Introducing the GIME chip

The GIME chip within the Color Computer 3 is an incredible piece of work. Not only does it almost completely emulate the functions of the 6847 VDG and 6883 SAM chip architecture of the earlier Color Computers but it adds a lot more! Advanced bank switching of RAM up to 512K, advanced interrupt handling and many more sophisticated graphics modes.

The Color Computer 3 with its GIME chip has expanded upon the earlier models with their more limited 128x192 four color and 256x192 two color modes. The GIME chip has added new modes of 16 colors with a resolution up to 320x225 and 2 color modes with a resolution up to 640x225, the colors being selectable from a total palette of 64.

An interesting observation!

In 1999, I pointed John Kowalski's attention to the block diagram of the GIME chip on page 96 of the Tandy Color Computer 3 service manual. John made an interesting observation that got us both curious. The diagram shows that the output of the "Reconfigurable (Graphics) Shift Register" is fed through a gate, followed by a "Register" and then on to the "Palette 16x6 Bipolar RAM", finally ending up at another register then on to the DAC for output. This seems sound and as one would expect. What got John suspicious was that the output of the "Reconfigurable (Graphics) Shift Register" ALSO routed to another unlabeled "Register" which appeared to bypass the "Palette 16x6 Bipolar RAM" and fed in the register that follows it. John came with the suspicion that this register allowed 8 bit data from the shift register directly past the 64 color palette RAM!!

We both puzzled over this but and no one had any ideas of what it could be. Suddenly I came into contact with a very prominant ex-radio shack employee who had a very close association with the development of the Color Computer 3 via one of the CoCo IRC chat sessions. To say that his question to me..."Have you found the 256 color mode?"...raised the hairs on my back was a gross understatement!!

"Have you found the 256 color mode?"

Here is an edited transcript of what he had to say about this mode...

Have you found the 256 color mode?

There is a real 256 color mode in there. I was hoping someone would discover it. It's a real byte level pixel, 320x200 mode. It uses a yyyyyrgb format, 5 bits of intensity, 3 bits of color.

The mode is rather complicted to get to and did not work reliably on the first run of GIME chips and its been too long, I don't remember the sequence anymore and I got overruled by Roach on the disclosure of the mode because it was too much competition with the 1000. It's really the last secret in the Color Computer 3.

You can only have one 256 color page but it was very complicated to set up. It could view 512K and that was one of the problems because of the page at the top of memory. I was not allowed to tell anyone exactly what to look for either.

The only thing I remember about how to get into it is that you have to switch the mode while you are in the IRQ in the protected page at $FE00. It had to be set as a mode on the IRQ, it didn't use the IRQ. While in that routine from an IRQ there is a machine state that changes the meaning of the graphics mode. I don't have a clue what that pattern is anymore.

John Prickett ran out of pins to make it work so it reused some of them in while in the IRQ state.

I, along with John Kowalski, was able to get in touch with this person for a second time to get a few more questions in and here is the edited transcript ...

[MR X] It is a machine state, so it does require the right set of events and I am probably the only one who ever turned it on and played with it. My demo program put up changing color patterns. It was designed to be for games.

[JOHN] So, you're saying that an IRQ routine that 'itself' is located in the $FEOO-$FEFF page is used to switch on the mode?

[MR X] That's correct, John and as I remember it's the only place you can do it.

[JOHN] Any hints as to what generates the IRQ signal? Does it matter?

[MR X] I did it on the timer I think.

[NICK] Is the 256 mode visible in both RGB and composite video output?

[MR X] I believe so, its a real mode.

[JOHN] So if I set up a TIMER based IRQ that jumps to the $FExx range... If I systematically trash random hardware registers $FFOO to $FFFF, eventually the mode should turn on.. shouldn't it?

[MR X] It sees to me some of the regs had to be written with a given interval so random data is not going to do it and I can tell you the routine is not in Tandy archives anymore either.

