ENIAC: THE WAY WE WERE

[JamesD]

http://hackaday.com/2017/01/31/eniac-the-way-we-were/

[GroovyBee]

Interesting read.

[SSG]

Really Neat!

[JamesD]

The thing that stood out the most to me, is the power requirements, price, and time it took to perform a calculation.
Definitely better than the 20 hours a human required... though I think that is an exaggeration.
But still, you couldn't do it on the fly with people unless you had pre-generated tables... which was the point.

For comparison, here is ENIAC vs several 8 bit machines in MIPS
(I understand MIPS is not the best comparison and some of the #s in the table these come from are clearly wrong):
Machine MIPS

• ENIAC 0.00289
• MITS Altair 0.01
• Apple II 0.02
• Commodore 64 0.02 (this was listed at 0.2 MIPS which was clearly wrong)
• TRS-80 0.04 (I think this is wrong, should be about 0.02)
• TRS-80 M3 0.04 (this machine is clocked faster than the Model I)
• Atari 800XL 0.165 (seems like a large jump from 1 MHz)
• TRS-80 M4 0.2 (and another big jump)
• Amiga 3000 12.5
• Raspberry Pi 2 9700

Even the lowly Altair is 3.46 times as fast. It also had more memory.
Calculations would go from 30 seconds to 8.67.
The Amiga calculates it in a tiny fraction of a second.

I want to take a Raspberry Pi back in time and sell it to the military.

[CatPix]

Always impressive to read.

It's mindblowing how, even at the time, things marched on quickly, with EDSAC coming, and other computers shaping the modern computers of today

 

Some technical details strike me :

18,000 tubes, 7,000 diodes

Of course I guess that they make a difference to know the "computing" part. But I think it might be important to know if the diodes here were tubes or solid state. I think they were tubes, which would explain the high failure rate, and how the technicians spotted "blakened tubes".

Worn out tubes don't blacken, but busted valves do, soit's probably valves that failed. ENIAC would have been more reliable if it used solid state diodes.

Unlike what people think, tubes are very reliable. Any tube radio made after 1920 and kept in a reasonnable state might need a capacitor change, but it's likely that the 80 or so years old tubes will work fine.

 

But tubes diodes, or "valves" are pretty much lightbulbs (and in many cases, you can use a 100w incandescent lightbulb in a radio to replace a diode). And those were highly prone to failure, it's even why many tube radio fans advise to use incandescent lightbulbs rather than diodes, or to use solid state diodes, because old valves will burn quickly.

 

A modified valve using solid state diodes and a fridge lamp :

 

 

Make me wonder how an ENIAC would be if it used transistors instead of tubes.

Maybe you could even have one at home?

 

In 1965, Olivetti rleased what's considered as the first home/office computer, the Olivetti Programma 101 :

It looks like a big calculator, on purprose, as olivetti feared that users would be afraid of owning a computer but would be comfortable with a "programmable calculator" But you can write programs, and even record programs on MAGNETIC cards.

The Programma 101 was used for the Appollo program, no less.

I can't find much figures to compare the computing mower of the Programma related to the ENIAC, but on such exotic systems, it's hard to make comparisons...

Edited February 3, 2017 by CatPix

[high voltage]

Here's some more info on the Olivetti 101:

http://royal.pingdom.com/2012/08/28/the-first-pc-from-1965/#respond

Edited February 3, 2017 by high voltage

[high voltage]

And some home computers from way back:

http://www.blinkenlights.com/pc.shtml

[simbalion]

I've never heard of using bulbs in place of tubes in radios or tubes burning out really fast in radios. Unless something is majorly wrong, a tube will long outlast a light bulb. Especially the cheap ones you find now, if you can find them.

In any case, it is interesting to see how far we have come from the first days of computers like Eniac.

[ClausB]

Where do those MIPS figures come from? Most 6502 instructions take 2 or 3 cycles, so at 1 MHz it's about 0.4 MIPS. A typical Z80 instruction executes in 7 to 10 cycles. At 2 MHz that's about 0.2 MIPS.

