TI-99/4a - How critical is RAM access time?

[jmetal88]

I'm down to what's pretty much my last-ditch effort in repairing my TI-99/4a - replacing the eight 4116 RAM chips.

The RAM I've removed is clearly marked at 150ns, and I've found replacements sold for as low as 99 cents per chip (without a minimum order).

However, I've found an equivalent chip marked at 200ns access time, but rated for the same length of RW/RMW cycle for only 14 cents per chip.

That said, just how critical is the address access time for RAM in the TI-99/4a? Would I be able to get away with ordering the 200ns chips since a read-write/read-modify-write cycle is the same length as on the 150ns chips, or would the longer address access time throw off other operations?

 

[wmmullaney]

It would most likely be fine, since the processor is only running at several mhz. But for only four, the extra pennies might be the best choice, just to make sure and for authenticity.

 

[jmetal88]

It would most likely be fine, since the processor is only running at several mhz. But for only four, the extra pennies might be the best choice, just to make sure and for authenticity.

Agh, I said four, didn't I? I meant eight, I'm replacing all eight.

Yeah, if it were four I probably wouldn't be asking, but for eight, there's a significant difference, especially with $7 shipping figured in for each order.

 

[Chuck(G)]

Jameco sells them for $0.99 each for the 150nsec parts, if it's any help.

 

[jmetal88]

Already found those, just trying to save a few bucks.

 

[wmmullaney]

The ti 99 is doing a memory access a max of once every 333ns. That's a close call, but probably safe if you ask me.

 

[jmetal88]

The ti 99 is doing a memory access a max of once every 333ns. That's a close call, but probably safe if you ask me.

Hmm, that's interesting. Not in itself, but in the fact that the TI-installed part is rated for a RW cycle of 375ns. That is, unless TI's datasheet is wrong, and it actually has a 320ns RW cycle, like what other brands of 16k 150ns chips list in their datasheets.

 

[wmmullaney]

I'm not familiar with the CPU, but at 3mhz, it should be going to the RAM once every 1/3us executing continual NOPs. Again, I know nothing of this computer, just educated guesses.

 

[jmetal88]

Eh, the more I think about it, the less I want to use 4116 chips. I was looking at the TI motherboard last night, and it looks like if I remove two components and add one jumper, the board would be 4164 compatible, and from what I've heard, 4164s are much more reliable than 4116s. Cheapest I've found 4164s is about 86 cents though, so I may end up just going with Jameco on those so I can get some extra 16-pin DIP sockets at the same time.

EDIT: And before anyone says anything, I already know the added 48k would be inaccessible.

 

[Chuck(G)]

One benefit is that the 4164s are single-supply. I don't know if this simplifies things for you or not. It would be interesting to be able to bank-switch the entire DRAM, however.

 

[geoffm3]

Probably wouldn't be of much use bank switching, since the RAM in question is directly accessed by the VDP only. The processor on this machine can only access the RAM through IO port read/writes from the VDP. TI really hindered the machine a lot with the memory design... the only directly accessible RAM is either 256 or 512 bytes (that's right, BYTES) of memory, since the 9900 has external registers, it needs some physical memory. And then of course only connecting 8-bits of the 16-bit wide databus was another problem. I guess they had too much development time invested in the 99/4 software to deviate once memory started coming down in price.

 

[Chuck(G)]

I understand that one of the driving forces behind the silly 99/4A design was that it should not compete with the 990 mini, in particular with the 990/M, which used the same TMS9900 CPU. With the 99/4A designed the way it is, there's no chance of being able to run 990 code on it.

 

[geoffm3]

I understand that one of the driving forces behind the silly 99/4A design was that it should not compete with the 990 mini, in particular with the 990/M, which used the same TMS9900 CPU. With the 99/4A designed the way it is, there's no chance of being able to run 990 code on it.

Seems like I've heard that before. I've also heard it said that, when you get calculator designers who are used to minimalistic designs to design a general purpose computer, the 99/4(a) is what you get.

Don't get me wrong, I had a lot of fun with ours back in the day, and I even learned to program in BASIC on it. That said it boggles the mind how expensive they were compared to other home computers at the time given the design tradeoffs they made and the fact that they were so vertically integrated (practically every IC in the thing is stamped TI). Commodore beat them at their own game obviously, but it's hard for me to believe the anecdotes that they were too expensive to make money with (PEB arguments aside).

 

[Ksarul]

Actually, I was recently talking to one of the TI engineers for the 4A system. He had a lower cost variant of the machine designed and tested in 1982, using a 9995 CPU and a gate array to significantly lower parts counts. It never made it into production because of the long-term part ordering strategies followed by management--they had already calculated so many sales of each part into their calculations for quarterly revenue for the year, that it made any change disastrous. A lot of innovative items never made it out of the labs and into production because of that. I have a TI-made 128K card for the 4A, an IEEE-488 interface, an EPROM burner for the PEB, a FORTI music card, and a TI DSDD Disk controller card. I've also seen a 1200 baud modem card and a Video controller card (this latter was used for computer aided instruction and would control multimedia presentations using a laser disk or a VCR).

