EPT![[image loading]](http://i59.photobucket.com/albums/g309/Tamerlane___/Gamespeedsresults.png)
So, I had done my own approximate testing to see what was the relation between game speed settings (only faster actually) and the clock displayed in the replays. That was before I saw this page http://wiki.teamliquid.net/starcraft2/Game_Speed and I had slightly different results from there, but I thought I was just wrong since I didn't have any timer or anything of the sort at the time... Nevertheless, I recently noticed that the ratio between game length displayed at the end of each game and the length of the replays did not match the values in this table, I wanted to find out why.
At first I thought it would be quite simple, simply reproduce the experiment that was done, so I grabbed a replay and timed the duration of the MULE as in the article. I got rather different results, by a few seconds!! Too big of a difference to be blamed only on reaction time or anything of the sort, so I decided I would repeat the experiment with the production of a unit that took a lot of time (I picked the Thor, but that information is irrelevant) and I decided to make more tests to complete a table with the different game speed settings.
And I've been... Thunderstruck!! (damn that song is so good, gonna play it right now)
In short, here are the main conclusions that I've come to :
=> The game duration that is displayed after each game is accurate (yeah no big surprise here / LDO, just keep reading)
=> The speed at which time flows between the game and the replay is not the same for a common speed setting (that is, if I play a game on "fast" and watch the replay at "fast", the replay will take less "real time" than the game took!)
Since I am quite positive that SC2 is not going at relativistic speeds, there must be some problem with the replayer and hence every calculated number (including the clock) you see in a replay - and possibly when observing a game - is inaccurate.
- Yeah, but we all knew that already!
- I didn't say I was finished...
=> The normal speed setting, when playing a game, is not even "normal" or if you prefer : the game is slower than it should be. That means, your 90 seconds timer on the mule will last longer than 90 seconds when playing at normal...I am obviously not talking about its length in a replay, but an actual game.
(If you're a bit confused up to this point, you might want to stop reading or jump straight to the results)
=> I expect that the build time indicated in the tooltips of all units and building is accurate (to less than .5%) with real time if you calculated that from a replay at "Normal" speed instead of an actual game.
Sadly, there is no normal speed in the replayer to verify this conjecture
So basically, the combined effect of wrong time elapsing in replays and game settings kind of gave me a nightmare to sort out. But let's start with the data I got from actual gameplay.
*** Empiracy *** (yeah, I know, the word doesn't exist but a contraction of "empire" and "piracy" is way cooler than "empiricism"!!)
I timed the building time of many SCVs [Why SCVs? Because they are fast to make, so I can make a lot without spending my evening doing just that. Why make a lot? Well, look at the methodology if you really want to know...] and did an average to have a pretty accurate building time. These are the results :
*Faster
39 SCVs in 486,24 seconds for an average of 12.468 seconds per SCV
=> 36,35% faster than reference.
*Fast
24 SCVs in 350,24 seconds for an average of 14.593 seconds per SCV
=> 14,16% faster than reference.
*Normal
23 SCVs in 402,75 seconds for an average of 17.511 seconds per SCV
=> 3,01% slower than reference
*Slow
9 SCVs in 196,40 seconds for an average of 21,822 seconds per SCV
=> 28,35% slower than reference
*Slower
11 SCVs in 321,07 seconds for an average of 29,188 seconds per SCV
=> 71,71% slower than reference
And here is the game length vs replay duration (not real time)
*Faster
8:44 => 11:50
524 seconds vs 710 seconds => 35,50% faster (or -26,20% slower if your cognitive process is...marginal)
* Fast
I got 2 sets of data here and you'll see why I post them :
5:07 => 5:56 ; 307 seconds vs 356 seconds => 15,961% faster
8:25 => 9:44 ; 505 seconds vs 584 seconds => 15,644% faster
[long and boring paragraph]
Now, since the game length and replay duration are rounded to the nearest second (well, somehow rounded, which method is irrelevant), that means there is a margin of error of a full 1 second for each set of data which translate into a margin of error of +/- 0,356% for the first set and +/- 0,171% for the second set. Not only does the real value has to be between 15,961 and 15,644 - but it has to be between the lower and upper (respectively) extremes within their margin of error. The new interval is [15,605%;15,814%] and this method could be used coupled with greater game length (not exceeding 9:59) to get a very accurate value.
[/long and boring paragraph]
*Normal
8:58 => 8:38
538 seconds vs 518 seconds => 3.86% slower
*Slow
8:53 => 6:52
533 seconds vs 412 seconds => 29,37% slower
*Slower
6:06 => 3:31
366 seconds vs 211 seconds => 73,46% slower
edit : I redid the same test, but this time with the MULE's timer and other units with a long build time to avoid the lag introduced by using an external software (see below).
