Captain Ned wrote:flip-mode wrote:Yes, they absolutely would. 99th percentile FPS will be a smaller number than the full average FPS. Think them marketers are going to use a smaller number? Even sill, we really have come back to FPS, even if only 99% of the way back
Ugh. All I can say is that you'd rather advertise big numbers over small numbers even when the small number better describes what's really going on.
EDIT: At the end of the day, this is the real problem. Far too many people have a visceral and negative reaction to a measurement where the smaller number is better.
In the car world people can deal with small 0-60/0-100 numbers, why can't we? We deal with memory latencies with low numbers too. So does "time to encode X video".
Ned, I do believe the OP and a few others are not saying they want bigger numbers. They just want the term brought back somehow, even if it is d(FPS)/dt (flip, you need to brush up on your calculus
). Unfortunately for game testing, you can't cut infinitely small where t -> 0. All we have are individual frames, and each frame does take non-zero amount of time to be rendered.
blitzy wrote:If someone says a game will run at 60 FPS I know roughly what to expect, in general anything less than 30 frames is not great and I can imagine just how smooth a game will run based on an FPS figure. An FPS value gives me a pretty good estimate of performance that I can visualize based on my own experience. I can watch examples of games running at different FPS values and get an understanding of the respective performance.
Actually that single number has been somewhat obsolete about 5 years ago, as review sites began to use other numbers in an attempt to get a more complete picture:
1. min/max FPS numbers
2. a somewhat coarse measurement of "at no point in time the fps number dropped below 30 (this is affected by the measuring interval where it can miss "slow enough" frames), this is kind of a precursor to the 99th percentile, but not quite.
3. numbers from game runs using different scenes/maps
The new line graphs actually show us more interesting things in addition to "summarized" results. Constrast the following 2:
1. You have a group of high latency frames only close to each other, but the majority of the time the line is pretty smooth.
2. You have high latency spikes almost all the time, but each occurrence lasts shorter.
#1 will give you long lags but fewer of them. #2 you will be stuttering almost all the time. The number of high latency frames, and the 99th percentile count, are the same in both cases. Which is more preferred? That may be up to the person. Personally I think #2 gives a worse experience overall because it is not smooth all the time. Something just listing a few numbers cannot show us.
We need the numbers
and the graphs.
The Model M is not for the faint of heart. You either like them or hate them.
Gerbils unite! Fold for UnitedGerbilNation, team 2630.