Thread: SysInfo 4.0
View Single Post
Old 08 November 2012, 08:57   #35
Join Date: Jun 2010
Location: out in the wild
Posts: 1,133
There's a few minor things that I learned to do (or not to do) on Sysinfo:

- don't trust the MHz output. I highly depends on memory speed. I suggest to replace this routine by something that completely runs in the CPU cache and that runs way longer. Calculation should not be referenced against the chipset, but against the RTC if found.

- don't move the mouse while measuring HD speed. Full multitasking is still in place during measurement, and if you're moving the mouse, it takes up a lot of CPU time, reducing the measured value. No idea how to work around this.

- a full 68020 is shown as EC020, although some addresses that Sysinfo is displaying are way outside the 16MByte barrier. A more complete mirror-test of the memory space will be required here.

- EClk calculation is questionable, probably derived from the Agnus/Alice type found. EClk is generated from the chipset clock (it is chipset clock divided by 40). This can (most probably) only be calculated if an RTC is found, because that has it's own crystal that can serve as a reference. The differencies between board clocks are tiny, so only a crystal reference will give the required precision. You'll have to measure a 0.92% difference if you want to detect the PAL vs. NTSC frequency (28.63636MHz NTSC vs. 28.37516MHz PAL). Some computers have a 28.322MHz clock installed, challenging you to measure a 0,19% difference between correct PAL and "standard VGA crystal" that was commonly used as a replacement part by repair places. It might take several seconds for a measuring routine to gain the required precision.

- Sysinfo opens it's own screen (std. hires). While this is good for vendors of flickerfixers, it might be an idea to have it clone the screenmode that it was launched from - a similar option that Dopus4 has. This makes it compatible with the largest number of monitors, but adds challenges to performance calculation due to different chipmem access speeds. An easy workaround would be to switch off all chipset DMA for performance testing.

That should serve as a filler for a few long winter nights ;-)

Schoenfeld is offline  
Page generated in 0.05464 seconds with 9 queries