- The LaST Upgrade -
PART 10 - GemBench6
Last updated - May 18, 2017
For many years GemBench has been one of the best benchmarking apps on the ST. Ofir Gal's last release was GB4 where he was kind enough to allow me to use his source code to update GemBench. After many weeks of work, GemBench5 was born and ultimately GB6.
Updates on GB5 were to take out the registration stuff, so it runs as registered even without the registration files or typing in codes etc. There was some hidden features which I have enabled some. The main one which GB was crying out for was a loop test. So now you can select 1, 10, constant loop testing. Reference boxes were removed and defaults are loaded from a file instead. All the values in GB seemed to be hard coded and the blitter tests seemed to just assume its xx% faster than a stock machine. Overall none of the results were very accurate. A new ROM test was added -which was more accurate on accelerated machines. There have been a lot of bug fixes with various routines. Though as the code was so complex it proved impossible to update GB5 any further. So GB5 was abandoned and never officially released.
GB6 series is a total re-write of GB4. The hisoft "HGK kit" has been totally removed. There seemed to be some odd bugs which were not easy to solve as the HGK framework was so huge. While the framework makes functions easier to use, when it goes wrong, its impossible to de-bug. With so little information about how it all works, It just had to be removed. This made a steep learning curve for learning how GEM/AES/VDI works, but in the long run it makes the new GB6 code a whole lot easier to maintain.
All tests have been tweaked in some way. Mostly test times have increased to take into account fast boosters like the CT60. Each test is clickable as before, and pressing "A" will run all tests, or can be selected from the menu. The blitter on/off has been simplified to one menu and with the addition of "system info" box it displays the current state of the system machine. NOTE - the blitter on/off does not work on the Falcon. This is NOT a bug. For some reason its impossible to turn off the blitter on the Falcon. So regardless of what you try and set the option as, it will always show as ON.
Thanks go to Rodolphe Pineau for recoding RAM/ROM/INT-DIV/TT-RAM tests in assembly. These can also be changed in the menus. It is possible with caches on that results can vary about 2% depending on what data is "cached". The TTRAM/ALT-RAM/FAST-RAM or whatever you want to call it, a test has been added to benchmark alt-ram. If alt-ram is not found, then no results will be displayed. The ROM test (actually the same test in GB5) was re-done as it wasn't accurate in GB4. 16mhz speeds showed almost zero increase in the tests. Now the ROM is read into CPU registers which makes the test generally show 170% on 16mhz speeds. The 30% "loss" in speed is down to RAM access slowing the test loop down. Technically it should be 200%, but can't have it all ways :)
Later builds d.m.l was kind enough to recode the assembly routines and
add some new features. dml has also added routines so GB6 can access the
instruction and data caches on later CPU models. HUGE thanks to dml for
taking the time to code those routines up for me!
GB6 has a new option in the file menu to benchmark your machine. This should only be used when your machine is in a "stock" condition. For example, you could benchmark your stock STFM and that will then be used as your "stock reference". So all your scores will show 100%. Then when you enable a accelerator, you will then have a direct comparison to your own machine. Things like TOS versions can vary in speeds a few % and there is no guarantee that all STFM's run at exactly the same speed either. While I will provide some benchmark files of my own machines, they may not be 100% accurate for other machines. So you should always benchmark your own machine which then "calibrates" GB6 to your own machine. As people can benchmark their own machine, they can send me the benchmark files and I can include them in future releases of GB6.
Quick people will notice "New dialogs" test has been removed. That test wasn't in GB3 even. The test was basically a repeat of the "GEM Dialog Box" only it used the HGK KIT's dialog functions. As nobody is likely ever going to program with the HGK kit the whole test was pointless. More to the point, as the HGK kit was not used in GB6 the test couldn't be done anyway. I found that using AES routines instead to do the test just made it a repeat test of "GEM Dialog Box" so in the end, "New Dialogs" test was taken out. In its place TTRAM test was added which is more useful.
The menu system has been greatly simplified. The sub menu is gone along with all the odd icons which were hard to understand what they were all actually for. Both menus seemed to be a repeat of what was in other menus anyway. The basics are still there in the menus so really there shouldn't be anything "missing". The reference and system info is now displayed on the main screen. It was mostly hidden away in a menu in GB4. This makes it a lot easier to see what your machine is being compared against.
There is also a new "Comparison" box. This draws 3 bars based on the "Statistics" values. By default 100% scores will show a dotted area bar. If your benchmark is 50% faster on the CPU for example, then the bar will be drawn with 50% solid black to indicate visually the speed increase. Similar if you had 25% increase in speed, then it would show 75% dotted and 25% solid black.
In reverse to this, if you benchmark a machine which is slower than the reference machine, they you can end up with negative values. A example would be benchmarking a STFM with a Falcon as the reference. All scores are going to show less than 100%. In which case the "drop" in speed is shown in RED. So if you are running 75% slower, then you will get a 75% RED bar and 25% black dotted area. Please see the readme file for more information on how GB6 works.
Overall GB6 has been totally recoded from the ground up. All tests have been updated in some way, though mostly they look the same as before. The blitter test is changed to a moving box, this puts a harder load on the blitter, so results will show lower than in previous versions.
Some other ramblings may be found on the Atari forum thread. http://www.atari-forum.com/viewtopic.php?f=28&t=29168
GB606 B26 - Released May 18, 2017
GB606 B24 - Released January 3, 2017
GB606 B22 - Released May 12, 2016
GB4.03 Registered is HERE