Oh, no. Of course not. Perish the thought. Sent from my Windows 10 phone From: Toby Thain Sent: Monday, January 23, 2017 12:45 PM To: cctalk at classiccmp.org Subject: Re: IBM 7074 and then some: "Systems we love" conference On 2017-01-23 5:16 PM, Chuck Guzis wrote: Thank God this could not happen today! --T

Consider Babbage's Analytical Engine. It was decimal and it was, not so much by intrinsic merit as by lack of competition, the fastest machine of its day. Admittedly, it ended up being little but vapourware until modern times. But applying "supercomputer" to vapourware machines is a longstanding tradition (though perhaps not one dating back quite to Babbage's day). /~\ The ASCII Mouse \ / Ribbon Campaign X Against HTML mouse at rodents-montreal.org / \ Email! 7D C8 61 52 5D E7 2D 39 4E F1 31 3E E8 B3 27 4B

On 01/23/2017 12:25 PM, Paul Koning wrote: The 7070/74 was just a member of the 7000 line. The 7030 STRETCH and even the 7090/94 were both binary and far faster. It's just that I bridle a bit when hearing the young 'uns refer to any physically large machine as a "supercomputer". It's the same feeling that I get when I see press releases today that relate that David Gelernter single-handedly developed the parallel computation. He's not old enough; at 61, he was still in high school during the ILLIAC IV era. Now, get off of my lawn! --Chuck

On 2017-01-23 6:55 PM, Paul Koning wrote: Has this been scanned anywhere? --Toby

Here is a translation of the relevant text from that report (page 6): General strategy. A very important question is how many operations are performed in parallel in the machine. Here we must distinguish between large scale parallelism and small scale parallelism. Large scale parallelism means the possibility of performing multiple arithmetic operations on distinct operands at the same time. This of course requires multiple arithmetic elements, for example 4 adders or 2 multipliers. The immediate benefit is faster calculation. In addition, such multiple arithmetic elements have a significant additional storage capacity. In the Eniac essentially the entire fast storage capacity is implemented by its adder elements. However, the distribution of the calculation program among these elements is not a trivial matter if we really want the whole system to be working efficiently. For this reason, the most effective approach to make Eniac a more manageable machine is that of von Neumann, in which 19 of the 20 adders are simply demoted to memories. In some machines (for example the relay machine of Stibits) great ingenuity has been applied to allow the machine to be split arbitrarily into a number of independent machines, or to acts as a single unit. ---- So the implication is that Eniac is a 20 unit MIMD computer. Amazing. paul

On 01/23/2017 09:04 PM, Chuck Guzis wrote: Yes! There's all sorts of little things that would drive you nuts. Things like cooling fans, power supply capacitors, cable routing hardware (clamps, ties, etc.) that they no longer make. it would be a constant job of finding suitable replacements for unavailable parts. The 7000 series had "pages" that made up a "book" with tons of old wiring that flexed every time you opened up the pages to access the circuit cards. I just CAN'T believe somebody is actually keeping such a machine in daily service. (On the other hand, CHM does have a working 1401, that also requires folding out racks of boards to access the cards, flexing similar cables.) Wikipedia says the 7070 had 14,000 SMS circuit cards, with 30,000 transistors and 22,000 diodes. Having worked on some much more recent gear with Germanium transistors, I saw about 10% of them were bad. I didn't run that gear long once I fixed it, I sold it on eBay before any more went out. But, I can't imagine that a machine with that many components could keep running awfully long between failures. As for the 1-6 ms response time, that is totally bogus. The article is complete gibberish, talking about a vast library of mag tape and ms response time in the same sentence. Maybe the 7074 prepares data weekly for some other (newer) system that is actually connected online. And, of course, to connect anything to the 7074, you'd have to build custom hardware. RS-232 had not even been invented when the 7000-series came out. They did have a 1414 unit that apparently was some kind of comm adapter, but I'll bet it took milliseconds to send one character. Oh, and the picture in the article is CLEARLY a posed IBM sales brochure photo, and not from the recent operation at the unnamed government agency. Jon

