Run benchmarks and share your results

[mk79]

Wow, that is a lot. I get some 350-400 on the machines I have here

[XorA]

Almost 1000 is what I got running sQLux on an M1 mac. So only 1/2 that score!

[pjw]

Both tests run using QPC2 4.95 straight from the qle folder and minimal boot:
1668 on i5 desktop (with all my normal stuff running)
1392 on i5 laptop. But scores vary widely on each run (980-1390).

[M68008]

3029

JS ROM, 4-year-old i7 PC, 'QemuFast' variant of Q-emuLator.
Standard Q-emuLator on the same machine gives 1359.

Perhaps the benchmark should run for a longer time for better accuracy? The INTMATH test only run for 3 ms?

[M68008]

A bit unstable at high speed. In some runs a test may report a duration of 0 or 1 ms, resulting in an inaccurate overall score:

[mk79]

Wow, that is a lot. I get some 350-400 on the machines I have here

OK, on the docking station with 512x256 QL colour it's 1598 now But as Daniele says, the speed is basically too fast to get meaningful results anyway. Not sure how you can measure <20ms execution speeds accurately on a standard system.

[QLvsJAGUAR]

Wow, that is a lot. I get some 350-400 on the machines I have here

That's a rather old PC then or running at an energy option which favours battery life very much.

[QLvsJAGUAR]

Both tests run using QPC2 4.95 straight from the qle folder and minimal boot:
1668 on i5 desktop (with all my normal stuff running)
1392 on i5 laptop. But scores vary widely on each run (980-1390).

When results vary, especially when no other jobs performing work/load under QDOS/SMS, chances are high that Windows energy features and clock rate throttling is the reason.

[QLvsJAGUAR]

3029

Gold medal!

Perhaps the benchmark should run for a longer time for better accuracy? The INTMATH test only run for 3 ms?

QTop-Index was written way back in 1988 and last amended in 1995. At that time QXL cards and ATARI TTs were the fastest QLs around.

With foresight I implemented two things:

1. Draw 1000 BLOCKs at startup and see how fast the machines completes that. Based on the results do a specific number of iterations of the banchmarks to get better accuracy.

Code: Select all

888    REMark BLOCKS to evaluate iterations
889    ticktimer%=0
890    SetStart
900    BEEP 0,16,34,2000,5,7
910    PRINT#cCh_UI_stdinp%\\"      random calculated BLOCKs" : REMark v107 Use "BLOCKs" instead of "BLOCK's"
920    FOR i=1 TO 1000
930       xpos%=RND(281)
940       ypos%=RND(23)
950       xdim%=RND(1 TO 25)
960       ydim%=RND(1 TO (24-ypos%))
970       colour%=RND(255)
980       BLOCK#cCh_UI_info%,xdim%,ydim%,xpos%,ypos%,colour%
990       CURSOR#cCh_UI_stdinp%,6,70 : PRINT#cCh_UI_stdinp%,i
1000    END FOR i
1010    blockspeed=fGetDuration
1011    CURSOR#cCh_UI_stdinp%,180,70 : PRINT#cCh_UI_stdinp%," in "&blockspeed&" seconds" : REMark v107 Adjusted CURSOR
1020    BEEP
1030    IF fCheckKeysForEvent%=-1 : RETurn
1040    :
1041    SELect ON blockspeed
1042       =0 : iterations%=100 : ticktimer%=1
1043       =1 : iterations%=50
1044       =2 : iterations%=30
1045       =3 : iterations%=20
1046       =REMAINDER : iterations%=10
1047    END SELect 

2. Implemented a so called TickTmer to "measure" parts of a second

Code: Select all

2297 DEFine PROCedure WaitForNextFullSecond
2298   LOCal d,loop%
2299    d=DATE
2300    REPeat loop%
2301       IF d<DATE : EXIT loop%
2302    END REPeat loop%
2303 END DEFine
2304 :
2305 DEFine FuNction fGetTicksPerSecond
2306   LOCal d,loop%,ticks
2307    WaitForNextFullSecond
2308    d=DATE : ticks=0
2309    REPeat loop%
2310       IF d<DATE : EXIT loop%
2311       ticks=ticks+1
2312    END REPeat loop%
2313    RETurn ticks
2314 END DEFine
2315 :
2316 DEFine FuNction fGetTicksRemaining(inp_par)
2317   LOCal loop%,ticks
2318    ticks=0
2319    REPeat loop%
2320       IF inp_par<DATE : EXIT loop%
2321       ticks=ticks+1
2322    END REPeat loop%
2323    RETurn ticks
2324 END DEFine
2325 :
2326 DEFine PROCedure SetStart
2327    WaitForNextFullSecond
2328    g_start=DATE
2329 END DEFine
2330 :
2331 DEFine FuNction fGetDuration
2332   LOCal d,ticks_remaining,ticks_per_second
2333    d=DATE                                : REMark PRINT#cCh_UI_stdout%,"ticktimer%     ",ticktimer%
2334    IF ticktimer%=0 : RETurn d-g_start
2335    ticks_remaining=fGetTicksRemaining(d) : REMark PRINT#cCh_UI_stdout%,"ticks_remain.  ",ticks_remaining
2336    ticks_per_second=fGetTicksPerSecond   : REMark PRINT#cCh_UI_stdout%,"ticks_per_sec. ",ticks_per_second : REMark PAUSE#cCh_UI_stdinp%
2337    RETurn d+1-g_start-ticks_remaining/ticks_per_second
2338 END DEFine

[QLvsJAGUAR]

Wow, that is a lot. I get some 350-400 on the machines I have here

OK, on the docking station with 512x256 QL colour it's 1598 now

See, it was your energy options (lower performance when on battery).

But as Daniele says, the speed is basically too fast to get meaningful results anyway. Not sure how you can measure <20ms execution speeds accurately on a standard system.

Without assembler, just BASIC (interpreted or compiler) I see at least two ways:
1. Count the number of 20ms to the next second with PAUSE 1
2. TickTimer which first evaluates the number of ticks per second, then counts the number of ticks to the next second

For best results and accuracy of QTop-Index it is adwised to have no other JOBS running. With QL/E simplest way to achieve that - as Per wrote - is to boot QL/E with the "<M>inimal boot flavour".