 

[geoffm3]

Actually, I was recently talking to one of the TI engineers for the 4A system. He had a lower cost variant of the machine designed and tested in 1982, using a 9995 CPU and a gate array to significantly lower parts counts. It never made it into production because of the long-term part ordering strategies followed by management--they had already calculated so many sales of each part into their calculations for quarterly revenue for the year, that it made any change disastrous. A lot of innovative items never made it out of the labs and into production because of that. I have a TI-made 128K card for the 4A, an IEEE-488 interface, an EPROM burner for the PEB, a FORTI music card, and a TI DSDD Disk controller card. I've also seen a 1200 baud modem card and a Video controller card (this latter was used for computer aided instruction and would control multimedia presentations using a laser disk or a VCR).

Very interesting. I really liked playing around with the 99/4A when I was a kid. I had a complete system with PEB and a couple disk drives, a 128k RAM expander (the manufacturer I couldn't tell you, but it had a ramdisk feature on it), and the RS-232 interface. The PEB is one of the things I really liked about the system, but it's probably also one of the reasons that a fully populated system was prohibitively expensive.

 

[Chuck(G)]

Were there any third-party systems made with the TMS9900 where memory is treated in a more conventional way (other than the TI 990/4 and later systems)? For example, were there any S100 or ISA cards made with the TMS9900?

Does anyone have any equipment made with the SBP9900--the I²L version of the TMS900? I can remember asking a TI Marketing guy about getting a sample at a Wescon sometime around 1979 and essentially being told "in your dreams...". I guess it was mostly the aerospace and defense crowd who got them...

 

[Dwight Elvey]

One benefit is that the 4164s are single-supply. I don't know if this simplifies things for you or not. It would be interesting to be able to bank-switch the entire DRAM, however.

Hi
Do be careful on the 4164s. Some of the early ones from ( I think )
samsung had 256 cycle refresh instead of 128.
I got stung with these on an Z80 project I was working on.
Since there were so many, I had to make the code do the
refresh for me. What a pain.
Dwight

 

[JGardner]

Actually, I was recently talking to one of the TI engineers for the 4A system. He had a lower cost variant of the machine designed and tested in 1982, using a 9995 CPU and a gate array to significantly lower parts counts. It never made it into production because of the long-term part ordering strategies followed by management--they had already calculated so many sales of each part into their calculations for quarterly revenue for the year, that it made any change disastrous. A lot of innovative items never made it out of the labs and into production because of that. I have a TI-made 128K card for the 4A, an IEEE-488 interface, an EPROM burner for the PEB, a FORTI music card, and a TI DSDD Disk controller card. I've also seen a 1200 baud modem card and a Video controller card (this latter was used for computer aided instruction and would control multimedia presentations using a laser disk or a VCR).

There was a thriving TI-74/TI-99 development community in Germany at one
time.

For reasons I don't understand, TI was more amenable to German independent
development than stateside - Or so it seems to me.

If you're interested, and have'nt run across this guy, take a look...

http://www.ti99hof.org/bios/MichaelBecker/MichaelBecker.htm

There's more out there, if you look for it.

best regards, Jack

 

[carlsson]

Perhaps Texas Instruments thought Germany (Europe) was such a small market it didn't matter if they got competition from 3rd party, unlike in the USA where they would make all the money from only selling their own produced stuff? ;-)

How much profit would TI make on the 990 series, compared to if they had developed proper memory refresh on the 99 and thus made it a lower end competitor to their own minis? Perhaps the 99 series had sold even better, in particular if the market had been open to 3rd party developers and no GROM (?) lockouts. Sure, it could have gone all 2600 but it could also have gone Apple 2 or C64 as regarding to quality vs selection on software.

Chuck: According to Old-Computers.com, there was a Sanyo PHC-3000 released in Japan 1979. It would have a TMS9900, 32K RAM, 2K VRAM, 80x24 monochrome text display etc. Someone called Peter mentions on this page this computer later was bundled with a typewriter and sold as a word processing system:

http://users.tpg.com.au/pschamb/work.html

 

[Chuck(G)]

Anders, thank you for the pointer to the Sanyo PHC-3000.

I suspect that TI never thought that the consumer/commercial market for the 9900 was that lucrative, compared to the radiation-hardened SBP9900 used in defense and space applications. Apparently, there were some much faster CPUs offered by TI in their SBP9900 I2L line that are almost never seen today.

Some of the TMS9900 architecture survives today in the MSP430 microcontroller, but the register "working set" pointer is gone.