Bob Bener has written a short squib about how the 7070 came into being: http://www.bobbemer.com/BIRTH.HTM Funny, in a tragic way. --Chuck

On 01/23/2017 07:45 PM, Jon Elson wrote: This blog seems to indicate that there is NO 7074, but an emulator running on 370 hardware. This makes a lot more sense, some of these microcode emulations were still available of fairly late machines. Also, it probably would not be real hard to write a decent emulator for the 7074 and run it on modern hardware. Once you are using emulation, why not keep the host hardware current? Too bad the excerpts from the original talk are totally scrambled. Jon

Chuck Guzis asked on Mon, 23 Jan 2017 22:00:15 -0800 In 1988 I designed an ARM2 based computer (my Merlin 4, which was only built in 1992 when the ARM2 was already obsolete) and wondered if it could emulate a PC fast enough to be usable. I had written an ARM assembler and a friend did an ARM emulator running in QNX on the PC, so we wrote an 8088 emulator in ARM assembly. We then ran some simple 8088 programs on the PC emulator running on the ARM emulator running on the PC. This allowed us to figure out that an 8 MHz ARM2 would be able to run PC programs at nearly the speed of a 4.77 MHz 8088. Though there were much faster PCs at the time, the original configuration was still being sold and was popular enough that this would have been considered usable. It was not enough to convince my partner, however, which is why the machine was not finished in 1988. -- Jecel

Toby Thain wrote: It was a simple interpreter. I had been aware of the 1984 JIT implementation of Smalltalk by Allan Schiffman and L Peter Deutch. The goal was to test the worst case, however, and that was the interpreter. Note that the PC emulator was incomplete - it only had enough code to correctly run the benchmark programs. That made us miss the most serious bottleneck, which is the correct emulation of all the 8088 flags (specially the parity flag and the auxiliary flag). I have heard that Dave Ditzel was looking at emulating the PC on the Sparc at Sun and figured out that adding a tiny piece of hardware to calculate these flags would nearly double the performance of the emulator and that this was one of the inspirations for Transmeta. Pete Turnbull wrote: I don't remember if I was aware of this at that time, but it is very likely and that could have been the inspiration for our project. Note that we did all our own tools even if Acorn was selling them to their customers. In the RiscPC era they used actual Intel processors instead of (or was it as an alternative to) software emulation. Many people think Intel was stupid to have the 80186 be incompatible with the PC but they forget that this processor was practically finished by the time the PC came out and was launched just a few months later, which is long before the importance of the PC became obvious. Even with the 8088 there were plenty of MS-DOS machines that were not PC clones for about two years before clones became the standard. -- Jecel

On 01/24/2017 01:50 PM, Pete Turnbull wrote: We started getting samples of the 186 in 1981. It was, as you might imagine, pretty buggy--IIRC, we were getting a new stepping about every month for a time. Some of the bugs were really bizarre, such as DMA clobbering the SI and DI registers during the DMA operation (at the conclusion, they were restored correctly, which made tracking that one down very difficult.) What many people don't realize is that the 80286 development was going on in parallel with the 80186. It was even more buggy, as one might expect, in the pre-release steppings than the 186. Had IBM had working samples of the 186, I don't doubt for a second that they would have used it as the CPU of the 5150--the advantages were too great to ignore--integrated 20-bit DMA, timers, interrupt controller and programmable chip selects would have certainly gone far toward reducing the component count on the PC. Somewhere along the line, Intel's much ballyhooed 432 platform quietly sank under the waves (Micro-mainframe). It was a multi-chip set and hideously expensive. The 8086 really was a 1970s CPU. --Chuck

On 01/24/2017 03:05 PM, dwight wrote: Maybe, but the 80188 was even later than the 186. Given the integrated peripheral support on the 186, however, it probably wouldn't have economized much on the support circuitry to use an 8 bit BIU. --CHuck

On 01/24/2017 05:49 PM, Eric Smith wrote: Admittedly, this was before the 432 was released in any form, but I recall "Fast Eddie" our Intel sales guy quoting us about $1K for a chipset--this would have been about 1982. That was expensive in anyone's book. --Chuck

