What's the highest cost vs computing power ever asked?

[TerraHertz]

Here's a PDP 8/S with a buy-it-now price of $45,000.00  OK, it is SERIAL NUMBER ZERO, but still...
https://www.ebay.com/itm/PDP-8-S-Serial-Number-0/324054935256

It's in the USA, I have nothing to do with that.
I came across that ebay auction while searching, related to a couple of PDP 8/S I'm restoring:
  http://everist.org/NobLog/20181104_PDP-8S.htm

Interesting machines, built entirely of discrete transistors, resistors and diodes. Possibly the slowest computer ever built, since it has a one bit serial shifting ALU.
Also the simplest power supply. Just a transformer, bridge rectifier and filter cap (Two rails.) No regulation needed, because the logic circuitry is so forgiving.

Or is it an Apple I board for nearly a million dollars?

[ebastler]

Here's a PDP 8/S with a buy-it-now price of $45,000.00  OK, it is SERIAL NUMBER ZERO, but still...
https://www.ebay.com/itm/PDP-8-S-Serial-Number-0/324054935256

It's in the USA, I have nothing to do with that.
I came across that ebay auction while searching, related to a couple of PDP 8/S I'm restoring:
  http://everist.org/NobLog/20181104_PDP-8S.htm

Interesting machines, built entirely of discrete transistors, resistors and diodes. Possibly the slowest computer ever built, since it has a one bit serial shifting ALU.
Also the simplest power supply. Just a transformer, bridge rectifier and filter cap (Two rails.) No regulation needed, because the logic circuitry is so forgiving.

Or is it an Apple I board for nearly a million dollars?

I guess you would have to go further back in time to find the highest cost/computing power offers. Just one example which I happen to be familiar with:

The Librascope LGP-30, a very successful computer in the late 50s and early 60s, was sold for $40,000 (1950s money, equivlent to $350k .. $400k today). That machine could execute around 500 instructions/second. Even though each instruction processed a 32-bit data word, the Apple I was more than 100 times faster, hence a bargain even at $1M. 

[gslick]

Here's a PDP 8/S with a buy-it-now price of $45,000.00  OK, it is SERIAL NUMBER ZERO, but still...
https://www.ebay.com/itm/PDP-8-S-Serial-Number-0/324054935256

That one has been relisted over and over for around a year or so that I have noticed, maybe longer, with the opening asking price of $26,500 never getting lowered. Probably still be listed there at the same price a year from now.

[obiwanjacobi]

Early quantum computers? 
Probably pretty expensive and the early ones didn't do much...

[jmelson]

Here's a PDP 8/S with a buy-it-now price of $45,000.00  OK, it is SERIAL NUMBER ZERO, but still...


Interesting machines, built entirely of discrete transistors, resistors and diodes. Possibly the slowest computer ever built, since it has a one bit serial shifting ALU.

Not even close.  The Bendix G-15 was seriously slow, but likely not the slowest.  The G-15 had a drum memory, and a word rolled off the drum every 55 us (18 KHz) I think.  All instructions had the next instruction address in them, so an optimizer could arrange instructions around a drum track so that the next instruction came up just as the current instruction was finishing.  But, unless you unrolled loops, you only got one loop every turn of the drum.  That REALLY slowed things down.  And the drum only held about 2000 words.

But, they managed to get a whole computer to fit in one box the size of a refrigerator, with 300 tubes and 3000 Germanium diodes.

Jon

[SiliconWizard]

Early quantum computers? 
Probably pretty expensive and the early ones didn't do much...

Because recent ones do?

[chmod775]

Hello everist, cool restoration of the PDP8/S. How it's going?
Do you have the schematic (or pcb photos) of the front light panel board?
I'm trying to make a 1:1 replica of it.

Thanks

[radiogeek381]

Here's the maintenance manual circa 1970: http://bitsavers.informatik.uni-stuttgart.de/pdf/dec/pdp8/pdp8s/F-87S_8sMaint_Oct70.pdf It has schematics at the end.

Bitsavers has a wealth of documentation on old machines.

