12 ms·

What a DEC-20 Cost in 1976 (2007)

53 points by cpach 11y ago

ctstover 11y ago
A quick search puts 350k in 1976USD at about 1.5m 2015USD.
ChuckMcM 11y ago
Loved how you could run your business with 2 x 100MB of hard drive space initially. That system though sounds similar to the ones we had at USC in the late 70's early 80's. The student CS program was taught using two of the named "JUGHEAD" and "ARCHIE" with a giant 120 port terminal multiplexer (which was basically a PDP-11/40 running a dedicated instance of RSX-11 doing nothing but connecting individual VT52's and VT100s.
imglorp 11y ago
My first encounter with a DEC-20 was MACLISP for an AI class. Of course, the first thing you write is factorial, but I forgot the termination condition. Before I figured out why it was taking so long and send a brk, my CPU quota for the semester got consumed and TOPS kicked me off and locked me out. Luckily the dept secretary took pity and gave me some more time. We also had APL for a numerical methods class on that system. Yes, we had some terminals with the special APL symbols; I think they were LA-36's. http://www.columbia.edu/cu/computinghistory/la36.html http://www.columbia.edu/cu/computinghistory/la36.html
- nickpsecurity 11y ago
  Probably where the trick of adding a counter to loops and stuff to ensure termination came from. People going over their hardware budget back in the day.
  - jacquesm 11y ago
    A termination guarantee is still good practice.
    - nickpsecurity 11y ago
      I agree. Just wondered if that's where it originally came from. Could've been just from formal methods with difficulty of doing termination proofs.
      - jacquesm 11y ago
        My money would be on latency guarantees. If you add up the longest path through all your loops and one of them is a while(1) then you can no longer give any guarantees for any latency at all.
        nickpsecurity 11y ago
        Interesting possibility.
- Animats 11y ago
  Around 1980, I caused thrashing on SRI's DEC-20 by using the Boyer-Moore theorem prover. I was working on some theorems involving 16-bit arithmetic, and had numbers such as 65536. Those were internally represented as (add1 (add1 (add1 ... (zero))))..., because Boyer-Moore theory is true constructive mathematics. This used enough memory to cause paging. Boyer and Moore put in an actual integer representation after that. We got our own VAX at Ford Aerospace and ported the prover to that. Then we could get some work done on proof of correctness.
karavelov 11y ago
That price included "Six months on-site resident software specialist"
nickpsecurity 11y ago
So, they want $300,000+ for a system and only offer a 90-day warranty? That's when you know it's going to be a piece of s*. Haha.
TheOtherHobbes 11y ago
36-bit processor, 25MHz clock, 128K system memory, 100MB drives, air-conditioned room... Totally killed by a modern $5 Pi Zero. If you took a Pi back in time no one would know what to do with it. Linux would be recognisable, Ethernet would be recognised by Xerox Parc and a few experts, and only at 10 MB/s - but USB, HDMI, SC cards, the SoC, and the web would all look like advanced alien technology.
- salgernon 11y ago
  You would be surprised. A lot of of those things were invented in those times - and with realistic timeframes for when they'd be available. PARC in particular had a "skate where he puck will be in 10 years" attitude.
- VLM 11y ago
  A classic display of latency vs bandwidth. The DEC box wins on latency but the pi wins on bandwidth, for large enough tasks, anyway. A 2040 was a PDP-10 with a 30 MHz KL processor running around 30 MIPS and TOPS-20 (the source of the "20" in the name) a quarter century ago. 30e6 * 60 * 24 * 365 * 25 = about 4 e 14 instructions. I can't get a straight answer but hundreds of MIPS sounds reasonable for a pi zero. 4e14 / 500e6 / 60 / 24 / 365 = about a year and a half So sometime in 2017 a pi zero would finally catch up to a DECSystem-2040 which had a quarter century head start. Of course the 2040 has been continuously producing intermediate results for the first 25 years of the "race" while the pi output no results at all having not being invented yet. Also after 27 or so years the pi will somewhat surpass the 2040. Also the IO bandwidth and scalability of old mainframes usually puts newer hardware to shame. "My calculator has more power than a 1960s NASA mainframe that sent men to the moon" "Oh so you're saying you've been to Mars, then?" On one hand, 4e14 instructions executed would have been enough to run entire civilizations, some years ago. On the other hand, its not enough to boot modern Windows, although despite the difficulty is none the less not accomplishing anything in the real world.
