to deploy a 2nd hand Cray-1 at UQ, we had to raise the ex-IBM 3033 floor, it turned out the bend radius for flourinert was NOT the same as a water cooled machine. We also installed a voltage re-generator which is basically a huge spinning mass, you convert Australian volts to DC, spin the machine, and take off re-generated high frequency volts for the cray, as well as 110v on the right hz for boring stuff alongside. the main bit ran off something like 400hz power, for some reason the CPU needed faster mains volts going in.
The flourinert tank has a ball valve, like a toilet cistern. we hung a plastic lobster in ours, because we called the cray "Yabbie" (Queensland freshwater crayfish)
That re-generator, the circuit breakers are .. touchy. the installation engineer nearly wet his trousers flipping on, the spark-bang was immense. Brown trouser moment.
The front end access was Unisys X11 Unix terminals. They were built like a brick shithouse (to use the australianism) but were a nice machine. I did the acceptance testing, it included running up X11 and compiling and running the largest Conways game of life design I could find on the net. Seemed to run well.
We got the machine as a tax-offset for a large Boeing purchase by Australian defence. End of life, one of the operators got the love-seat and turned it into a wardrobe in his bedroom.
Another, more boring cray got installed at department of primary industries (Qld government) to do crops and weather modelling. The post cray-1 stuff was .. more ordinary. Circular compute unit was a moment in time.
400 Hz is really the next best thing to a switching supply, as the transformers and filter capacitors can be smaller than they would need to be at 50/60 Hz. It can save cost and space for filter capacitors, especially in a three-phase system where there's not as much ripple to deal with.
Another rationale may have been that the flywheel on the motor-generator would cover a multitude of power-quality sins.
> the main bit ran off something like 400hz power, for some reason the CPU needed faster mains volts going in.
Aerospace originally did that to reduce component size, CDC and IBM took advantage of the standard in the early 60's.
Strangely, it seems mainframes didn't adopt switching power supplies until the end of the 70's, despite the tech being around since the end of the 60's.
I used a Cray C-90 and T3D 256-core machine from 1995-1999. The T3D used commodity Alpha 21164, and was already behind the T3E when we got it (Cray refurbished.) By the end it was outclassed by an SGI Oxygen box with 8 CPUs. I’d already ported a lot of software from SunOS and HP-UX to Irix and Unicos (Cray) and it was easy to move it to Linux in the end.
IIRC the desk side onyx had the royal purple stripes and only accepted one CPU board, the rack mount version were that blue-purple color, more indigo (color)
So there were two computers made with the same bits mid 90s. Origin(compute, blue) and onyx(graphics, purple). Both had deskside and rack systems.
Onyx had a few slots reserved for graphics, original could have more compute boards. But you could certainly put two couple cpu boards in an onyx deskside or rack.
It must have been more complex then that, I was at the skywalker ranch when onyx was being replaced by o200/2000 and never saw a purple onyx at Kerner(ILM), I had a purple Indy impact as my home machine and was looking for purple. I hated teal with a passion.
Perhaps all the purple rack onyx had been dumped but we dug through ILMs boneyard looking to add to our cxfs cluster, but the FC bus speed was too low.
It is possible that R10k was different or that there were multiple chassis. The desk sides I had experience with required RAM in slot 1, with CPU in slot 2, with up to 4 CPUs on the board.
o200 was more restrictive, with 2 CPUs per chassis, with the ability to cray-link two chassis for a total of 4 CPUs, more required o2000.
But this was a quarter of a century ago or more by now…so I may misremember things.
I actually forgot that they made a low end ‘Indy’ and never saw one. We called the Indigo2 Indy
There were some oddities with ad dollars and prestige clients in the 90s, where systems would be upgraded, avoiding swapping out the serialized, asset tagged parts, yet upgrading systems.
The teal Indigo2s were the original with the impact graphics ones being purple.
But all the Indigo2s I saw typically were r10k + max impact graphics despite the color.
The cases were identical.
But Evans & Sutherland and Lucas are the only places I dealt with SGI, so probably not a good source for typical customer’s experience
There a lot of discussion here https://retrocomputing.stackexchange.com/questions/7412/why-... but nothing seems conclusive.. I would wager the last answer, "IBM was using 400Hz", to be most directly causal reason. The motor-generator configuration might provide galvanic isolation and some immunity to spikes and transients as well?