[JamesD]

Where do those MIPS figures come from? Most 6502 instructions take 2 or 3 cycles, so at 1 MHz it's about 0.4 MIPS. A typical Z80 instruction executes in 7 to 10 cycles. At 2 MHz that's about 0.2 MIPS.

It' s Dhrystone 2.1 MIPS, which is a benchmark, not counting clock cycles for a handful of instructions.

The 6502 doesn't rank higher largely due to the lack of 16 bit support, which means more instructions are often required to do the same thing.

You also jump in number of clock cycles for instructions that index data.

 

My 64 column graphics text code shows about a 2:1 MHz ratio for 65816 vs Z80. On a 6502 the ratio is even lower.

But in that case, the Z80 is actually doing more because of the hardware it's running on and some corners cut on the 6502 code for speed.

Given the amount of data that must be accessed, it's probably not as favorable to the 6502 as some other programs though.

It requires a LOT of indexed addressing.

 

I'm guessing they estimated the ENIAC number or used an emulator.

 

[CatPix]

I've never heard of using bulbs in place of tubes in radios or tubes burning out really fast in radios. Unless something is majorly wrong, a tube will long outlast a light bulb. Especially the cheap ones you find now, if you can find them.

In any case, it is interesting to see how far we have come from the first days of computers like Eniac.

As I said, it's not the amplifier tubes, triodes, etc... It's only a specific type of tube that is usually called valve, and that act as a diode and power regulator.

Valves in older radios became less common in the 1930's when radio makers switched from autotransformers to safety transformers.

Thinking about it, since those transformers were used because you had all kind of voltages in France/Europe (110V AC and DC, 130Volts AC, 220Volts AC, 240Volts AC) you might have less of those valves tubes in the US....

[simbalion]

Ah, ok. Basically, what we call a rectifier tube over here that helps to regulate the power and convert it to a more smooth DC flow in our old radios. I must admit, most rectifier tubes here are pretty rugged, though there are exceptions to the rule! Some transformerless radios here used higher voltage filaments in the rectifier and output tubes and those can be prone to burning out. Actually, what old radio fans over here are doing is adding an extra type of resistor into their sets to bring the line voltages down. Our radios here were originally designed for 110-115 volts and now we are seeing 120-125 volts AC. That's hard on the old electronics. Actually, wonder how that is effecting our gaming and computer hobbies as well? Poor Eniac would probably be blowing more tubes and parts than ever if it was still complete and functional! They were pushing the tubes hard enough in those early beasts.

[CatPix]

Well aside for some machines like the Commodore 64 that have AC unregulated power going in, the rise in voltage is not a huge concern. It's probably harder on the power supplies, but most of our old machines have regulators inside them that drop the voltage (in most cases) from 9V/10V to 5Volts DC.

Normally, power supplies and most electric stuff can accept the indicated voltage with a 20% tolerance.

 

And if diodes noted in the specs are rectifiers (I guess they have to be, as calculations would need a "pure" voltage" then there is 2.57 tubes for one diode.

I guess it was required for better power filtering, but you also increase the chance of failure. Even if only 0,1% of rectifier tubes are likely to fail every day, then you have 7 rectifier failure per day.

That sounds consistant with what I read, with ENIAC operating without failure for 24 hours at the very best.

[JamesD]

I wonder what the annual budget for tubes was.

[CatPix]

Good question.

From the comments of the article, the engineers used "older" types of tubes, probably the ubiquitous octal tubes, instead of "miniatures" tubes. It's likely that it's because octal tubes were made in larger quantities than miniatures tubes (that were introduced in 1939, so probably not mass produced because of war time), much cheaper, and probably more know and available in military version unlike miniature tubes, that would have saved space and power.

[JamesD]

...

My 64 column graphics text code shows about a 2:1 MHz ratio for 65816 vs Z80. On a 6502 the ratio is even lower.

...

 

That should read a 2:1 speed ratio at the same MHz.