[JOHN] What would be the reason for the program code to be in $FExx range? It's not like the GIME can tell where the CPU's PC is.

[MR X] That's not true John, there are times the GIME does know. AII the coco files were distroyed 2 years before I left Tandy. No, once in the mode the computer acts as it did before, but no way to exit the video mode without a reset because the MMU is part of the addressing so the MMU has to intercept the address.

[JOHN] So.. the GIME remembers the last address issued on the CPU bus, and if that address is not exactly 'some number', the mode will not activate?

[MR X] The special page is treated differntly by the MMU and the memory map becomes unstable during the mode change so it has to be one in that page system needs to be in the RAM mode.

[NICK] MMU on or off?

[MR X] On.

[NICK] CoCo1/2 compatible mode on or off?

[MR X] CoCo 3 mode.

GIME chip designer found!

My next avenue of pursuit was to locate the man responsible for the design of the GIME chip and I was given the name, John Prickett. My search on the internet located a John Prickett working for AMD at the time and here is his response to my question...

I'm John Prickett, and I did design the chip you referenced, but that was 14 years and 50 designs ago. I couldn't possibly tell you any of the internal intricacies of the chip now. You probably know more about it that I do right now.

I do seem to remember that it was 64 colors only -- we only used a 6-bit palette, and so I very much doubt that there is a 'hidden 256 color' mode. But I could be wrong.

Sorry I couldn't be more help, John

John appears to have moved on and I have lost contact with him.

More Proof!

During my job as PennFest 2000 co-organiser, I came into contact with Mark Hawkins of Microware. Mark is one of the 3 people who developed the CoCo3's Super Extended Basic and is one of the "Three Mugsters" that appears on the CTRL-ALT-RESET page. I asked Mark about this mode and he was unfamiliar with it but he said that he would look through the archives at Microware for old documentation about the GIME chip. He found the original "Color Computer Custom Video Proposed Feature List" document which came from Tandy's R&D section. All of the info is as found in the service manual except for the cover page which lists the 256 color mode!

This proves that the mode was at least in the planning stages. What we are unsure of  from this page is what the references are about 1 and 2 banks of RAM and how that effects the modes. We have seen references to this in CoCo3 memory maps but it has always been ignored.

With closer scrutiny of this R&D document, we find only one additional piece of information that is not to be found in the normal CoCo3 memory maps. On page 7 of this document that covers the CoCo1/2 compatible registers of $FF20 to $FF23, we find handwritten at the bottom of the page the following...

Could this be one of the mystery registers or was it an abandoned idea?

Infinite Scanlines Mode

Another interesting observation that has me curious (but may not be related to the 256 mode) is the function of BITS 6 and 7 in the FF99 Video Resolution Register. The functions are listed in the CoCo3 memory maps as...

The Lines Per Field is the number of vertical scanlines or vertical resolution of the mode. All of these work fine except for the 210 setting. This does not create 210 lines. Instead it appears to create an infinite number of lines. With this setting, the display stretches past the top and bottom and it seems to pick whatever color the GIME was processing at the time the mode is set. If it was in the border region of the screen, it keeps generating this color otherwise it repeats the current color in the active portion of the screen.

Current Status

John and I have run tests on the various GIME chip locations, running random data and experimenting with all the areas of the map that are marked as undefined  or not used. The mode, if it truly exists, does not take the form of a normal register function. As our friend on the IRC chat explained, the designers ran out of pins to make it work properly and had to re-use pins during a specific machine state of the GIME. This makes it extremely difficult to locate because even if we are looking at the right places to setup the mode, the mode is non-functional and therefore will not present itself unless this specific state is met.

We need more information and vital clues in order to proceed further. What are those 2 registers that appear to be part of some bypass of 8 bit video data in the GIME chip and how are they accessed? Could the mode have been removed from final production GIME chips or does the mode still exist but has been carefully "hidden" under the request of Tandy management so as not to conflict with the sales of the Tandy 1000 line of PC compatibles in the product line at the time.