Smaller transformers and capacitors in all the linear power supplies.
400Hz is still common in aircraft. Distribution losses are higher, but you're going across the room, not across the country.
There you go. Not to doubt what you say, but we definitely had the love seat yet we also had a tank of vaguely flourescing green liquid. Maybe we had some intermediate state, the cray-1 cpu form but the cray-2 upgraded coolant.
It wouldn't surprise me if we had the bastard love-child of leftovers from Boeing.
The GP also mentions X11 Terminals. My wiki-fu shows the X Windowing System came about on or around 1983, while Cray-1 was 1970s vintage. I assume that was an upgrade at some later point.
X Window Release 3 (X11R3) was introduced on Cray into UNICOS (a UNIX variant of Cray OS, COS) in late 1989 using ported 64-bit Xlib. But it was not widely used within small Cray community.
But MIT cooked up X11 "PROTOCOL" of Xlib in late 1985 to 1986 on Univac and Unix in C with many other X libraries written in Common Lisp.
X10R3 mostly stabilized the Xlib around a few platforms and CPU architecture (DDX) in a"long" preparation for X11R1 in September 1987.
It was fat finger memory. it was X10R3 or something similar, which I had previously used in UCL on Ultrix machines in the 80s. I don't think it was R5, I don't think much got upgraded in that space but .. it was a long time ago.
This was a SECOND HAND cray. It was a tax contra made in the 90s when Boeing sold a lot of stuff to the Australian Defence forces, and to avoid a massive tax burden donated science trash to the uni.
I knew a guy who worked at one of the national labs that had its own Cray supercomputer, in a computer room with a big observation window that visitors could admire it through, of course (all Crays required observations windows to show them off).
Just before a tour group came by, he hid inside the Cray, and waited for them to arrive. Then he casually strolled out from the back of the Cray, pulling up the zipper of his jeans, with a sheepish relieved expression on his face, looked up and saw the tour group, acted startled, then scurried away.
The Cray 1-ish machines I had access to at Shell and Chevron were most definitely tucked away in rooms with no visibility into them. In fact the Chevron machine room had pretty stern "no photography" placards, which I took seriously and is sadly why I don't have a photo of sitting on the loveseat of their machine.
Getting access took just short an act of God and I was a sysadmin in the central support group! They didn't want us puttering on the machines, so as far as I could tell it mostly sat idle.
I was part of a group from the University of Minnesota that traveled to Chippewa Falls to tour the Cray plant. I expected to see exactly what you describe, the computer behind a big observation window. But instead they took us to a machine that was undergoing final testing before delivery, I think it was serial #5. They were extremely proud of their cooling and invited us to put our hands on the panels in the opening of the C to see how cool it was. I still freak out over thinking about what would have happened if someone had tripped and fallen into all those wires on the inside.
I was a heavy user of the Cray-1, Cray-XMP and Cray-2 at the magnetic fusion computing center at Lawrence Livermore National Lab. One of the important things to remember was that the Crays of this generation were vector processors, that is they carried out operations across multiple elements of arrays at once. If you didn't structure your code properly, you didn't see these extraordinarily high instruction rates.
And, yes, quite some time ago I noticed that my cell phone had surpassed the capabilities of these early Crays :)
The CRAY-1 was so ridiculously ahead of its time that it took until the Pentium MMX (1997) for “ordinary” computers to catch up to its raw performance.
That’s 20 years or about 10,000X the available VLSI transistors via Moore’s Law.
With the visual analysis of LLMs, throw in some extra ingredients with a camera, microphone, speaker, and display, and were getting close. Add Bluetooth for biosensors and... Iterate a couple of generations...
"The 160 MFLOPS Cray-1 was succeeded in 1982 by the 800 MFLOPS Cray X-MP, the first Cray multi-processing computer. In 1985, the very advanced Cray-2, capable of 1.9 GFLOPS peak performance
...
By comparison, the processor in a typical 2013 smart device, such as a Google Nexus 10 or HTC One, performs at roughly 1 GFLOPS,[6] while the A13 processor in a 2019 iPhone 11 performs at 154.9 GFLOPS,[7] a mark supercomputers succeeding the Cray-1 would not reach until 1994."