On 01/24/2017 06:13 PM, Eric Smith wrote: It could have been 1981--I'm a little fuzzy about dates back then. It could be that Bill Davidow thought that the 432 was a lost cause and wanted to discourage us--after all he was on our BOD. But I do recall the $1K chipset number and how that elicited a gasp. --Chuck

An example where that could be useful is in validating an emulation. I did that recently: running an EL-X8 emulation implemented in ALGOL (by Dick Gruene) on an EL-X8 emulation in SIMH. Run the test program that comes with it both on the nested emulation, and "standalone" on the outer emulation. Compare the outputs from the two runs, and fix or justify any differences. It helped significantly, especially because the inner emulation came with a bunch of important details about strange machine quirks. paul

On 24 January 2017 at 18:46, Jecel Assumpcao Jr. <jecel at merlintec.com> wrote: It was usable enough that Acorn sold a PC Emulator for the Acorn A305/A310 -- the original, first-series Archimedes with 8MHz ARM2 chips. E.g. http://web.archive.org/web/20021213050437/www.maxandanna.pwp.blueyonder.co.… I used it on my A310, in 1988/1989, to do real work in QuickBasic 3.0, brought home from the office. It ran at something around 2-3MHz as equivalent to an original 8088 PC, but with a much faster screen and hard disk, so overall performance was good and quite usable for MS-DOS stuff. -- Liam Proven ? Profile: https://about.me/liamproven Email: lproven at cix.co.uk ? Google Mail/Talk/Plus: lproven at gmail.com Twitter/Facebook/Flickr: lproven ? Skype/LinkedIn/AIM/Yahoo: liamproven UK: +44 7939-087884 ? ?R/WhatsApp/Telegram/Signal: +420 702 829 053

On 01/24/2017 09:25 AM, Jon Elson wrote: That would seem to me to be an insane decision also. How many 2400' 556 bpi 7-track tapes can you fit on a 1TB PC drive? Why fool with maintaining a bank of drives in that light? The biquinary coding used on the 7070 different from that of the 650. A two-out of 5 bit scheme was used (01236, with 0 being represented as 12). IIRC, the 650 used 7 bits. A word was 10 digits plus sign; a reference to 55 bit length is made, which would seem to devote a whole digit to the sign--it's not clear why this was done, as the sign appears to have only 2 values. https://www.computer.org/csdl/proceedings/afips/1959/5054/00/50540222.pdf Has a good detailed run-down. There are some interesting details; for example, although it employed a digit-sequential ALU, circuitry apparently did leading nonsignificant zero detection, so that only the number of significant digits was operated upon. Another interesting aspect was the scatter/gather tape I/O facility. One normally thinks of this as a feature on later gear; to see it in 1959 is a bit surprising. Also interesting is hardware prioritizing of I/O operations. Was the 7070 IBM's first machine with a wire-wrapped backplane? --Chuck

On 01/24/2017 06:37 PM, Jon Elson wrote: The 7070 was very early (1958) and is probably the first (or close to the first) IBM transistorized computer. The 1620 and 1401 were both 1959 introductions. First delivery of STRETCH to LANL was 1961. --Chuck

On 01/24/2017 07:55 PM, Jon Elson wrote: Yup, I was going by delivery dates. "If you can't deliver it, it doesn't exist". IIRC, Gene Amdahl and friends originally proposed a design using point-contact transistors (trying to get a uniform batch of those was "interesting"). The proposal died a couple of times, until IBM pitched what was then vaporware to LANL (then LASL). Development was started in 1956, according to WikiP. Given that the whole SMS thing was the common denominator among the 7000 and 1400 and 1620 lines, it's hard to say who was first. But the 7070 was first delivered. --Chuck

On 01/24/2017 10:01 PM, Jon Elson wrote: I find the whole period amazing. Consider that the 709 was introduced in mid 1958. It's only 6 years to the S/360. It seems that when transistor manufacture matured enough, the floodgates really opened. After 1958, nobody but developing or Warsaw pact countries even considered vacuum tube computing. --Chuck

