[RESOLVED] Timer() vs GetTickCount()

[Ellis Dee]

I've seen it asserted many times on the boards that GetTickCount() is more accurate than Timer(), but frankly I don't see it.

Just now I tossed in a sample check of GetTickCount() into my sorting program (linked in sig) to see how they compare, and set it to benchmark sorting an array with only two (Long) elements using bubblesort.

The results:

Timer: 0.00078 seconds
GetTickCount: 0 milliseconds

Seems to me that Timer() is the clear winner. Am I missing something?

[Logophobic]

Private Sub Command1_Click()
  Dim lngTickA As Long
  Dim lngTickB As Long
  Dim sngTimerA As Single
  Dim sngTimerB As Single
  Dim i As Long
  
  lngTickA = GetTickCount
  Debug.Print lngTickA
  For i = 0 To 9
    Do
      lngTickB = GetTickCount
    Loop While lngTickB = lngTickA
    Debug.Print lngTickB, lngTickB - lngTickA
    lngTickA = lngTickB
  Next i
  Debug.Print
  sngTimerA = Timer
  Debug.Print sngTimerA
  For i = 0 To 9
    Do
      sngTimerB = Timer
    Loop While sngTimerB = sngTimerA
    Debug.Print sngTimerB, sngTimerB - sngTimerA
    sngTimerA = sngTimerB
  Next i
End Sub

 4243671 
 4243687       16 
 4243703       16 
 4243718       15 
 4243734       16 
 4243750       16 
 4243765       15 
 4243781       16 
 4243796       15 
 4243812       16 
 4243828       16 

 64049.06 
 64049.08      0.015625 
 64049.09      0.015625 
 64049.11      0.015625 
 64049.13      0.015625 
 64049.14      0.015625 
 64049.16      0.015625 
 64049.17      0.015625 
 64049.19      0.015625 
 64049.2       0.015625 
 64049.22      0.015625

It should be quite clear that both functions are accurate to 1/64 second.

[Ellis Dee]

Wow, great post Logophobic. Took me a bit to grok what your code proves, but yeah, that's pretty irrefutable.

I wonder why the belief is so pervasive that GetTickCount() is more accurate? Probably nothing more than pro-API anti-VB bias.

Anyone know any way to get accurate timing results with a precision of a single millisecond or less?

[Logophobic]

You can use the QueryPerformanceFrequency and QueryPerformanceCounter APIs. Merri has posted code for these functions wrapped up in a nice little class module. link to post

[Ellis Dee]

Sweet, thanks.

[Limited Atonement]

I wonder if there is a problem right there in your Dimension statements. TimerA has to be a Single and Tickcount uses longs in order for timer to be of an effective precision. I see that long and single both use 4 bytes, but is 'single' access any slower than Long? The only reason I might think so is that it may somehow take time for some sort of scientific notation check through which the single has to go that the long might not have to endure. Let me know if you know the two data types to be the same speed, I'm always interested in that sort of stuff!

Private Sub Command1_Click()
  Dim lngTickA As Long
  Dim lngTickB As Long
  Dim sngTimerA As Single
  Dim sngTimerB As Single
  Dim i As Long

It should be quite clear that both functions are accurate to 1/64 second.

[Logophobic]

Single is slower than Long because floating-point types are slower than integer types. This fact isn't relavent here, as both loops run thousands of cycles before a change is detected. The GetTickCount call seems to be roughly 15 to 16 times faster than the Timer call, but again, this is irrelavent.

Scientific notation is just a way of representing a number in string format, and has nothing to do with numeric data types.

[Limited Atonement]

Dear Logophobic,

Why do you say that GetTickCount being fifteen times faster is irrelavent? That is the point of the original post, right? OHHH! I just took time to re-read the post and now see that precision was the issue. But, my first question remains.

Aaron Laws

Single is slower than Long because floating-point types are slower than integer types. This fact isn't relavent here, as both loops run thousands of cycles before a change is detected. The GetTickCount call seems to be roughly 15 to 16 times faster than the Timer call, but again, this is irrelavent.

Scientific notation is just a way of representing a number in string format, and has nothing to do with numeric data types.

[Merri]

If you're making timing in a loop, it would be insane to keep either one of the calls in the tightest loop, both would slow down the code dramatically (of course Timer would slow down it more noticeably i that case). When placed in a less critical position the net effect of the function call's speed becomes close to irrelevant, so it is more of a matter of taste which one is used.

Do
    ' loose loop
    Do
        ' tight loop doing some very fast math
        ' = stupid place to put Timer, GetTickCount or DoEvents
    Loop
    ' loose loop
    ' = a much better place to put Timer, GetTickCount or DoEvents
Loop

[Ellis Dee]

If you're making timing in a loop, it would be insane to keep either one of the calls in the tightest loop, both would slow down the code dramatically (of course Timer would slow down it more noticeably i that case). When placed in a less critical position the net effect of the function call's speed becomes close to irrelevant, so it is more of a matter of taste which one is used.

Exactly. Nobody cares how long one loop iteration takes; the relevant info is how long the entire loop takes.

Timing such a thing only requires two timing calls, whether they be Timer(), GetTickCount(), or QueryPerformance(): One to start the timer and one to stop.

This thread was a reaction to seeing a lot of recommendations to use GetTickCount() instead of Timer() because it's more accurate. That's clearly not true; you have to use QueryPeformance to enhance timing precision.