These flops are not the same. The 2013 phone flops are fp32, the A13 flops look to be fp32 as well (not entirely sure), while the Cray numbers (like the rest of the HPC industry) are fp64 (Cray 1 predates what would become IEEE 754 binary64 though, so not same exact arithmetic but similar in dynamic range and precision).
A modern Nvidia GB200 only does about 40 tflop/s in fp64 for instance. You can emulate higher precision/dynamic range arithmetic with multiple passes and manipulations of lower precision/dynamic range arithmetic but without an insane number of instructions it won't meet all the IEEE 754 guarantees for instance.
Certainly if Nvidia wanted to dedicate much more chip area to fp64 they could get a lot higher, but fp64 FMA units alone would be likely >30 times larger than their fp16 cousins and probably 100s of times larger than fp4 versions.
Oh but we had The Forbin Project, its sequel Colossus, and later Wargames. Not to mention Star Trek episodes with malignant computers. And I have No Mouth But I Must Scream.
In the 70s, science fiction fed me Cold War fears of world-controlling mainframes.
Colossus: The Forbin Project is simply a renamed release of The Forbin Project, a few months after the later had a poor opening. Didn’t help the box office much. I liked it, back when it was easy to dismiss as an impossible dystopia.
> The CRAY-1 was so ridiculously ahead of its time that it took until the Pentium MMX ...
You'd need a different comparison to show how the Cray-1 was special. If the comparison is to single commodity CPUs, like the Pentium MMX, you could make much the same comparison for many mainframes and supercomputers. Several supercomputers in the 1980s exceeded 1 GFLOP, for example, and it wasn't until the 2000s that you could get commodity CPUs with that performance.
If these kinds of machines interest you I highly recommend the book "The Supermen" by Charles Murray. It has all the details you would ever want on Seymour Cray and others in the business.
I was working at a geophysical company in the 80's and we lusted after a Cray-1. Best we could afford where array processors (CSPI) connected to VAX systems.
Inmos not imos, if my memory cells serve ne correctly. I lived overseas at the time, so I did not hear about the Cray till like 1980/81. My friend (we were like 12) had an idea to write a simulator for digital circuits, and I was puzzled as to why you would want to simulate a circuit when you can build it and test it. He was way ahead of his time.
> The aesthetics of the machine have not been neglected. The CPU is attractively housed in a cylindrical cabinet. The chassis are arranged two per each of the twelve wedge-shaped columns. At the base are the twelve power supplies. The power supply cabinets, which extend outward from the base are vinyl padded to provide seating for computer personnel.
They needed someplace to put the power supplies anyway. The seat cushions just added a touch of whimsy. Could you imagine IBM putting seats around one of their computers?
According to this link https://www.datacenterdynamics.com/en/news/cray-1-supercompu... these things retailed for $8 million (in 1975 dollars!). That's almost $50,000,000 today - I can't even imagine how much work went into justifying a purchase like that.
I was able to pick up recycled flourinert, which I used for an immersed dual celeron setup. It was mind boggling to see the submerged motherboard chugging away and silence beyond the the soft whir/gurgle of the water pumps. My first CRAYon machine was so messy. I always hoped that it was coolant from our U of MN's Cray.
Compare this to modern system: we could fully simulate this kind of slow computer in real time, I would guess, with all the wheels, clockwork and mechanics as well as logic circuits with full electricity. Deciding the simulation level would be a bit challenging - is this an atomic, an electron or more common sense simulation -; still I think the simulation would work as well as this kind of document.
Absolutely not on the electrical level. ISA emulation, probably, maybe, though these Crays are not IEEE machines so you have to actually fully emulate the floating point operations, instead of punting them to the FPU. That may make getting hundreds of MFLOPs on difficult.
2. It’s a secure facility and wanted to prevent people from looking in.
3. To not have to look at something outside.
4. It’s a secure facility and wanted to prevent the chance of taking a picture of someone or something outside that could compromise the location of the computer or someone’s identity; sometimes the first place a photogenic computer was built was at a customer site.
