IBM Processor Technology - An Evolution 3 Dec 2104
This is where it all starts... Here are two single-gate tube modules and a ferrite core memory card. A transistor replaces a tube in logic chips, and one can build over one million transistors in the space of the dot above this "i". A tube module is 1" X 1" X 6", so a million tubes, tightly packed, would take up 6 million cubic inches. My kitchen is 14' X 15' X 8' which is just over 3 million cubic inches, or one half the space of a million tubes. And imagine the heat produced!
This is a picture of a fairly current technology Intel Celeron processor sitting next to something called a Thermal Conductivity Module, or TCM. The TCM is out of an IBM 4381 processor, vintage 1983, and has a ceramic substrate which acts a printed circuit board and heat sink. The ceramic substrate has a capacity of 36 processor chips. There were, on average, 30 of these TCM modules in a 4381 mainframe computer. So given that each TCM held up to 36 processor chips, and there were up to 35 of the TCM modules in a mainframe, the 4381 must have been a mighty computer, yes?
The 4381 could handle about one hundred concurrent users and we all know that a laptop handles one user, but let's put this into perspective. The 4381 could be supported by as much as 64MB of RAM and usually had about 4-16 disk arrays, each with about 1GB of storage. The Celeron can support more than 64GB (this is 1000 times the memory) of RAM and it is usual to see laptops with 500MB to 1TB of disk. Bluntly, the Celeron dwarfed the 4381 in all respects except user concurrency, but I would argue that the Celeron could handle 1000s of users doing the very simplistic computing the then current 4381 users were up to. The 4381 would not even be able to load a modern operating system, much less run it in anything faster than glacial speeds. A 4381 sold for about (depending on model) $500,000 and required a air-conditioned raised floor, and a boatload of power. Each of the TCMs could dissipate 90 watts of heat for a total of 27KW of pure summer misery. You need to know that in 1983, no one did much in the way of graphics on a mainframe, and the displays were mostly one color - a shade of mint. Modern operating systems devote most of their power to handling the user interface, the 4381 devoted 0% of its power to the user experience.
A last look at what was a technological marvel at the time, this TCM. This was state of the art technology and exceptionally powerful relative to the other midrange systems out there at the time. It is also hopelessly weak by today's standards. The single Celeron processor can crunch numbers much faster and only requires a single processing module and produces (wastes) a small amount of heat. It took 30 of these TCM modules to power a large 4381 and it was a slug by comparison.
One more thing: There is a measure of processor performance called Millions Instructions Per Second, or MIPS. This is often called Meaningless Indicator of Processor Speed because there is no way to compare real work across different architectures using MIPS. All that said, the 4381 model line ranged from 1.8 to 9.9 MIPS, and the Celeron line ranges all over the place, but consider its mean of about 500 MIPS.
One more thing: There is a measure of processor performance called Millions Instructions Per Second, or MIPS. This is often called Meaningless Indicator of Processor Speed because there is no way to compare real work across different architectures using MIPS. All that said, the 4381 model line ranged from 1.8 to 9.9 MIPS, and the Celeron line ranges all over the place, but consider its mean of about 500 MIPS.
4 Feb. 2015
I just received this wonderful TCM in the mail, and have to share it with respect to the other multi-chip modules. This is from 1986 and an IBM 3090 class processor - which one I am not sure. The 3090 machines ranged in speed, but topped out at about 120 MIPS. There were quite a few of these TCM modules in a 3090, dozens of them in the 120 MIPS model, the 3090-600. Dozens, and still not adding up to one, fast, modern Intel chip.
I also need to point out that the 4381 and 3090 mainframes are in no way equivalent to an Intel Celeron processor. The 3090 had a lot of processor offloading in its architecture. For instance, a keystroke did not cause an interrupt, and requests for data were handed off to outboard support processors. Still, the processor speeds and chip density are pathetic by today's standards.
The TCM is ~6" X 6" and 1-1/8" thick. This is a very heavy module.
I also need to point out that the 4381 and 3090 mainframes are in no way equivalent to an Intel Celeron processor. The 3090 had a lot of processor offloading in its architecture. For instance, a keystroke did not cause an interrupt, and requests for data were handed off to outboard support processors. Still, the processor speeds and chip density are pathetic by today's standards.
The TCM is ~6" X 6" and 1-1/8" thick. This is a very heavy module.
Here is the 3090 TCM with an Intel Celeron processor sitting atop it, and a 4381 TCM next to it. The tiny Celeron is much more powerful than the 3090 or 4381. That said, the 3090 TCM is a wonder of engineering and reflective of the day when IBM could produce a great product.
I opened up the TCM to see how many chips are in there. I count 76 chips, all connected up via the ceramic substrate. A very impressive piece of technology.
Each of the chips has a small aluminum piston poised atop it to pull heat out of the chip. That's a wee little diode off the left.
Each of the chips has a small aluminum piston poised atop it to pull heat out of the chip. That's a wee little diode off the left.
One more look at the substrate...