Page 1 of 3

Run benchmarks and share your results

Posted: Wed Mar 02, 2016 8:43 pm
by QLvsJAGUAR
What about running the “new PCW Benchmarks” implemented in the QTop-Index program? This can be found in the QL/E part (QLE.win) of THE DISTRIBUTION. See message footer for download link.

There are different ways to invoke the Benchmark program. QL/E comes ready with:

1. A started instance of QTop-Index visible as a button “TINDEX” at the bottom of the Desktop.

2. A Qascade menu entry in the Start Menu (“EXEC Benchmark (QTop-Index) TINDEX”).

3. A path to the exe, just enter EX TINDEX in a SBASIC CLI once the QL/E BOOT process has been completed.

4. For a quick start with a completely fresh machine without booting in the Desktop first, just press <ESC> during the 10 seconds BOOT countdown and then enter:

Code: Select all

LRESPR WIN1_sys_thorst_bin
EX WIN1_app_qtop_tindex
This will then look like this:
Benchmarks_using_QTop-Index_of_QLE_A_How_to_start_standalone-PrtScr.png
On the program’s main screen you will be asked for a “Testdevice” where you please enter a valid directory device to benchmark.

After the benchmark results are displayed you have the possibility to Save the results to a plain text file. You just have to enter a valid directory device as storage location and the program will save a the result file and screen-dump to it. The files are named TINDEX_* (e.g. TINDEX_1740949417_HBA_res and TINDEX_1740949417_HBA_res_scr).

This will then look like this:
Benchmarks_using_QTop-Index_of_QLE_B_Results-PrtScr.png
This Benchmark run resulted in a QTop-Index of 248.751 which is about 250 times faster than a good old unmodified 128KB Sinclair QL.

The TINDEX*res file will hold Content like this:

Code: Select all

TINDEX_res
1.07
2016 Mar 02 21:03:23
1740949403
32372
30846
HBA
x.xx
163840
0
0
0
0
512
256
0
TurboBASIC
0
100
1
win1_
0.023
0.070
0.139
0.349
0.093
0.013
0.687
248.751
QL/E v3.11 under SMSQmulator v2.12 on a Notebook with i7 CPU under Windows 7 64-bit 
Feel free to post your results here or email me the TINDEX_*res* files.

Thanks for sharing!

QL forever!

Re: Run benchmarks and share your results

Posted: Mon Nov 20, 2017 9:06 pm
by Peter
QTop-Index v1.07
Q68: 23.718
Q60/80: 195.892

Is there any table which includes the GoldCard and SuperGoldCard results?

By the way, I just saw a wrong result for the Q60/80 under:
http://qlheaven.blogspot.de/2016/04/rpi ... peeds.html
Couldn't comment without account, so "Artificer" if you read this, feel free to contact me.

Re: Run benchmarks and share your results

Posted: Fri Nov 24, 2017 11:05 am
by Artificer
Just joined the forum so I am not allowed to email peter privately at the moment.

I would say that the QTOP figure posted on the QLHeaven website was not "wrong" as such.
The results of QTOP performance testing for the Q60 depends on cache settings, screen size and colour mode as well as processor speed.

The figure posted on the QLHeaven blog for the Q60 was with the settings in day to day use at QLHeaven.
QLHeaven has investigated the effects of cache etc on the QTOP result and will be posting the findings shortly but can confirm a QTOP index of 194.845 in mode 4, 512x256, copyback enabled and no other software running.

Also Supergoldcard with Aurora gets 13.448, mode 4 and 640x480.

Re: Run benchmarks and share your results

Posted: Fri Nov 24, 2017 11:09 am
by vanpeebles
PM's unlocked for you, welcome! :)

Re: Run benchmarks and share your results

Posted: Fri Nov 24, 2017 11:36 am
by QLvsJAGUAR
Two more hints:
- Do not use the mouse or the keyboard during the benchmark tests.
- To compare two systems, make sure you did use the same directory device in the STORE benchmark.

Re: Run benchmarks and share your results

Posted: Fri Nov 24, 2017 12:05 pm
by Peter
Hi Artificer and thanks for joining! :)

Copyback cache is the normal way to run a 68060. If it must be disabled for everyday use, there is definitely a problem that deserves fixing.
With correct MMU tables, all hardware access is not cached anyway. Self-modifiying executable code is an issue of course, but should be rare, and deserved to be dealth with.

Under Linux, the Q60 always runs with copyback cache activated, and for my personal use, SMSQ/E 2.98 also didn't cause problems with copyback. It could make sense to report your problems, maybe they can be fixed in SMSQ/E.

Peter

Re: Run benchmarks and share your results

Posted: Fri Nov 24, 2017 12:12 pm
by Peter
Artificer wrote:Also Supergoldcard with Aurora gets 13.448, mode 4 and 640x480.
Thanks. That would mean the Q68 is almost twice as fast as SuperGoldCard, which I don't think it is under the same conditions.
If it isn't too much effort, could you please try MODE 4 in 512x256?

Re: Run benchmarks and share your results

Posted: Fri Nov 24, 2017 1:18 pm
by Artificer
Its not much trouble.

SuperGoldcard with aurora, mode 4 512x256 gets a QTOP index of 13.770, not much different from 640x480 as you might expect.
The screen organization is the same which is not the case when comparing the Q60 mode 4, 512x256 with 16bit colour and 1024x512 resolution.

Q60 1024x512 16bit colour, copyback enabled, no jobs other than the TINDEX manages 138.096.

The problem with copyback does not seem to be in smsq/e but mainly in QLiberated programs and older bits of software, enabling copyback restricts what can be run without getting
unexpected effects.

I prefer the RPi for linux.

Re: Run benchmarks and share your results

Posted: Fri Nov 24, 2017 2:19 pm
by Peter
Thanks for retrying the GoldCard benchmark with 512x256. Not bad for the Q68 then - it wasn't designed for speed, but I wanted it not slower than SuperGoldCard at least.

Copyback cache: I wonder whether the apparently self-modifying code is only needed immediately after execution start. That could probably be worked around.

Re: Run benchmarks and share your results

Posted: Sat Dec 02, 2017 10:41 pm
by QLvsJAGUAR
Peter wrote:QTop-Index v1.07
Q68: 23.718
I get almost the same index result when having no other jobs running and having started the program with EW.