Many computer rooms had windows. Computers were expensive things that companies often wanted to show off - but they needed special rooms with raised floors, HVAC (which also make them look even more exclusive/impressive and worth showing off), and the cost (millions of dollars when senior operators might be making $50k/year) meant you wanted to be careful who could touch them. Thus a window so that you could bring outsiders by the room on the way to a meeting and they could look in and see that shiny equipment.
Remember in 1980 the modern PC wasn't even invented (Apple II was) yes. Most people didn't have access to any terminal as part of their job, only specialized positions did. Typewriters where was everyone had on their desk. As such a computer was something a company could wow visitors with. Times have changed.
In college I had an account on our ACM chapter's DEC Alpha, which I used primarily for mudding. Its DNS name was cray-ymp.xxx.xxx.edu, which resulted in more than a few moments of shock/consternation from mods. "You're mudding from a CRAY Y-MP???"
"Seymour said he thought it was odd that Apple bought a Cray to design Macs because he was using Macs to design Crays. He sent me his designs for the Cray 3 in MacDraw on a floppy.” reports KentK.
Cray himself. Here's a talk about designing it. My favorite part is his description of the aluminum machining they had to invent in order to move the freon through the frame to keep the machine from melting. It's a great talk.
In the past few years whenever I re-watch 2001 when Dave is shutting down HAL, I see a spaceship capable data center. And HAL sings "Daisy.." finally at the foundational, bare metal layer.
I remember doing a report on this in high school in the late 80s. I'd love to do an order of magnitude comparison to a modern M4 Mac... Amazing how far we've come.
I just did a BOTE calculation for my iPhone (A17 Pro chip; GPU rated at 4 Tflops). According to the sales blurbage in TFA, the Cray 1 performed at 80 Mflops. (Yes, that is OBVIOUSLY not comparing apples to Apples -- pun intended). Unless I've dropped a decimal point, my iPhone is (capable of) 50,000 times the floating point speed of a Cray 1.
I remember computer magazines of the time talking about "a Cray on a chip" in their April 1st jokes.
Well... we're there. Far past, in fact. We live in the future that then was so far out of reach that people could only joke about it, not consider it a realistic possibility.
The "R exp" is subroutine call (which saves return address to register B00), and I believe "J Bjk" is the subroutine return.
The Cray-1 didn't have a hardware stack, so subroutine call is basically just jump there and back, using a register for the return address rather than pushing/popping it to/from the stack.
Another oddity of the instruction set that stands out (since I'm in process of defining a VM ISA for a hobby project) is that the branch instructions test a register (A0 or S0) rather than look at status flags. In a modern CPU a conditional branch, if (x < y), is implemented by compare then branch where the compare instruction sets flags as if it had done a subtraction, but doesn't actually modify the accumulator. In the Cray this is evidentially done by doing an actual subtraction, leaving the result in A0, then branching by looking at the value of A0 (vs looking at flags set by CMP).
I recall when reading TAOCP that Knuth's MIX assembly supported subroutines by requiring the caller to modify the RET call to it's own address (obviously not re-entrant!). This sort of thing was common when Knuth started in the early 60's, may have still been around by the time of the Cray.
This Cray version isn't so bad - it just requires that if the callee itself calls other subroutines, then it has to save/restore this B00 return address register. You could still support re-entrant routines with this as long as you saved/restored the return address to a software stack rather than a fixed location, but I wonder if Cray compilers typically supported that?
Apparently the reason for using a register vs stack for return address was because memory access (stack) was so much slower.
I'm kinda tempted to use this for the VM I'm designing, which will run on the 6502 8-bit micro. The 6502 doesn't have any 16-bit operations, so pushing and popping using a software defined 16-bit stack pointer is slow, and saving return address to zero page would certainly be faster. It'd mean that rather than always pushing/popping the SP, you only do it in the callee if needed. It's an interesting idea!
Cray's design for Control Data before starting his own company was interesting. You were required to start each subroutine with a jump instruction, and the subroutine call instruction would modify the memory at that location to a jump back to the caller. To return from a subroutine you would just branch back to its beginning.
