99/4A accelerator

[mnbvcxz]

Many other computers have had accelerators and processor upgrades designed for them, would there be any interest for one in the TI 99/4a community?, I do not have the skills to design it myself (although that won't stop me from trying to do it one day).

 

I thought about having an accelerator cartridge that plugged into the cartridge port, it would contain a 9995 based single board computer with 1MB static ram (2 chips), maybe a CF or SD socket, and it would use the 9900 in the console, as an in/out processor, moving data between the accelerator cartridge and the devices attached to the 99/4A such as the keyboard, screen and disk drive Etc.

 

Programs running on the accelerator would not be affected by the crippled architecture of the 99/4A, it might even be possible to copy GPL and a modified BASIC into the cartridge, although that might slow things down too much, but a simple monitor program will allow programs to access the 99/4a hardware via the 9900, there would also need to be code that tells the 9900 how to be the I/O processor.

Edited February 5, 2012 by mnbvcxz

[rocky007]

yes...it's called Geneve or nearly...

[mnbvcxz]

Geneve is not designed to plug into the cartridge port, but needs an expensive and in the UK, unobtainable PEB, the geneve cannot be bought now and the geneve 2 is not yet available.

The Console on its own is not very useful, but an accelerator like this would give it a whole new lease on life.

[rocky007]

the problem with this type of extension is the lack of program developpement.

 

take the example of supercpu: it was really an incredible & powerful interface, but it was distributed in very small quantities. finally, i can count on my hand the number of games exploiting it. A developer prefer reach a maximum users rather than a minority.

 

there are also purists who like to develop on a computer with the original limitations

[+Vorticon]

In the mean time, how about upgrading the crystal and placing the 32K RAM expansion on the 16bit bus? This combination will give you about a 54% performance boost from my testing:

 

8bit RAM and stock crystal: 39s (baseline)

16bit RAM and stock crystal: 21s (46% faster)

14.3 MHz crystal + 8bit RAM: 32s (18% faster)

14.3 MHz crystal + 16bit RAM: 18s (54% faster)

 

This is where you can find the detailed instructions:

http://www.mainbyte.com/ti99/upgrades.asp

[moulinaie]

In the mean time, how about upgrading the crystal and placing the 32K RAM expansion on the 16bit bus? This combination will give you about a 54% performance boost from my testing:This is where you can find the detailed instructions:

 

Yes, that's an interesting thing. What language did you use to make the benchs?

 

Guillaume.

[+OLD CS1]

I am pretty certain you cannot run the core system from the cartridge port as it does not present all the control lines which would be necessary to do so. You would have to use the side expansion port.

[mnbvcxz]

I am pretty certain you cannot run the core system from the cartridge port as it does not present all the control lines which would be necessary to do so. You would have to use the side expansion port.

 

It does not need them, the accelerator acts like a cartridge passing instructions and data to the 9900, whilst the application program is actually executed by a 9995 in the cartridge, the 9900 in the console controls all the hardware attached to the 99/4a and moves data to/from the cartridge as dictated by the program running on the 9995. The application program will be loaded into the cartridge by the 9900 from CF, disk or cassette, so a small rom is needed in the cartridge to provide simple BIOS functions.

[+Vorticon]

In the mean time, how about upgrading the crystal and placing the 32K RAM expansion on the 16bit bus? This combination will give you about a 54% performance boost from my testing:This is where you can find the detailed instructions:

 

Yes, that's an interesting thing. What language did you use to make the benchs?

 

Guillaume.

I used straight assembly language and created a very short and simple timing loop to compare results.

[moulinaie]

I used straight assembly language and created a very short and simple timing loop to compare results.

 

that's interesting. But I read some time ago the page for the upgrade and found it hard to do. I only changed the crystal.

 

Do you know if the 32K can be used with the CF7?

 

Guillaume.

[Stuart]

Do you know if the 32K can be used with the CF7?

 

Yes, but you'll need the version of the CF7 which doesn't have the 32K on-board. Or be inventive and mod your CF7 to disable the 32K chip.