On 01/25/2017 09:39 AM, Jon Elson wrote: Try, say, the PB-250, with 400 transistors; a 22 bit machine. There was a healthy mistrust of early transistors. Witness the one-transistor DTMF encoder used on early Touch-Tone phones or the very low transistor count in the early Dataphones. Nowadays, of course, we think of a million transistors as being modest. --Chuck

On 01/25/2017 01:37 PM, ben wrote: Were computrons ever deployed in a real product? I was under the impression that they were stillborn. --Chuck

On 01/25/2017 02:07 PM, Eric Smith wrote: According to IBM's DPD Chronology for 1959: "On August 3, DPD introduces the IBM Datacenter -- facilities in which customers rent the use of IBM 7070 systems by the hour and supply their own programmers and operators. DPD foresees a nationwide network of 25 to 30 Datacenters in major cities, with the first three located in New York City, Chicago and Los Angeles. " As a matter of fact, here's a 7070 on its way to an installation in Naples in 1959: https://www-03.ibm.com/ibm/history/exhibits/vintage/vintage_4506VV2070.html And I *did* specify computer, not calculator with regard to transistors. So who you gonna believe--a photo taken in 1959 or some guy writing 27 years later saying it didn't exist? Perhaps they've got the 7070 confused with the 7074. --Chuck

On Wed, Jan 25, 2017 at 4:00 PM, Chuck Guzis <cclist at sydex.com> wrote: Based on the bios of the authors, I'm actually more inclined to believe the book than annotation of a photo on the IBM web site. Doubtful. They do mention the 7074 but don't give much detail.

On Wed, Jan 25, 2017 at 8:07 PM, Chuck Guzis <cclist at sydex.com> wrote: That's certainly more convincing, but it still seems to be a bit of a mystery. The four authors of IBM's Early Computers were all engineers and engineering managers intimately involved with the development of the computers in that time frame, were researchers at IBM when the book was written, and based much of the book on material from IBM archives. I could imagine them getting a date wrong, but it seems pretty surprising that they would specifically claim that the 7070 was announced far before the 7090 but shipped six months after, if that wasn't true. IBM generally didn't consider a data processing system to have "shipped" until it passed the field acceptance criteria, e.g., assembled on-site and passed diagnostics. Perhaps the 7070 didn't pass acceptance testing until April 1960?

On Thu, 26 Jan 2017, Eric Smith wrote: On a system like that, how much time could elapse between, "We (customer) are switching to..."? decision planning negotiating contract begin manufacturer of first bespoke component finish manufacture of last bespoke component begin shipping delivery of last awaited part begin installation complete assembly and turn on power initial diagnostic IPL diagnostics redo whatever needs redoing first IPL to actually begin customer's data processing 'twould seem that there could be a moderately substantial amount of time between the customer saying, "we are getting" V "we are now using" Even the phrase "shipped" need not be when the customer starts printing out the nine billion names of god. (1953, so it predates the 7074) Yep, That is implied, but not explicitly confirmed by the customer saying, "we are switching to..." ANY "FIRST" will run into overlaps between the ways that the "first" date can be chosen and marked for multiple products in development. -- Grumpy Ol' Fred cisin at xenosoft.com

On 01/26/2017 07:50 PM, Fred Cisin wrote: I remember reading that one many years ago. The bit about the stars quietly winking out at the end stuck in my mind.-- Nine Giganames doesn't seem so much in our days of tera- and peta- and yottawhatsis. --Chuck

On 1/26/2017 8:37 PM, Fred Cisin wrote: That was a printout ... not card I/O. (1953, so it predates the 7074) Great Story!

On 1/25/2017 9:54 AM, Ray Arachelian wrote: There is a blog which exposes the video publicly, but the video itself is marked to do limited sharing. I think it is perfect for here, but maybe share the referenced blog pages, which has an expanded index in case they want to take it down or change it. Looks like it was an interesting meeting. thanks jim