The following quote gives some sense of how "manual" this was:
> "On execution of the return jump instruction (007), register Boo is set to the next instruction parcel address (P) and a branch to an address specified by ijkm occurs. Upon receiving control, the called routine will conventionally save (Boo) so that the Boo register will be free for the called routine to initiate return jumps of its own. When a called routine wishes to return to its caller, it restores the saved address and executes a 005 instruction. This instruction, which is a branch to (Bjk), causes the address saved in Bjk to be entered into P as the address of the next instruction parcel to be executed."
Details were up to the compiler that produced the machine code.
Essentially, the B00 register is a Top Of (Return) Stack or TOS register. It’s great for leaf routines.
You have to push it to your preferred stack before the next operation. You do the cycle-counting to decide if it’s a good ISA for your implementation, or not.
Obviously, ISAs with a JSR that pushes to stack are always using an extra ALU cycle for the SP math, then a memory write.
Doing it with a (maybe costless) register transfer followed by (only sometimes) a stack PUSH can work out to the same number of cycles.
With improvements in memory speed or CPU speed, that decision can flip.
Consider that in this era, your ALU also had the job of incrementing the PC during an idle pipeline stage (maybe the instruction decode). Doing a SP increment for a PUSH might compete with that, so separating the two might make the pipeline more uniform. I don’t know any of the Cray ISAs so this is just a guess.
Before IBM's deep blue, apparently there was a chess program called Cray Blitz, I'm sure the cray had the raw power(20MHz? But with 64 parallel operations) needed to beat the top humans, but the software just wasn't there yet.
Blitz was ported to C and continued development FOSS as Crafty, mainly by a U of Alabama professor, but to this day it can't beat top humans on modern cpu (topping at 2650 elo instead of 2850 of, say, carlsen.)
The latest version of Crafty has a significantly higher rating on CCRL than Fritz 10, the version that defeated Kramnik in 2006. He was the World Champion and was rated 2750 at the time. I do not know what source you used for Crafty’s rating but ratings from different lists are not comparable. It is highly probable that Crafty running on a Ryzen could defeat any human.
I am also of the opinion that with an optimised program the CRAY-1 would have been on par with Karpov and Fischer. I also think that Stockfish or some other strong program running on an original Pentium could be on par with Carlsen. I am not sure if Crafty’s licence would count as FOSS.
to deploy a 2nd hand Cray-1 at UQ, we had to raise the ex-IBM 3033 floor, it turned out the bend radius for flourinert was NOT the same as a water cooled machine. We also installed a voltage re-generator which is basically a huge spinning mass, you convert Australian volts to DC, spin the machine, and take off re-generated high frequency volts for the cray, as well as 110v on the right hz for boring stuff alongside. the main bit ran off something like 400hz power, for some reason the CPU needed faster mains volts going in.
The flourinert tank has a ball valve, like a toilet cistern. we hung a plastic lobster in ours, because we called the cray "Yabbie" (Queensland freshwater crayfish)
That re-generator, the circuit breakers are .. touchy. the installation engineer nearly wet his trousers flipping on, the spark-bang was immense. Brown trouser moment.
The front end access was Unisys X11 Unix terminals. They were built like a brick shithouse (to use the australianism) but were a nice machine. I did the acceptance testing, it included running up X11 and compiling and running the largest Conways game of life design I could find on the net. Seemed to run well.
We got the machine as a tax-offset for a large Boeing purchase by Australian defence. End of life, one of the operators got the love-seat and turned it into a wardrobe in his bedroom.
Another, more boring cray got installed at department of primary industries (Qld government) to do crops and weather modelling. The post cray-1 stuff was .. more ordinary. Circular compute unit was a moment in time.
(I think I've posted most of this to HN before)
400 Hz is really the next best thing to a switching supply, as the transformers and filter capacitors can be smaller than they would need to be at 50/60 Hz. It can save cost and space for filter capacitors, especially in a three-phase system where there's not as much ripple to deal with.
Another rationale may have been that the flywheel on the motor-generator would cover a multitude of power-quality sins.
> the main bit ran off something like 400hz power, for some reason the CPU needed faster mains volts going in.
Aerospace originally did that to reduce component size, CDC and IBM took advantage of the standard in the early 60's.
Strangely, it seems mainframes didn't adopt switching power supplies until the end of the 70's, despite the tech being around since the end of the 60's.