  - rjsw 11y ago
    I don't think a PDP-10 does all that well on a bandwidth comparison with a modern system, I would expect something like a Cubietruck with SATA and Gigabit Ethernet to win fairly easily.
timonoko 11y ago
I remember that bare-bones Nova-1200 costed exactly 10 times more than Fiat-600 automobile in 1972 (70,000 and 7,000 Finnish markkas (1 Fmk was 3 USdollars) and with heavy import taxation). I had both some years later. Both equally out-of-date. The car actually fell apart somewhat catastrophically on the road in 1982, because no rust-proofing.
sailfast 11y ago
I'd be interested to see something like this but rather than a straight cost comparison, an analysis of what it got you in the marketplace. If you had this kind of compute power in 1976, what could you do that your competitors could not, and what was the multiplier to the business. With computing ubiquitous now the hardware doesn't always provide you with a significant competitive advantage, but I can imagine in 1976 you could own an industry that way, and $300K may have seemed like a steal. EDIT: The post might be something like: "Why we bought a DEC for 300K and never thought twice about it"
- marshray 11y ago
  > what could you do that your competitors could not Input customer orders, schedule production, track inventory, print shipping labels, do accounting, send invoices, and type memos... all in one integrated system. And if you had multiple offices, they could each have one and they could exchange data throughout the day. It was truly a revolution in productivity.
- WalterBright 11y ago
  > what could you do that your competitors could not Obviously, you could run Empire on it! http://www.classicempire.com http://www.classicempire.com
ratsmack 11y ago
That is about the same price as the VAX 11/780 we bought in 1979 (serial number 13) to run CAD/CAM software. I remember the price was around $350K, which included three RM80 drives, a vacuum column 9-track tape drive and giant battery backup. We added third party memory to bring it up to 4MB and a 40 port Emulux serial interface. It was just a few years later that a 16Mhz i386 was as fast as this system.
- hga 11y ago
  It was just a few years later that a 16Mhz i386 was as fast as this system. Wouldn't that be more like 8 years later? 1986 for the 12MHz release (originally targeted at 16), add at least a year to get up in speed and out in people's hands. And didn't the VAX have a much faster disk system, and few i386's tape backup? BTW, how reliable were your RM80s?
  - ratsmack 11y ago
    When you're having a good time a "few years" is pretty short. :-) Our RM80's were trouble free from what I can remember. We had an air conditioned, electrostatic filtered clean room for our systems which also included some PDP11's and PDP11/40's with RL02's and RK07 disk drives. The room had very limited access and you had to wear booties. The only time we lost a drive is when one of the RK07's went crazy, slamming the heads against the hard stops causing the heads to crash onto the disk.
    - hga 11y ago
      From everything I've heard, RK07s were ... subpar. I'm not surprised one failed like that. RL02s were meh, but they worked, the first system I was responsible for had 1 or 2 of them. I asked because the RA81 which was released in 1982 was, in MIT's Project Athena experience, awful, they had some sort of problem with glue, and they were losing 1 a week out of a population of no more than 40 drives (!). But it sounds like they were more careful, more lucky, whatever with the earlier RM80s, which I don't remember hearing anything bad about.
    - linker3000 11y ago
captaincrowbar 11y ago
There's a quote I heard somewhere: "A given speed of computer is available from Cray in 1980 for $10,000,000, from Sun in 1990 for $100,000, from Dell in 2000 for $1000, and from Oxfam in 2010 for $10."
- donatj 11y ago
  I feel like the thousand dollar Dell was actually profoundly more powerful.
  - rbanffy 11y ago
    Not really. It probably spent most of the CPU cycles waiting for IO and a little less time waiting for memory. Also, its SIMD units were probably limited in comparison with what Cray was doing 20 years prior. As for the Sun, using SCSI disks and a decent OS more or less compensated for limitations similar to Dell's circa 2000 PC.

