Digital Foundry vs. Console Lag: Round Two • Page 3

PlayStation 3 and Arc camera put to the test.

The notion of dropped frames impacting on controller response and thus the gameplay experience often crops up in the Face-Off features, and probably the most memorable example of this is Capcom's Resident Evil 5. In this video you'll see first how the game responds when running at its optimum 30FPS, and then you'll see how controller response is affected when frames are dropped.

At best, Resident Evil 5 runs with 100ms latency, but dropped frames can up response time by 50 per cent.

It's not pretty. At its best, we see Resident Evil 5 respond at 100ms, but as frame-rate decreases, latency does the opposite. The same actions within the game can take as long as 150ms. As players we adapt to the lag in our games. Shooting in Halo is consistent at 100ms. Actions in Mirror's Edge are consistent at a higher 133ms. So long as latency is low enough, and it's consistent, we know what to expect from the game and how it should respond. Sharp changes to this state of affairs aren't so welcome and it's at this point that it feels intrusive to the gameplay experience.

In our initial feature, a few unwritten "laws" came into effect based on the testing we did, which backed up previous findings published on Gamasutra by Mick West. While the lowest possible latency was found to be 50ms (on the PS3 XMB), the fastest a 60FPS game can typically respond is in four frames, or 66ms. Testing the likes of Call of Duty: World at War, Street Fighter IV and Ridge Racer 7 confirms that this is the case on PS3 too.

However, a number of examples have cropped up which show that although 60FPS can produce a smoother-looking game, there's no guarantee that you'll hit that magical 3-4 frame response time.

While 66ms, or even 50ms, is the aspirational response time for a 60FPS game, many fall short. The difference is unnoticeable in some cases but anything but in others...

WipEout HD consistently took five frames, or 84ms to respond, while LEGO Batman took anything from 84ms to 216ms (!) to react depending on the action chosen - a fact we can only assume is by design, with the programmers introducing different levels of latency according to the player movement. In our previous feature we measured this at 133ms on Xbox 360 and found that the same motion incurred the same response time on PS3. Dante's Inferno - the most consistently "60FPS" game released in recent times - surprised with a relatively slow 100ms. Visceral's latest looks like a 60FPS game, but it responds like a fast 30FPS title. However, the smoothness of the visuals combined with that consistency in the controller response, plus the fact that we've become conditioned to a response time greater than 100ms, diminishes those shortcomings.

A fair amount of testing was carried out for this feature, not all of which we have the time to present in video form, but to wrap up the practical side of this piece, here's a results compilation:

Game Latency Measurement
BioShock 2 Frame-rate Locked 133-150ms
BioShock 2 Frame-rate Unlocked 100-150ms
Call of Duty: World at War 66ms-100ms
Dante's Inferno 100ms
Killzone 2 150-183ms
LittleBigPlanet 100ms
Mirror's Edge 133ms
MotorStorm: Pacific Rift 116ms-133ms
Resident Evil 5 100-150ms
Ridge Racer 7 66ms
Street Fighter IV 66ms
Unreal Tournament III 100-133ms
WipEout HD 84ms

Any discussion of lag wouldn't be complete without factoring in online gaming. So where does PSN and Xbox Live or video-streaming services like OnLive fit into the equation? It's important to stress that the kind of "system latency" we are talking about bears little relation to the issues you may be having while playing online. However, with OnLive and its ilk, the testing methodology will be of crucial importance in determining how successful they will be.

In terms of PSN and Xbox Live, game code is designed on the principle of prediction. It has been since the days of John Carmack's Quakeworld back in the mid-nineties. The idea is simple. If you lob a grenade, or fire a gun, the consequences of those actions can be calculated before they actually happen. The grenade and bullet trajectories have been calculated and they can be rendered client-side while the host is being informed. Good prediction code makes actual gameplay look seamless, even though it really isn't.

Ever pumped tons of bullets into an opponent and had him gun you down like a stinking pig much sooner? Ever see a Modern Warfare Killcam section look markedly different to your recollection of the encounter? That's lag in effect. But it's not the kind of lag we're discussing here.

Services like OnLive are entirely different. Their very nature makes prediction impossible as the gamer only has a dumb terminal at his command. It decodes and displays video, it plays streamed audio, it takes your controller inputs and beams them back to the host. That's it. So the 60FPS camera tests we use to measure game lag is perfectly valid as a testing mechanism for OnLive. Company front man Steve Perlman is on the record as saying that latency on his service is 80ms. Remarkably - some might say unbelievably - that is faster than playing the vast majority of console games locally.

Hopefully, like the Sony motion controller, this is something we can put to the test Stateside...

Many thanks to Neversoft co-founder Mick West for his pioneering feature on this topic.

Comments (53)

Comments for this article are now closed, but please feel free to continue chatting on the forum!

  • Loading... hold tight!