People often used to piggy-back chips 30 years ago, I did it once to get 2 Mbytes of DRAM on a 1 Mbyte transputer board of mine. Someone I knew had a tower of eight Z80 processors, all operating in parallel, interfaced to a TRS-80 Model I, for doing Mandelbrot calculations. He built his stuff in cardboard boxes, that one had a chimney made of rolled up cardboard for cooling.
I first used stacked chips in 1971. In high school we piggybacked TTL decade counters (SN7490 was it?) so as to get a required frequency divider ratio.
Piling up static RAMs was also a favorite. Until recently I still had an old prototype 8088 card with stacked up RAMs in my office.
How did that stacked Z80 thing work? Surely must have had RAM to go with each processor. Hmm.. stacked up Props running ZiCog...
I think he did everything in registers, with an SRAM shared by the CPUs containing the program, and loaded from the TRS-80. It was very clever.
If I can dredge up his name, or that of a friend of his (they both belonged to the TRS-80 users group that I formed), from my long-term memory, I might be able to contact him for the details. I just remembered his first name, Simon, which is a start. The annoying thing is that I have a mental image of the pair of them. His friend belonged to a well-known amateur radio club, I might be able to find him that way.
I've just remembered the Z80 stacker's name - Simon Langton. I'll see if I can track him down.
That's what I don't understand.
I would not have thought a Z80 has sufficient registers to do a sensible Madlebrot calculation even on a single point.
We might assume there is only one copy of the code being fetched by all the Z80's. But that would imply they all do the same number of iterations and take the same code path all the time (SIMD) which would be very in efficient.
At some point the coords have to be fetched by the Z80's and the results written out, at which point it looks like bus contention to me.
Must have been something devious going on there.
They might have had a 2k SRAM each, it was many years ago. It really did speed up the Mandelbrot calculations, but wasn't anything like as fast as one of my transputer systems.
Has anyone piggybacked more than two Propeller chips? (i.e. one chip on top of another chip, etc.)
Sal said he used to flatten out the pins on DIPs, stack them up, and wirewrap directly to the leads, and there's no reason we can't do the same with props.
I have plans to make a stack of P8X32A-Q44 when the multiprop support is a little more mature. The idea is to solder surface mount resistors and capcitors directly to the pins for the multiprop, and wirewrap to the remaining pins for peripheral connections. I haven't figured out what I would do with it, but a pen containing a stack of several props and a couple batteries would be fun if nothing else.
I thought the benefit of stacking chips was when they shared multiple pins in common (clock, latch, Vdd, Vss, etc.). A stack of Props would have some of its pins shared but I wouldn't think you'd want or need very many of the IO pins soldered together. You might want a multiple bit data bus but I'd think you'd want most of the IO pins available for other uses. Having to bend to pins outward to access them make this more awkward than if the were just side by side on a circuit board.
You'd want to make sure that two Props in a stack didn't both set a shared pin as output with one Prop holding it high and the other Prop holding it low.
The Z80, if I remember correctly(I'm on vacation in Bangkok, so far away from my books, and can't be bothered to search) has pins for 'BUS Request', which in theory could be used to arbitrate access to a common memory block.
As or stacking SRAM, this was also done on commercial products back in the 80s.
I know I have a 'laptop' from that time which has a memory exansion that consists of 4 SRAM chips stacked on top of each other.
(The Chip Select pin was bent out and a wire soldered to each. All the other pins were soldered straight down the stack)
I just want to make use I have my alphabet soup correct. RPC stands for Remote Procedure Call, right?
I'm constantly amazed at all the amazing Propeller projects that have been done that I didn't notice at the time. I think I've just recently learned enough to appreciate what many of these past threads (eg. OctoProp) are talking about.
Many years ago there was an MCU interconnect system called ChipRack, I had some involvement with it. The system consisted of small PCBs about 2" square, in little plastic frames with contacts round the edges. One could have a CPU on one layer, memories on other layers, and I/O on other layers, built up in stacks like little towers, and mounted on a larger PCB. The chips were intended to be in die form, with wire bonding. They built some demonstration systems using a 68000 die, using a stack of four ChipRack assemblies. I used to have one.
I was interested in using it for building large transputer systems. Mike Anstey, who invented it, is still around:
People often used to piggy-back chips 30 years ago, I did it once to get 2 Mbytes of DRAM on a 1 Mbyte transputer board of mine. Someone I knew had a tower of eight Z80 processors, all operating in parallel, interfaced to a TRS-80 Model I, for doing Mandelbrot calculations. He built his stuff in cardboard boxes, that one had a chimney made of rolled up cardboard for cooling.
Leon, I'm very interested in your work with the Transputer. Perhaps you can start a thread for it. Much of that information can be related to Propeller projects.We should not forget those important accomplishments made.
It would be very helpful to learn more information about the tower of eight Z80s in parallel! Another man far ahead of his time!
