ACA1221lc@40

The first thing where the card will fail is bus operations. The one bus operation that gets really tight is fastmem access. So the best things you can do is to run programs that use lots of memory, like compiling a large project, creating large LHA archives in RAM: and running a test over these LHA archives over and over again.

When I did the final fixes to the ACA1221lc at 40MHz (=80MHz memory clock), I wrote a simple shell script that would first create a large LHA in RAM, then test it over and over again over night. In most cases, it would make it through the night without a heatsink, but some cards crashed pretty quickly, which is why I decided to put a heat sink on all of them.

I also had in mind to collect data from users about overclocking stability of different CPU masks, which is why every ACA1221lc knows not only about it's own Warranty ID, but also the CPU mask that it was shipped with. So far, it turned out that all masks are working fine at 40MHz if the power supply has proper regulation, so the ACAtool displaying the mask is merely informational, as that's covered by the heat sink.

Running a memory test program isn't equivalent to a real-world test, but it will of course test if memory access is working correctly.

Please note that you're not testing the memory itself if you're overclocking. The 80MHz pace is a "relaxed sunday walk" for the 166MHz-rated SD-Ram, and the local 3.3V regulation will provide much lower ripple to the 3.3V parts of the card than the power supply produces. Before this fails, a lot of other things have failed.

What you are testing when overclocking is the CPU's ability to take over and provide data at the right time. This timing shifts with remperature and 5V power supply level. So your goal is not to give the memory a workout, as that's safely running at less than 50% of it's rated speed. You want to give the CPU a workout, so it heats up as much as possible - this in turn will break communication with memory, if the CPU shifts timing too far.

Jens