[+OLD CS1]

I am pretty certain you cannot run the core system from the cartridge port as it does not present all the control lines which would be necessary to do so. You would have to use the side expansion port.

 

It does not need them, the accelerator acts like a cartridge passing instructions and data to the 9900, whilst the application program is actually executed by a 9995 in the cartridge, the 9900 in the console controls all the hardware attached to the 99/4a and moves data to/from the cartridge as dictated by the program running on the 9995. The application program will be loaded into the cartridge by the 9900 from CF, disk or cassette, so a small rom is needed in the cartridge to provide simple BIOS functions.

 

Interesting. I suppose you could turn the 9900 into a simple job processor, handling hardware jobs for the 9995 when needed -- akin to how the Commodore 128 runs in CP/M with the Z80 as the primary processor and the 8502 running the hardware interface. You have memory space to make common as a way to pass structures to the 9900 as a job processor to handle DSRs and what-not, even execute console ROM/GROM routines, but direct access to hardware, such as the 32k memory expansion and VDP, would not be possible. And anything in the console would be executed at the 9900's speed. Though, I suppose you could copy the entire system over to the 9995 space during boot-up. But, again, hardware access would have to be mediated by the 9900.

 

Do you have any specs worked up for this idea? It sounds pretty complex. Might make for a fun project, though I do not see it gaining wide acceptance due to the complexity and inherent incompatibilities -- at least at first. When speaking of accelerators on other platforms, the new processor usually completely takes over the system rather than act as a coprocessor.

 

The only exception I know of are the PowerPC accelerators for the Amiga, which require the 68000 processor to run the base operating system. There is a good write-up at Amiga.org which talks about the slowness inherent to the 68k/PPC coprocessing design, particularly dealing with context changes moving between the two processors, and how this was mostly overcome with later systems. I believe OS4 Classic runs completely on the PPC, thus eliminating the 680x0 altogether, but it again has full access to all system hardware.

[mnbvcxz]

I do not have the skills and knowledge to design this yet, which is why I started this thread, in the hope that others could do it, but I thought that the powertran cortex circuit could be adapted for this, I am very new to the 99/4a and do not yet understand the complexities of the operating system, I only have a console so I cannot be sure that anything I come up with will work with anything else that might be attached, this must be a community project to ensure it works with any hardware.

 

The limiting factor in this idea is the availability of the 9995 processor which are getting difficult to find or expensive.

[Willsy]

You'd probably find that it would be simpler to simply add a 9995 directly to the consol in place of the 9900.

 

Colin Hinson, an employee of TI UK did this very mod to his console, and it worked. If I remember, it was a three chip solution. Physically, it was a board that plugged into the board where the 9900 used to be. If I remember correctly, the add-on board contained a crystal, a RAM chip (to replace the 256 bytes scratch-pad ram - the 9995 has 256 bytes of ram built-in, but it's not in the >83xx range that the console ROMs and GROMs expect) and some programmable logic.

 

I remember Colin explaining to me that it was actually quite a simple mod, since the mother-board is essentially engineered for an 8-bit world, with a crippled 16-bit chip shoe-horned into it. Fitting the 9995, which is around twice as fast as the 9900 makes life somewhat simpler, since it has an 8-bit data-bus.

 

Thus (I imagine) circuits such as the 8 to 16 bit mux/de-mux, which occupies a sizeable portion of the discrete logic on the mother board simply get disabled. The wait-state generator too...

[eck]

OPA announced a TMS99xxx processor plug in for the console in MicroPendium with 12 MHz operation speed. As far as I remember the board should be clipped just over the 9900. I ordered a TIM and hoped to buy these device too, but the TIM was not shipped to me and the 'Speedboard' was never released. A few days back I was looking for the SBP9989NL (4.4MHz, 64 pin, 16 bit data bus, 9995 instruction set, 128K ram addressable, ... ) on the internet, if it might be a replacement for the TMS9900 in the console, but the price is exorbitant. So this is cancelled. There was also an advertisement in MicroPendium issues about a p-box card with a forth microprocessor on it, which promised a lot of speed. Sad to say that nobody ordered this card.

