there is independent corroboration for Sony's sub-frame latency claim at the mechanical level, but the key measurement comes from gameplay.
So, this is indeed a tentative analysis with one big unknown remaining (the lag of the display itself), and others still undefined, for example whether the demo itself is smoothing the data and introducing further latency as a result.
However, as an indication of the response level of the games played at the GDC reveal (and I played them all), this overall measurement seems sound in the meantime. Once the controller is here in the office, we can get it running on properly calibrated equipment to see how close we are.