What accounted for the cost of ENIAC?

Question

I'm used to the fact that first-generation computers were very expensive, which I had always assumed was because they contained large numbers of vacuum tubes, each of which is a rather complex, high precision little machine in its own right. So it's not too surprising that ENIAC cost $487,000, which was a lot of money in those days. According to that article,

By the end of its operation in 1956, ENIAC contained 18,000 vacuum tubes; 7,200 crystal diodes; 1,500 relays; 70,000 resistors; 10,000 capacitors; and approximately 5,000,000 hand-soldered joints.

Which sound like large numbers! So today I tried to find out exactly how much vacuum tubes cost, and found a 1948 issue of Billboard magazine (it's on Google Books, with one of those links that's a whole paragraph long in its own right; are those considered suitable for linking here?), listing prices for several kinds of vacuum tubes at less than a dollar each, including e.g. the 6SN7 which was one of the more important ones for building computers, at $0.75 each. That seems to be retail price in quantity one by mail order; wholesale would probably be less. Say it's $0.50 wholesale. So 18,000 of those would be $9,000.

So what was the other $478,000 spent on?

Other costs of using vacuum tubes, like electricity, and the technicians needed to be constantly replacing burned-out tubes? Perhaps significant, but out of scope here, being operating costs, whereas the above figure seems to be purely upfront purchase cost.

Other electronic components? Doesn't look like it. The only more numerous one in the above list is resistors, and those have to be pretty cheap.

Peripheral equipment like punch card readers? I don't think so. Everything I have read elsewhere says it was considered preferable to use standalone punch card equipment for any job it could do (like keying data onto cards, or sorting cards by sequence number) precisely because it was much cheaper than a computer.

The five million hand-soldered joints? Probably a significant cost, but doesn't seem like it should account for the bulk of it. A quick skim of some hourly wage figures for the late forties suggests factory workers might've been typically paid more than a dollar an hour, say round up to two dollars for total cost of employment. To account for half of the total cost, the soldiered joints would've needed to cost 5 cents each. That would match if it takes 1.5 minutes to solder each joint. It's been three decades since I picked up a soldering iron, and I wasn't an expert with it even then, but that sounds on the high side. I don't think it should take that long.

Inaccurate cost figure, or ENIAC unrepresentative? I don't think so; other computers of the era were even more expensive, for example Univac which ended up costing over a million dollars.

Difference between manufacturing cost and sale price to account for manufacturer profit margin? Okay, that should be a significant chunk of the sale price, but not 90% of it.

R&D cost? That would make sense for ENIAC, but not for Univac, because that ended up selling a total of 46 units, so R&D cost would be amortized to a much smaller sum per unit, yet the unit price was over double ENIAC.

What am I missing?

Answer 1

One of the biggest factors is that when you have a machine that requires 5,000,000 successful solder joints to function properly, you need to make sure that all of your solder joints are really really good. If 1 in 10,000 of your solder joints is subtly bad, that means that the first time you try to put everything together you will have 5000 bad solders and the machine will be no where near functioning. Any time you have to troubleshoot this, it's probably taking at least 30 minutes due to the size of the machine, and the fact that hardware bugs aren't obvious. Now throw in the fact that all of the 18,000 vacuum tubes, 7,200 crystal diodes, and 1,500 relays are mechanical devices and will all fail at various rates. These aren't 21st century electronics, at least one of these components probably fails on your average day. Each time this happens, you have to go through and figure out what went wrong and replace it. On a related note, much of the engineering on early computers wasn't focused on speed as much as minimizing components. Why? Because if your machine was too big, the odds of being able to complete a calculation before hardware failure goes down dramatically.

This gets especially complicated once you realize that one possible side effect of a component failing is letting through too much current. The reason this is so bad is it means one component failing can fry many other things in the vicinity.

As such, the number of components in the final machine isn't that useful a number, because the number used over the lifespan of the project is likely much higher, and all the adding and removing of components during troubleshooting is much more expensive than initial construction.

Answer 2

R&D stuff isn't manufactured (at first).

It's usually partially constructed, ripped up, and redone, multiple times, with long testing and debug cycles in between, all the while with payroll running up the tab. Tons (literally) of fried or used components and partial assemblies can go into the junk bin. The specifications then often evolve with the early test results, requiring more redo's. Then include building and rebuilding all the mechanical stuff as well as the electrical circuits and assemblies. Add tooling, management, operations, floor space, clerical (women not-yet-called-programmers), and accounting, etc.

Answer 3

So what was the other $478,000 spent on?

Paying people to design and build it would have been a fairly big component. People often underestimate the cost of labour, particularly if it is their own time. Also looking at the photos of it on wikipedia, there were a lot of components besides valves. There were racks and other cabinets and what looks like a fairly substantial air-con system. The power supply probably wasn't exactly "off the shelf either". Also, I expect whatever the IO devices were wouldn't have been cheap.

Answer 4

Here's a video about manufacturing UNIVAC 1108 (circa 1965, so a bit later but still illuminating). Notice how much is done by hand, even things like winding coils.

According to the University of Minnesota the UNIVAC division employed about 10,000 people.

Price list for some UNIVAC 1104 components (long page of text, search for "UNIVAC 1004 PRICE LIST"). Price list for some UNIVAC 1108 components. Another, longer, price list for UNIVAC 1106 and 1108 (PDF, price list starts on page 10).

So, labor costs. Making the parts, manual QA on the parts, assembling the parts into a whole. Loss on parts that don't pass QA. Maybe some jobs can be done by a worker with only basic training, but a lot of the tasks look like they need quite a lot of skill.

Some of the machines to make and test parts look pretty expensive just by themselves, so you have to amortize those across all of the units.

You have to pay for the R&D of the previous system, but also have enough profit to fund R&D for the next system.

Advertising costs.