Today, at school, I have been transferring my OverClui overclocking utility to people who requested it. Looks like not all Prizms will overclock to 94.3 MHz without completely locking up or return a system error.

I had transferred Pover, and later, the more user-friendly utility OverClui, to other people before. On their calcs, the CPU frequency could go as high as 94.3 MHz without hanging (and return a system error or lock when going as up as 101.5 MHz, unless USB power was connected).
I made my tool so that it wouldn't let go as high as 101.5. It lets go up to 94.3 since this was considered safe.
Today, I transferred OverClui to other people. On two out of the three calculators I transferred to, they only go as high as 87 MHz. When USB is connected, overclocking to 94.3 MHz is possible (but when USB power is disconnected, the calc stops working and a reboot is necessary).

At first I thought it had something to do with battery levels and capacity, even though all batteries were alkaline. I confirmed this doesn't seem to be the case: I switched the batteries with mine, and it still didn't work (their batteries in my calc made it work fine at 94.3 MHz too).

About OS version: one of the calcs where overclocking didn't work, was running OS 1.02 and the other OS 1.04. Both are fx-CG 20.

It's not OverClui either, as the problem is reproducible with Gravity Duck too. I didn't have Pover in my calc to test with it

I guess I could run Insight on these calculators as well as take note of serial numbers and etc., but I don't want to annoy other people with my technical curiosity In fact, I must thank a friend for letting me run into system errors and rebooting her Prizm for over 5 times in an hour...

Why this matters to Prizm add-in creators
(and regular users that install certain software on their Prizms)
There is software like the latest version of Gravity Duck that automatically overclocks to 94.3 MHz when starting. The problem is that this will render the game unplayable to some people (and scare users at the "system error" message).
So, like I did before, I suggest again add-ins do not overclock without users' consent. Not everyone is able to overclock to the same speeds, and having certain software overclock without permission renders that software unusable on certain calculators.

This may also be the reason why Casio made the CPU run at 58 MHz: perhaps they know certain CPUs won't go higher than that safely, unless the calc is being powered though USB.

Looking for more insights and discussion on this topic

I completely agree that programs shouldn't be overclocking without the user's permission. I even, in fact, think that overclocking should be left to the specific overclocking programs. Other programs should at most have a message at the start that tells the user they should overclock, and put something in the readme the explains how (maybe even bundle OverClui/Plover with their release, assuming the creators are OK with that).

Or they should leave the choice to the user I guess. Like "Overclock? Y/N"

I think even that shouldn't be in there. Users are dumb, they'll just hit "Y" without thinking about it. On the other hand, if they're required to go to a different program, they may be more likely to think it through. Also, when it then fails, they won't go to the author and say "YOUR PROGRAM SUX I H8 U", they'll (hopefully) be more aware that it was the overclocking not working correctly with their calc.

Does overclocking to 87mhz work on all calcs?

I'm totally with merth. I found some people don't like the idea of having their calculators overclocked, not to mention the fact it consumes much more battery, so having software overclock without consent is not good.

Software can ask if users want to overclock, but I think it must leave very clear there is no warranty and that it may result in system errors, random lockups, data or physical damage, well, that users are on their own. Perhaps with a countdown for the Yes button like certain browsers do before installing add-ins?

flyinhfisch, I'm not totally sure. Right now, I don't recall if both problematic calcs worked at 87 MHz or if there was one that stopped working at 72.5 MHz. I'll try to check when I can.

Until we're able to test on every single calculator, I think we won't really be able to say. Hopefully we'll be able to get closer to discovering the safest speed, but I feel Casio might've already done that, and that's why they settled on the speed they did.

Well, once we test it more, I think that overclocking in-program is actually a good thing. Some games need the extra speed to function well, and if users don't overclock, the game will be much easier (or even unbearably slow) to play. Even now it's technically "safe", it sounds like it just ensues harmless resets or exceptions to occur. none of that should happen at 87MHz as for what we've tested so far.