And as far as cost per performance, the 8S and the rest of the 8 line existed because there was a market for a computer that a lab could afford. The price was the first consideration, performance the second.  As a laboratory instrument, even a slow computer could collect, analyze, and record data faster than a graduate student.  Labs (academic, industrial, and government) was the market target for a number of the little guys at the time. (DEC wasn't entirely alone.) The 8 and 11 series showed up everywhere from nuclear blast detectors, to animal environment controllers and recorders, numerically controlled machine tools, access/perimeter control, academic computing services, coal mines, and even the logo/light controller for the Westinghouse building in downtown Pittsburgh.

I think there were two things that made the line successful in the market: It was relatively inexpensive, and priced so that it wouldn't require approval in most companies to go all the way to the CEO or board. And the PDP-8 and PDP-11 series were made for their peripherals. There were lots of peripherals in the catalog, and some were created as building blocks for customer designed stuff. DEC made a lot of money from peripheral sales. Yet they were more than eager (in my personal experience) to make it possible to build custom modules for both lines.  The documentation was pretty good, and the Omnibus and Unibus specifications were easy to read and easy to build to.  (Modern eyes may find Omnibus and the earlier IO signaling schemes a little hard to grok. Take a deep breath. Forget all you know about TTL and CMOS. Forget modern notions of synchronous two-phase design. Embrace the current loop. Imagine you're in a large woolen mill in 1966.)

matt (Digital 1983 to 2001 (Digital/Compaq/Intel))

[Doctorandus_P]

I ask you to pay me 20 Billion Euro's, and in return I will give the answer to the sum of 34 and 8.

[granzeier]

I ask you to pay me 20 Billion Euro's, and in return I will give the answer to the sum of 34 and 8.

Yeah, that is kind of what I was thinking. I can always ask for more money to do the same thing - even compute. How about "I ask you to pay me 21 Billion Euros" and on, and on...

[jwet]

There's an Apollo Guidance computer on Sotheby's - estimated between $200-300K.  Speed .043 Mhz!  If the metric is dollars per MHz, that's almost $7M/Mhz.

Link
https://www.sothebys.com/en/buy/auction/2021/space-exploration/apollo-guidance-computer

[macboy]

ENIAC performed 0.005 MIPS at a cost of around USD $400000 (ca. 1943). The equivalent purchasing power today is about USD $7.1M
So that's $1.4B/MIPS accounting for inflation or a still rather impressive $80M/MIPS without. By contrast, a year or two ago I replaced the very old CPUs in my PC with slightly less old ones, two XEON E5-2667v2. These are 3.6 GHz with 8 cores. The seller asked, and I paid US$20 each. Intel's export compliance documentation lists the performance of this model at 211 GFLOPS. That is a cost of ~$0.0001/MIPS (MFLOPS technically, but I'll just wave my hands around and equate the two figures for this purpose). If we equate the $ value of one XEON "FLOPS" to one ENIAC "calculation per second" then we could declare that ENIAC was about 15 trillion times as expensive as a "modern" 2nd hand CPU.

The Charles Babbage Differencing Engine was a mechanical computer (though maybe not Turing complete?). A good reference I found on Quora seems to be from a professor at Stanford University who operated and maintained a working copy of it. He claims about 1 calculation (high precision addition) per second throughput. The British government spent 17000 pounds between 1823 and 1846 building an incomplete version of it. That's very roughly US$1M in today's money according to google, but I suspect that estimate is low (over two decades of research and fabrication!), and more money would be needed to complete it. Even going with that low-ball estimate, that's US$1M for 1 calculation per second, or US$1Trillion/MIPS. That's 1e16 or 10 quadrillion times as expensive as my used XEON CPU from ebay.

[brucehoult]

ENIAC performed 0.005 MIPS at a cost of around USD $400000 (ca. 1943). The equivalent purchasing power today is about USD $7.1M
So that's $1.4B/MIPS accounting for inflation or a still rather impressive $80M/MIPS without.

ENIAC was my immediate thought, if talking about electronic computers.

Although the customer (the US government) had bottomless pockets and wanted performance at any cost, I'd suspect that all successors gave better price/performance.

The Charles Babbage Differencing Engine was a mechanical computer (though maybe not Turing complete?).

Certainly not!

Babbage's "Difference Engine No.1 1" was designed (but never completed) with seven registers each holding a 20 digit number.  The ONLY calculation it can do is:

Code: [Select]

d5 += d6
d4 += d5
d3 += d4
d2 += d3
d1 += d2
d0 += d1
print d0

This allows calculating and printing successive values of an up to 6th order polynomial by setting up appropriate initial values in each register.