It was a long time ago, when I was playing around with the Inmos transputer. I was attracted to it at the time because I liked the idea of parallel processing, Inmos was a British company, and it was the most powerful device that was generally available (20 MIPS). It also needed minimal hardware support - just a handful of DRAM and a couple of logic chips for a complete processing system that had four high-speed serial links to other transputers, enabling large parallel processing arrays to be built - systolic arrays, pipelines, hypercubes, etc. One very nice feature was that the device could be booted from a link.
When Inmos launched the T414 32-bit transputer around 1982 I bought a couple of chips (about £500 each!!!!), wire-wrapped a little board with one on it, and a parallel interface chip, connected it to my PC printer port, hand-assembled a little program to light an LED, and it worked!
I then wire-wrapped a similar board with eight 256k DRAMs (64k x 4) on it, and that worked, so I designed a little pluggable module with a 16-way 0.1" Molex connector, paid someone £750 to design a PCB for it on a 4-layer board and a couple of prototype boards made. I showed it off the first transputer conference and a company called Transtech said that they'd buy lots of them if it had 1 Mbyte of DRAM, and I could supply a motherboard holding 16 of the modules, with Inmos link switching devices on it so that the connection topology could be modified by the host PC. They gave me £2000 there and then, to finance the development! I went ahead with the project - redesigned the module for 1 Mbit 256kx4 DRAMs (they were about £25 each!) and designed the motherboard. I was one of the first people to use those DRAMs in a commercial product. Everything worked, and Transtech (they are still around) bought a total of about 30 of the motherboards and 500 of the modules, over the next year or so. They built them into a very substantial enclosure with a massive power supply, and sold them for about £15,000 a time. I also sold a few of my modules to hobbyists, at £750 a time with a T800 floating-point transputer. The DRAM was intended to run with three wait-states, but I provided a header and jumper so that they could be set to run with two wait-states; they actually worked reliably, even with my piggy-backed chips, although the timing was way out of spec.
I also developed an ISA card that plugged into a PC with one of my modules, a PAL, and an Inmos interface chip, that could be used to host the Inmos TDS development software, and compilers available from other suppliers. It was much cheaper than the Inmos equivalent, and I sold quite a few of those.
I didn't make all that much money, though, as I sold Transtech the bare assembled and tested boards and they supplied the chips. Many universities and research establishments bought my system as it offered more computing power for the money than anything else on the market, at the time. Then Inmos developed their own transputer modules (TRAMs) which were technically better than mine, but much more complex, and the market dried up for my system. My modules did have the advantage of being easily removed and replaced, and Plessey Research at Roke Manor used them for a fault-tolerant system - they'd invite a customer to unplug two or three modules at random and the system would keep running. I never did find out what many of the people using my boards were doing with them, systems went to places like RARDE where military weapons are designed, and GCHQ, the UK intelligence centre. I heard from someone who used to sell stuff to the Soviet Union that the president of the Soviet Academy of Sciences wanted to buy some of my kit, even though it was embargoed under the export regulations in force at the time. I heard that he had acquired a Sinclair QL, although that was technically embargoed, because of the storage density of the Microdrives and the MC68008 chip it used. To see what would happen, I applied for an export license, as I did for export to approved countries, and it went through! I then told the DoT that they'd made a mistake, and they were very embarrassed and asked me to return the documents immediately. I later heard that I could have been in trouble, even with a valid export license, if I had actually shipped the boards. I'd have probably got a few months inside. Apparently, the way to get stuff like that into the Soviet Union, was to ship it to a middleman in Finland, and it would then make its way across the border into Russia, and no one would know about it.
I managed all the above as a "one man band".
Iann Barron, one of the founders of Inmos, has written this interesting account of the company history:
Humanoido
Posts: 5,770
Comments
I first used stacked chips in 1971. In high school we piggybacked TTL decade counters (SN7490 was it?) so as to get a required frequency divider ratio.
Piling up static RAMs was also a favorite. Until recently I still had an old prototype 8088 card with stacked up RAMs in my office.
How did that stacked Z80 thing work? Surely must have had RAM to go with each processor. Hmm.. stacked up Props running ZiCog...
I think he did everything in registers, with an SRAM shared by the CPUs containing the program, and loaded from the TRS-80. It was very clever.
If I can dredge up his name, or that of a friend of his (they both belonged to the TRS-80 users group that I formed), from my long-term memory, I might be able to contact him for the details. I just remembered his first name, Simon, which is a start. The annoying thing is that I have a mental image of the pair of them. His friend belonged to a well-known amateur radio club, I might be able to find him that way.
I've just remembered the Z80 stacker's name - Simon Langton. I'll see if I can track him down.
I would not have thought a Z80 has sufficient registers to do a sensible Madlebrot calculation even on a single point.
We might assume there is only one copy of the code being fetched by all the Z80's. But that would imply they all do the same number of iterations and take the same code path all the time (SIMD) which would be very in efficient.
At some point the coords have to be fetched by the Z80's and the results written out, at which point it looks like bus contention to me.
Must have been something devious going on there.