53 points by cpach 11y ago

A quick search puts 350k in 1976USD at about 1.5m 2015USD.

Loved how you could run your business with 2 x 100MB of hard drive space initially. That system though sounds similar to the ones we had at USC in the late 70's early 80's. The student CS program was taught using two of the named "JUGHEAD" and "ARCHIE" with a giant 120 port terminal multiplexer (which was basically a PDP-11/40 running a dedicated instance of RSX-11 doing nothing but connecting individual VT52's and VT100s.

imglorp 11y ago

My first encounter with a DEC-20 was MACLISP for an AI class. Of course, the first thing you write is factorial, but I forgot the termination condition. Before I figured out why it was taking so long and send a brk, my CPU quota for the semester got consumed and TOPS kicked me off and locked me out. Luckily the dept secretary took pity and gave me some more time. We also had APL for a numerical methods class on that system. Yes, we had some terminals with the special APL symbols; I think they were LA-36's. http://www.columbia.edu/cu/computinghistory/la36.html http://www.columbia.edu/cu/computinghistory/la36.html

nickpsecurity 11y ago
Probably where the trick of adding a counter to loops and stuff to ensure termination came from. People going over their hardware budget back in the day.
- jacquesm 11y ago
  A termination guarantee is still good practice.
  - nickpsecurity 11y ago
    I agree. Just wondered if that's where it originally came from. Could've been just from formal methods with difficulty of doing termination proofs.
    - jacquesm 11y ago
      My money would be on latency guarantees. If you add up the longest path through all your loops and one of them is a while(1) then you can no longer give any guarantees for any latency at all.
      - nickpsecurity 11y ago
        Interesting possibility.
Animats 11y ago
Around 1980, I caused thrashing on SRI's DEC-20 by using the Boyer-Moore theorem prover. I was working on some theorems involving 16-bit arithmetic, and had numbers such as 65536. Those were internally represented as (add1 (add1 (add1 ... (zero))))..., because Boyer-Moore theory is true constructive mathematics. This used enough memory to cause paging. Boyer and Moore put in an actual integer representation after that. We got our own VAX at Ford Aerospace and ported the prover to that. Then we could get some work done on proof of correctness.

karavelov 11y ago

That price included "Six months on-site resident software specialist"

nickpsecurity 11y ago

So, they want $300,000+ for a system and only offer a 90-day warranty? That's when you know it's going to be a piece of s*. Haha.

TheOtherHobbes 11y ago

36-bit processor, 25MHz clock, 128K system memory, 100MB drives, air-conditioned room... Totally killed by a modern $5 Pi Zero. If you took a Pi back in time no one would know what to do with it. Linux would be recognisable, Ethernet would be recognised by Xerox Parc and a few experts, and only at 10 MB/s - but USB, HDMI, SC cards, the SoC, and the web would all look like advanced alien technology.

salgernon 11y ago
You would be surprised. A lot of of those things were invented in those times - and with realistic timeframes for when they'd be available. PARC in particular had a "skate where he puck will be in 10 years" attitude.
VLM 11y ago
A classic display of latency vs bandwidth. The DEC box wins on latency but the pi wins on bandwidth, for large enough tasks, anyway. A 2040 was a PDP-10 with a 30 MHz KL processor running around 30 MIPS and TOPS-20 (the source of the "20" in the name) a quarter century ago. 30e6 * 60 * 24 * 365 * 25 = about 4 e 14 instructions. I can't get a straight answer but hundreds of MIPS sounds reasonable for a pi zero. 4e14 / 500e6 / 60 / 24 / 365 = about a year and a half So sometime in 2017 a pi zero would finally catch up to a DECSystem-2040 which had a quarter century head start. Of course the 2040 has been continuously producing intermediate results for the first 25 years of the "race" while the pi output no results at all having not being invented yet. Also after 27 or so years the pi will somewhat surpass the 2040. Also the IO bandwidth and scalability of old mainframes usually puts newer hardware to shame. "My calculator has more power than a 1960s NASA mainframe that sent men to the moon" "Oh so you're saying you've been to Mars, then?" On one hand, 4e14 instructions executed would have been enough to run entire civilizations, some years ago. On the other hand, its not enough to boot modern Windows, although despite the difficulty is none the less not accomplishing anything in the real world.
- rjsw 11y ago
  I don't think a PDP-10 does all that well on a bandwidth comparison with a modern system, I would expect something like a Cubietruck with SATA and Gigabit Ethernet to win fairly easily.