MULE's timer
*Faster
Ran it 5 times : 64,48 ; 64,76 ; 64,87 ; 64,67 ; 64,87
Average of 64,75 which is 39,00% faster than reference
*Fast
Ran it 5 times : 76,68 ; 76,61 ; 76,62 ; 76,53 ; 76,57
Average of 76,60 which is 17,49% faster than reference
*Normal
Ran it 5 times : 92,56 ; 92,82 ; 92,35 ; 92,68 ; 92,81
Average of 92,64 which is 2,93% slower than reference
Thor's production
*Faster
9 Thors in 489,42 seconds for an average of 54.38 seconds per Thor
=> 37,92% faster than reference.
*Fast
17 Thors in 1088,82 seconds for an average of 64.05 seconds per Thor
=> 17,10% faster than reference.
*Normal
11 Thors in 846,60 seconds for an average of 76.96 seconds per Thor
=> 2,62% slower than reference.
*** Additional test I ran ***
=> I replayed the same game that I played on "Fast" setting, the game time was 8:25 and I measure the real time of the entire game using the software I had : 8:09.23!! That's a difference of 15,77 seconds or if you will : a game replayed using "Fast" setting will be about 3,22% faster than a game played using the same setting!
Note that the value of 15,77 has an unknown margin of error because this is the only instance where there was unremovable delay between the time at which the replay started and me starting the timer. ALTHOUGH, since there was a bit of lag in SC2 to start the replay, the difference is probably insignificant. Therefore, this value probably matches the one reflecting the difference between game time and replay time at Normal setting.
edit :
=> I redid the trick I wrote down in the "Fast" setting game time vs replay time using these sets of data :
1:02 => 1:34
62 seconds vs 94 seconds => 51,61% faster
1:02 => 1:25
After doing the previous 2 tests, I realized that the time spend under any other application than SC2 had an impact on the game time, slowing it down, while the replay time stayed consistent with how long the game length SHOULD have been... So I redid the same test while staying inside the SC2 application, and I got : 1:10 => 1:36 ; 70 seconds vs 96 seconds => 37,1% faster
2:35 => 3:36
155 seconds vs 216 seconds => 39,35% faster
8:44 => 11:50
524 seconds vs 710 seconds => 35,50% faster
9:47 => 13:38
587 seconds vs 818 seconds => 39,35% faster
9:49 => 13:40
589 seconds vs 820 seconds => 39,22% faster
Bottom line is : the ratio between game length and replay duration seems to be fairly inconsistent, which is a bit surprising to me as even if there is a bias introduced by the replayer, I thought this bias would be the same throughout every game played, it would be constant if you will.
*** Methodology and Terminology ***
* Terms I am using
game time = the time elapsed while playing (which is displayed at the end of each game)
replay time = the time elapsed in a replay (by the clock)
real time = the time elapsed in a replay in reality
snap = the interval between production of 2 SCVs, measured using a stopwatch
spread = amplitude between the fastest and slowest snap(s)
* How I proceeded to get these data
- I used the freeware XNote Stopwatch to calculate these times
- I ran SC2 in windowed mode since I have only 1 screen
- I always started the clock when a SCV from the queue started (so, never the first SCV built to remove the delay between starting production and clicking on the stopwatch)
- I didn't take into account SCVs built before the production was stopped due to food cap (I sometimes messed up the timing), that explains partly the big difference between the built SCV count ; boredom and game speed are the 2 other reasons
- I used the snap feature of the software to measure my reaction time and exclude this exogenous variable from the conclusions.
- Snap times are fairly consistent, I excluded the snap times where I messed up because I wasn't paying attention or being distracted, this is to reflect a real reaction time as the beginning and end were always on point (otherwise I restarted)
*** Margins of error ***
As you will see, these margins are very small, almost insignificant, but I thought I had to include those in order to show the reliability of the numbers above. Note that these margins are for the production of the SCVs only, I did not take the time to calculate those of other tests, but the unit production tests will be in the same order of magnitude (but not necessarily for timers).
* Faster
fastest snap : 11.92
slowest snap : 13.03
spread : 1,05
margin of error : 0,53
12.468 +/- 0,014 or 0,109%
* Fast
fastest snap : 14,19
slowest snap : 15,00
spread : 0,81
margin of error : 0,41
result : 14,593 +/- 0,017 or 0,117%
* Normal
fastest snap : 17,04 (data not representative, second fastest snap was 17,25)
slowest snap : 17,89
spread : 0,85
margin of error : 0,43
result : 17,511 +/- 0,019 or 0,107%
* Slow
fastest snap : 21,59
slowest snap : 22,00
spread : 0,41
margin of error : 0,21
result : 21,822 +/- 0,023 or 0,107%
* Slower
fastest snap : 28,97
slowest snap : 29,53
spread : 0,56
margin of error : 0,28
result : 29,188 +/- 0,025 or 0,087%