[GroovyBee]

In my mind if you are going to put a chip in a cart as a "co-processor" you might as well make it an ARM. They are easily available and have free programming tools.

[+RXB]

I had the TIM and it was great. Made the TI console alone and a supercart loaded with patched FW do 80 col in Editor. Of course you had to have the 32K in the console also.

 

As for the Forth card are you talking about the one made here in NW of the USA that was the fastest TI Card I ever saw?

 

It had a dedicated Forth Processor on the card so when I saw the demo it was insane how fast it was.

[humeur]

OPA announced a TMS99xxx processor plug in for the console in MicroPendium with 12 MHz operation speed. As far as I remember the board should be clipped just over the 9900. I ordered a TIM and hoped to buy these device too, but the TIM was not shipped to me and the 'Speedboard' was never released. A few days back I was looking for the SBP9989NL (4.4MHz, 64 pin, 16 bit data bus, 9995 instruction set, 128K ram addressable, ... ) on the internet, if it might be a replacement for the TMS9900 in the console, but the price is exorbitant. So this is cancelled. There was also an advertisement in MicroPendium issues about a p-box card with a forth microprocessor on it, which promised a lot of speed. Sad to say that nobody ordered this card.

 

the SBP9989 is I2L technologie is not a same of the tms9900.

 

jl

[Tursi]

I had put some time into the same idea as you, mnbvcxz -- in fact that concept developed into my Linux expansion board (which, if I do it now, won't include the accelerator part anyway, but it was the whole idea of a co-processor).

 

It's pretty much the only way to do it without mods. It would have been nice if we could plug in a side-car expansion that replaces the 9900 but has access to the bus, but a motherboard modification would be necessary to do it, since the 9900's DMA line (I can't remember off the top of my head what it's actually called) is hard wired, and so the 9900 already drives the bus. Of course, even there the modification should be simple - rerouting a single pin. (The accelerator itself would be more complicated though.)

 

Turning the 9900 into nothing but essentially a server still works, though.

[JamesD]

The Apple II had the Zip Chip to speed it up. The Zip Chip plugged in to the CPU socked and used cache memory and a higher clock speed controllable via POKE commands to speed it up. A similar approach with the TI would probably be better than something in the cartridge port.

[unhuman]

IIRC there was some project to replace the crystal in the TI to make it faster...

[Tursi]

The Apple II had the Zip Chip to speed it up. The Zip Chip plugged in to the CPU socked and used cache memory and a higher clock speed controllable via POKE commands to speed it up. A similar approach with the TI would probably be better than something in the cartridge port.

 

Unfortunately, the CPU is not socketted, so it can't be trivally replaced. Most of the chips on the TI mainboard are soldered in.

[JamesD]

The Apple II had the Zip Chip to speed it up. The Zip Chip plugged in to the CPU socked and used cache memory and a higher clock speed controllable via POKE commands to speed it up. A similar approach with the TI would probably be better than something in the cartridge port.

 

Unfortunately, the CPU is not socketted, so it can't be trivally replaced. Most of the chips on the TI mainboard are soldered in.

That would pose a problem.

[humeur]

OPA announced a TMS99xxx processor plug in for the console in MicroPendium with 12 MHz operation speed. As far as I remember the board should be clipped just over the 9900. I ordered a TIM and hoped to buy these device too, but the TIM was not shipped to me and the 'Speedboard' was never released. A few days back I was looking for the SBP9989NL (4.4MHz, 64 pin, 16 bit data bus, 9995 instruction set, 128K ram addressable, ... ) on the internet, if it might be a replacement for the TMS9900 in the console, but the price is exorbitant. So this is cancelled. There was also an advertisement in MicroPendium issues about a p-box card with a forth microprocessor on it, which promised a lot of speed. Sad to say that nobody ordered this card.

 

i have a SBP9989 a picture of this

 

[Willsy]

How does the 9989 address 128K? Is there some sort of internal page register? Does it have any user/supervisor mode type exception traps/vectors?