Latency Measurement

[psorcerer]

Oh, finally somebody put a good coherent read on latency measurement.
http://bravenewgeek.com/everything-you-know-about-latency-is-wrong/
I hope fools from Digital Foundry and other sites that measure "frame rates" and "frame times" will finally read it.
But who am I kidding...

 

[tuna]

Some of this is relevant to games, most is not.

 

[function]

The talk appears to be about measuring the time of individual operations, calculating the likelihood of a user action (requiring multiple operations) running into an operation within a given percentile of operation times.

Not sure how this translates to what DF should be doing? DF won't have access to benchmark individual operations within a frame.

Frame rate(s), percentile frame time (which exposed issues with PC drivers a few years back), and input -> output latency would seem to be the best an outside party could do.

 

[DieH@rd]

I wonder why did nobody measure latency of RotTR on Xbone? Many complained lag severley impacts combat, but even DF did not measure it.

 

[Shifty Geezer]

Oh, finally somebody put a good coherent read on latency measurement.
http://bravenewgeek.com/everything-you-know-about-latency-is-wrong/
I hope fools from Digital Foundry and other sites that measure "frame rates" and "frame times" will finally read it.
But who am I kidding...

Nice attitude. "Everyone's dumb and wrong but won't listen to me." How much of that statistical representation about 99% and 9.999% percentiles will be effectively communicated with DF's typical reader? The wonky metrics, flawed as we all know and have discussed, do at least provide a consistent reference point in an unimportant, non-critical hobby.

If you wanted to discuss how latency and framerate metrics could be improved, you could have started a decent discussion with an offering and some ideas. But to open with a broadside that doesn't appreciate the context of the system you're criticising? Yeah, good luck finding any takers for a serious conversation here...

 

[VGA]

Displaying frame times in real time is probably the best way to measure the smoothness of a game. I don't see anything wrong with Digital Foundry's measurements in that regard.

 

[AlNom]

But to open with a broadside that doesn't appreciate the context of the system you're criticising? Yeah, good luck finding any takers for a serious conversation here...

"Perhaps today is a good day to die post! Prepare for ramming speed latency! "

"Sir, there's another starship mod coming in. It's the Enterprise Shifty Geezer.

 

[Arwin]

Oh, finally somebody put a good coherent read on latency measurement.
http://bravenewgeek.com/everything-you-know-about-latency-is-wrong/
I hope fools from Digital Foundry and other sites that measure "frame rates" and "frame times" will finally read it.
But who am I kidding...

Having (admittedly speed-) read the article, I don't actually see the difference here. Digital Foundry always focusses on what the end-user might actually notice, and gives relative performance indications, shows performance over time, and the graphs overlayed on the video don't actually average out. They indicate what part of the screen tearing happens because they know top screen tearing is less annoying, and they've done a fair share of actual latency measurements going from controller input to on-screen response, that variation in response times is worse than a consistent response time, etc. Lots of nuance and user focussed stuff in there, in other words.

And so I don't really see your point here, or much that conflicts with the article at all.