Sal said he used to flatten out the pins on DIPs, stack them up, and wirewrap directly to the leads, and there's no reason we can't do the same with props.
I have plans to make a stack of P8X32A-Q44 when the multiprop support is a little more mature. The idea is to solder surface mount resistors and capcitors directly to the pins for the multiprop, and wirewrap to the remaining pins for peripheral connections. I haven't figured out what I would do with it, but a pen containing a stack of several props and a couple batteries would be fun if nothing else.
I tried stacking 595s to make this LED display.
I also used the piggy-back trick to make a 256KB SRAM module.
I thought the benefit of stacking chips was when they shared multiple pins in common (clock, latch, Vdd, Vss, etc.). A stack of Props would have some of its pins shared but I wouldn't think you'd want or need very many of the IO pins soldered together. You might want a multiple bit data bus but I'd think you'd want most of the IO pins available for other uses. Having to bend to pins outward to access them make this more awkward than if the were just side by side on a circuit board.
You'd want to make sure that two Props in a stack didn't both set a shared pin as output with one Prop holding it high and the other Prop holding it low.
Duane
I would like to know how you used your stack of Props.
BTW I didn't see your post before my first reply to this thread. I too busy glorifying my stacks.
Duane
As or stacking SRAM, this was also done on commercial products back in the 80s.
I know I have a 'laptop' from that time which has a memory exansion that consists of 4 SRAM chips stacked on top of each other.
(The Chip Select pin was bent out and a wire soldered to each. All the other pins were soldered straight down the stack)
From OctoProp thread: http://forums.parallax.com/showthread.php?115925-OctoProp-Demo&p=839672&viewfull=1#post839672
Anyways, look for my announcement shortly - therotically you could stack a huge number of these pcbs
Done here http://forums.parallax.com/showthread.php?132769-Cluso-s-CpuBlade-1-quot-x1-quot-Stackable-Propeller-PCB-(Also-in-panels-of-28)
I just want to make use I have my alphabet soup correct. RPC stands for Remote Procedure Call, right?
I'm constantly amazed at all the amazing Propeller projects that have been done that I didn't notice at the time. I think I've just recently learned enough to appreciate what many of these past threads (eg. OctoProp) are talking about.
Thanks for posting the picture and reminder.
Duane
I was interested in using it for building large transputer systems. Mike Anstey, who invented it, is still around:
http://www.mike.anstey.btinternet.co.uk/Chiprack/essen.doc
http://forums.parallax.com/showthread.php?132769-Cluso-s-CpuBlade-1-quot-x1-quot-Stackable-Propeller-PCB
It would be very helpful to learn more information about the tower of eight Z80s in parallel! Another man far ahead of his time!
When Inmos launched the T414 32-bit transputer around 1982 I bought a couple of chips (about £500 each!!!!), wire-wrapped a little board with one on it, and a parallel interface chip, connected it to my PC printer port, hand-assembled a little program to light an LED, and it worked!
I then wire-wrapped a similar board with eight 256k DRAMs (64k x 4) on it, and that worked, so I designed a little pluggable module with a 16-way 0.1" Molex connector, paid someone £750 to design a PCB for it on a 4-layer board and a couple of prototype boards made. I showed it off the first transputer conference and a company called Transtech said that they'd buy lots of them if it had 1 Mbyte of DRAM, and I could supply a motherboard holding 16 of the modules, with Inmos link switching devices on it so that the connection topology could be modified by the host PC. They gave me £2000 there and then, to finance the development! I went ahead with the project - redesigned the module for 1 Mbit 256kx4 DRAMs (they were about £25 each!) and designed the motherboard. I was one of the first people to use those DRAMs in a commercial product. Everything worked, and Transtech (they are still around) bought a total of about 30 of the motherboards and 500 of the modules, over the next year or so. They built them into a very substantial enclosure with a massive power supply, and sold them for about £15,000 a time. I also sold a few of my modules to hobbyists, at £750 a time with a T800 floating-point transputer. The DRAM was intended to run with three wait-states, but I provided a header and jumper so that they could be set to run with two wait-states; they actually worked reliably, even with my piggy-backed chips, although the timing was way out of spec.
I also developed an ISA card that plugged into a PC with one of my modules, a PAL, and an Inmos interface chip, that could be used to host the Inmos TDS development software, and compilers available from other suppliers. It was much cheaper than the Inmos equivalent, and I sold quite a few of those.
I didn't make all that much money, though, as I sold Transtech the bare assembled and tested boards and they supplied the chips. Many universities and research establishments bought my system as it offered more computing power for the money than anything else on the market, at the time. Then Inmos developed their own transputer modules (TRAMs) which were technically better than mine, but much more complex, and the market dried up for my system.
I managed all the above as a "one man band".
Iann Barron, one of the founders of Inmos, has written this interesting account of the company history:
http://www.cs.man.ac.uk/CCS/res/res32.htm#c
http://www.cs.man.ac.uk/CCS/res/res33.htm#c
I agree with Humanoido on this one.
Thanks for sharing the story Leon.
Duane