The problem is, most people will prefer to play a game like Gravity Duck slower, than not playing it at all. Also, at least until more information is known, there's no safe way to know to which speed a specific Prizm can overclock safely.

Also, the exceptions are not harmless - you can't expect everyone to have your skill level and know one should press EXIT on the error screen, or press the reset button on the back when the calculator locks up (even because I've noticed many people think pressing the reset hole makes them lose all the stuff, i.e. reset the flash memory). Of course, you'll say "then don't tell noobs to overclock!". But when any add-in does overclock without warning or confirmation, how will the user know he shouldn't overclock?
For example, and sorry for using Gravity Duck once again, let's say I transfer it to someone and that I am not aware of this problem (in fact, it happened today). The other person, when launching the game, will see a system error message (or a lock up), and act like "OMG OMG this is a virus!", then go tell everyone the add-in in question is harmful.

I don't mind if my calc gets overclocked at two conditions:

1) Once the program exits, my calculator reverts back to its original speed (or the one I chose) before running it
2) It's safe

That said I hope Casio didn't do something like the new TI-84 Plus hardware, on which old versions of TI-Boy SE, MSD8x, Omnicalc restoremem and RealSound won't run.

I think it has something to do with hardware revisions, perhaps they used a slightly different power scheme for the CPU on some revision, which results in more power being available to the CPU on certain calcs (those that can overclock to 94.3 MHz). This is my bet, because from what I have seen, all calcs can go as high as 101.5 MHz as long as they have USB power, but stop working if USB is not present (some stop working at 94.3 and others at 101.5). It either has something to do with the amount of power, or to do with from where the power comes from.

If it is the bootloader that sets the initial CPU clocks... can it have something to do with how the bootloader sets the initial values after a reboot (those values that Pover lets you adjust and Overclui does not)?

If the cause of those is the CPU behaving erratically, that is not safe - although unlikely, it could end up writing to flash and corrupting the OS.

It's not the CPU itself from what I've dug into, it's the peripherals (such as memory and USB) that are the problem with the overclocking. [I'll explain more when I come back in an hour or so]

Hmm... that's odd. Is there any hardware version difference?

And yes @Ashbad, peripherals are terrible. For example from Renesas, the CPU can hit 500MHz whereas the peripherals are stuck at ~80MHz. And if you don't make sure that the peripherals are 100% happy, the hardware fails.

@merth: I agree, the programmer shouldn't assume that the calc can and should be overclocked. Personally, I would recommend doing what GlassOS has done. Make a configuration file in Main memory (not storage) that will tell programs what frequency to go to at max, and whether they must ask the user, go ahead, or not overclock at all. The config file will be adjusted only from overclocking addins, like Pover and OverClui. (GlassOS uses configs for greyscale and uses the same model stated above, and it works just fine. User-friendly)

We could investigate a bit more, because on Cemetech and Omnimaga, several people are using CG-10 and on Planet-Casio several people are using CG-20.
Since the first version of Pover, in any case the calc returns a system error when it works at 94.3 MHz.

I understand you don't want to annoy your friend with test, but we should research why it doesn't work in some cases.

Such as grabbing the serial number? Does the prizm provide any other info?

If the serial number is the barcode on the back of the calculator, then mine is almost completely blanked out If this is the case, then I guess people won't get upset with me if I ask to see the serial numbers of the problematic calculators.

One thing I know: one of the two calculators that didn't work well was ordered from the internet (from the FNAC website I think, so it may come from France). I don't know about the other one but I suppose it was bought in the same store as mine, in Portugal.

I've tried overclocking with +/- 10 fx-cg20 calculators and all worked (os 1.03 and 1.04) until 94.3 MHz

Today I tried with yet another calculator, and it worked fine at 94.3 MHz too. I think I have tested OverClui on at least seven fx-CG20, and two shown the problem of not being able to go up to 94.3 MHz, only 87 MHz.

2 possibilities:

1-exposure calculator to heat, humidity, constant movements (when transporting...), pressure, may reduce the vitality of the internal circuits

2-as some guys say, this is a lot of calculators with small differences in hardware (maybe different circuits)...