Some speed tests on QL hardware
Posted: Fri Mar 27, 2020 12:46 am
A discussion on a different thread lead to some timings being run by Steve Poole with the table below. Thought I'd share with everyone since I found them useful:
I will try and run this on my unexpanded QL when I get a minute.stevepoole wrote:Hi Bwinkel69,
You may be interested in some more timing figures for the QL :
Here is the program I used for the timings :Code: Select all
CONFIGURATION CYCLES/SEC +/- % SPEED REMARKS --------------------------- ----------- ------ ------ ------------------------------------------------------ Ql 128ko JS ? ? ? I don't wish to unplug my SuperGoldCard ! SGC QL under QDOS 1864 .25 x1 No extra multitasking from me. SGC QL with SMSQ/E 8440 .11 x4.5 " " ROMdisc and Hermes. Compaq, 1core, QPC2, 2.8Ghz 173894 37.2 x93 Very variable timings due to monocore multitasking. Asus, 3core, QPC2, 1.9Ghz 138683 4.0 x90 Triple core means better variability when multitasking. " " " " ? ? ? C++ code runs 400 tmes faster than superbasic version !
I tested each basic operator for 600 seconds, to avoid bias from mutitasking, several times, with and without web access....Code: Select all
100 :: 110 REMark Code_timer_bas by S.Poole, v25mar2020 120 REMark empty=168316 on QPC2 3core at 1.9Ghz... 125 CLEAR: overhead=93195: REMark {if d<>date: end if}... 130 CLS: y=PI: sum=0: Max=0: min=1E15: secs=10 135 z='3.141592654': BORDER 0 140 d=DATE: FOR wait=0 TO 1E9: IF d<>DATE: d=DATE: EXIT wait 150 : 160 FOR COUNTs=1 TO secs 170 FOR cycles=0 TO 1E9 180 IF d<>DATE: d=DATE: EXIT cycles 190 REMark 200 END FOR cycles 210 sum=sum+cycles: AT 1,1: PRINT secs-COUNTs!!! 220 IF cycles>Max: Max=cycles 230 IF cycles<min: min=cycles 240 END FOR COUNTs 250 dif=Max-min: hlf=dif/2: pc=(hlf*100)/Max 260 PRINT INT(sum/secs)!!'+/-'!pc;'%': BEEP 3276,1 270 :: 280 DEFine FuNction pie 290 RETurn 3.141593 300 END DEFine
Usually, figures are fairly OK using 10 seconds !
The code to test, for example x=0, is placed between lines 180 and 200. The overhead is a constant and is best ignored.
You may wish to compare the timings with your QL configuration ...
The C++ speed was obtained using the transcribed 'QPC2 Travelleing Salesman Program', some 80ko long...
How does the bare 128ko QL fare ?
Regards,
Steve.