Hello,
in my mission to build 50 Apple-1 clones by its 50th anniversary, I have done some testing to see what's needed to create clones that will run solid. So, obviously, the first approach is applying Uncle Bernie's "Reliability Mod" as described here.
All of the 50 builds will use the same type of DRAM ("Intersil") which threw a lot of DRAM error running Bernie's RAM test in my test board with no mods attached. So I first soldered in the caps on the back of the 74S257 to see if that changed something which it did. Far less DRAM errors but still too many to have a reliable machine. Basic would not even startup most of the time.
I asked Bernie what the lowest hanging fruit would be to achieve a reliable board and the mentioned to that adding the damping resistors would be a good start.
And that's what I did – just adding the six resistors of Bernie's mod (removing the caps I had already sodered in).
Result: The board has been running the PROM RAM test for a few hours now with no error at all. What a difference!
Best part of this: As it stands now, I can even use 74LS257 instead of the more expensive and harder to get 74S257. In fact, the machine running the test now for a few hours now, uses four 74LS257.
Now: By no means do I suggest that just adding the six resistors (and not ignoring all the caps) will do the trick for everybody, but as I will build many more board I will have collected enough data to (hopefully) make a strong case for an easier mod.
I will also test this "shortcut" with different RAMs (Intersil, Mostek, Fairchild) to see whether I get any different results.
Fingers crossed!
Best
Armin
Hi Armin -
as much as I appreciate your work in stripping my "reliability mods" down to the bare minimum, keep in mind four facts:
1. All DRAM manufacturers recommended *plenty* of bypass capacitors in their updated datasheets (after the first datasheets tried to hide the current spikes).
2. Every type of DRAM may have a different sensitivity to these spikes on its supply voltages, to the ground bounce, and to the ringing address and control lines. Even different lots of the same DRAM from the same manufacturer may be different, as the process moves under process controls.
3. Differences in supply voltages and the ambient temperature can shift the logic level thresholds quite a bit and this can make the quirks of the original Apple-1 disappear or reappear.
4. At the time I developed the "reliability mods" I could not find DS0025 and had to ship my kits with the more aggressive DS0026, so a few more added bypass capacitors were needed in and around the terminal section. Again, if those are really needed depends also on the type and lot of the 1404 shift registers needed, which are the accomplice in the crime, as they couple the fiercer current spikes from the faster DS0026 into their -5V supply lines ... which may affect all ICs that also tap into the -5V supply (like the DRAMs).
I still recommend "my" builders to implement all the reliability mods, because they are field proven and always have lead to a 100% success rate, meaning no DRAM errors. I know that some builders left them out and their build still works. Some builders have only put in every other added bypass capacitor on the DRAM banks.
I get these emails all the time asking: "look I did this, I did that, I left out some parts, is it safe ?". How can I know ? Same thing as with a car mechanic having some seemingly superflous parts left after the repair calling the factory asking if it's safe ! As far as I'm concerned I'm not engaging in such speculations.
Here is an allegory for the situation with the Apple-1:
Imagine a man standing at a steep cliff, just on the edge. This man may be perfectly safe unless some external factors destabilize the situation (wind, rain, earthquake). Now bus 150 men to that cliff and position them there. How many will fall into the abyss ? These are the ~150 Apple-1 the two Steves did build. Individual tolerances caused some to crash their programs occasionally, while others worked OK. Now add Uncle Bernie's "reliability mods". The 150 men will take one step back, away from the cliff. The overall situation will be much safer. But still, one or two may fall under certain individual tolerances / dispositions. Now, somebody suggests that instead of taking on full step away from the cliff, three inches should be enough (less "reliability mod" components to solder in). It is safe ? Will it work ? Who knows !
My builders can do their experiments with the "reliability mods" because I provide the extra diagnostics page in my A1, A2 PROMs, which includes a powerful DRAM test based on the special algorithm the industry used back in the day to test these DRAMs. But will this algorithm catch all the subtle and random errors brought about by the hostile environment of the Apple-1 with all its bouncing ground (actually, "solid ground" is nowhere to be found in this machine), the polluted supply lines, and the ringing DRAM multiplexed bus and control lines ? Probably not. The "MadHammer" test program I published here on Applefritter does a better job to expose the effects of the ringing multiplexed bus lines after a shorter run time, but of course, with a long enough run the DRAM test from the diagnostics page will also trigger such an error event, sooner or later. Note that "MadHammer" is not a complete DRAM test - it is specialized to provoke errors caused by the ringing multiplexed address lines. I have to mention this because some people think it's a general DRAM test. Which it is not. But it provokes the ringing many thousands times more often than a general DRAM test can do in the same run time, and so has a better chance to reveal the problem.
The Apple-1 rabbit hole runs deep. Don't dig it any deeper ! Of course, any builder of the Apple-1 has the right to build his or her machine as seen fit, but if you deviate from the path to success your Uncle Bernie has shown you, and it does not work, or is too wonky, don't whine and ask me to fix it for you !
Comments invited !