timonoko 11y ago

I remember that bare-bones Nova-1200 costed exactly 10 times more than Fiat-600 automobile in 1972 (70,000 and 7,000 Finnish markkas (1 Fmk was 3 USdollars) and with heavy import taxation). I had both some years later. Both equally out-of-date. The car actually fell apart somewhat catastrophically on the road in 1982, because no rust-proofing.

sailfast 11y ago

I'd be interested to see something like this but rather than a straight cost comparison, an analysis of what it got you in the marketplace. If you had this kind of compute power in 1976, what could you do that your competitors could not, and what was the multiplier to the business. With computing ubiquitous now the hardware doesn't always provide you with a significant competitive advantage, but I can imagine in 1976 you could own an industry that way, and $300K may have seemed like a steal. EDIT: The post might be something like: "Why we bought a DEC for 300K and never thought twice about it"

marshray 11y ago
> what could you do that your competitors could not Input customer orders, schedule production, track inventory, print shipping labels, do accounting, send invoices, and type memos... all in one integrated system. And if you had multiple offices, they could each have one and they could exchange data throughout the day. It was truly a revolution in productivity.
WalterBright 11y ago
> what could you do that your competitors could not Obviously, you could run Empire on it! http://www.classicempire.com http://www.classicempire.com

ratsmack 11y ago

That is about the same price as the VAX 11/780 we bought in 1979 (serial number 13) to run CAD/CAM software. I remember the price was around $350K, which included three RM80 drives, a vacuum column 9-track tape drive and giant battery backup. We added third party memory to bring it up to 4MB and a 40 port Emulux serial interface. It was just a few years later that a 16Mhz i386 was as fast as this system.

hga 11y ago
It was just a few years later that a 16Mhz i386 was as fast as this system. Wouldn't that be more like 8 years later? 1986 for the 12MHz release (originally targeted at 16), add at least a year to get up in speed and out in people's hands. And didn't the VAX have a much faster disk system, and few i386's tape backup? BTW, how reliable were your RM80s?
- ratsmack 11y ago
  When you're having a good time a "few years" is pretty short. :-) Our RM80's were trouble free from what I can remember. We had an air conditioned, electrostatic filtered clean room for our systems which also included some PDP11's and PDP11/40's with RL02's and RK07 disk drives. The room had very limited access and you had to wear booties. The only time we lost a drive is when one of the RK07's went crazy, slamming the heads against the hard stops causing the heads to crash onto the disk.
  - hga 11y ago
    From everything I've heard, RK07s were ... subpar. I'm not surprised one failed like that. RL02s were meh, but they worked, the first system I was responsible for had 1 or 2 of them. I asked because the RA81 which was released in 1982 was, in MIT's Project Athena experience, awful, they had some sort of problem with glue, and they were losing 1 a week out of a population of no more than 40 drives (!). But it sounds like they were more careful, more lucky, whatever with the earlier RM80s, which I don't remember hearing anything bad about.
  - linker3000 11y ago

captaincrowbar 11y ago

There's a quote I heard somewhere: "A given speed of computer is available from Cray in 1980 for $10,000,000, from Sun in 1990 for $100,000, from Dell in 2000 for $1000, and from Oxfam in 2010 for $10."

donatj 11y ago
I feel like the thousand dollar Dell was actually profoundly more powerful.
- rbanffy 11y ago
  Not really. It probably spent most of the CPU cycles waiting for IO and a little less time waiting for memory. Also, its SIMD units were probably limited in comparison with what Cray was doing 20 years prior. As for the Sun, using SCSI disks and a decent OS more or less compensated for limitations similar to Dell's circa 2000 PC.