I used a Cray C-90 and T3D 256-core machine from 1995-1999. The T3D used commodity Alpha 21164, and was already behind the T3E when we got it (Cray refurbished.) By the end it was outclassed by an SGI Oxygen box with 8 CPUs. I’d already ported a lot of software from SunOS and HP-UX to Irix and Unicos (Cray) and it was easy to move it to Linux in the end.
You means SGI onyx? The cool looking purple one?
IIRC the desk side onyx had the royal purple stripes and only accepted one CPU board, the rack mount version were that blue-purple color, more indigo (color)
So there were two computers made with the same bits mid 90s. Origin(compute, blue) and onyx(graphics, purple). Both had deskside and rack systems.
Onyx had a few slots reserved for graphics, original could have more compute boards. But you could certainly put two couple cpu boards in an onyx deskside or rack.
Uhoh, I think I mistook Oxygen for Origin. It was 25 years ago.
The funny thing was that our Cray sales team, from the C90 and T3D, jumped ship from Cray to SGI before their merger.
So got to set up some Indys as web servers/ Oracle database. Surprise! Same sales team. Then the merger happened and we got the Origin boxes.
It must have been more complex then that, I was at the skywalker ranch when onyx was being replaced by o200/2000 and never saw a purple onyx at Kerner(ILM), I had a purple Indy impact as my home machine and was looking for purple. I hated teal with a passion.
Perhaps all the purple rack onyx had been dumped but we dug through ILMs boneyard looking to add to our cxfs cluster, but the FC bus speed was too low.
It is possible that R10k was different or that there were multiple chassis. The desk sides I had experience with required RAM in slot 1, with CPU in slot 2, with up to 4 CPUs on the board.
o200 was more restrictive, with 2 CPUs per chassis, with the ability to cray-link two chassis for a total of 4 CPUs, more required o2000.
But this was a quarter of a century ago or more by now…so I may misremember things.
Oh I think you may be right here. I was thinking originally 2000/onyx 2 which were basically the same system just with or without graphics pipes.
And you’re right origin had three models (maybe more) a tower system (o200), a deskside system (same as onyx2) and a rack (same as onyx2).
For Indy impact so you mean the teal/purple indigo2? Weirdly the teal desktop was named indigo2..
Weird but cool stuff.
I actually forgot that they made a low end ‘Indy’ and never saw one. We called the Indigo2 Indy
There were some oddities with ad dollars and prestige clients in the 90s, where systems would be upgraded, avoiding swapping out the serialized, asset tagged parts, yet upgrading systems.
The teal Indigo2s were the original with the impact graphics ones being purple.
But all the Indigo2s I saw typically were r10k + max impact graphics despite the color.
The cases were identical.
But Evans & Sutherland and Lucas are the only places I dealt with SGI, so probably not a good source for typical customer’s experience
I'm tugging at my memory with this page: http://www.sgistuff.net/hardware/systems/index.html
I had a teal Indigo2, an O2, and an Indy at various points. Neat stuff, but time moves on.
There's some complex setups listed in the origin/onyx pages. You could mostly mix and match boards to get what you wanted.
I should add, the o200 was a 4u chassis with a desk side skin option.
If you popped off the plastic and had rails you could convert them.
The o200 had stonith (shoot the other node in the head) ports like the o2000 for clustering etc..
There a lot of discussion here https://retrocomputing.stackexchange.com/questions/7412/why-... but nothing seems conclusive.. I would wager the last answer, "IBM was using 400Hz", to be most directly causal reason. The motor-generator configuration might provide galvanic isolation and some immunity to spikes and transients as well?
Smaller transformers and capacitors in all the linear power supplies. 400Hz is still common in aircraft. Distribution losses are higher, but you're going across the room, not across the country.
Cray-1 or Cray-2? IIRC Fluorinert was new with the Cray-2, while wikipedia suggests that the Cray-1 used a freon as coolant.
There you go. Not to doubt what you say, but we definitely had the love seat yet we also had a tank of vaguely flourescing green liquid. Maybe we had some intermediate state, the cray-1 cpu form but the cray-2 upgraded coolant.
It wouldn't surprise me if we had the bastard love-child of leftovers from Boeing.
You consistency in spelling fluor as flour gave me visions of using pancake batter as a coolant :-D
The GP also mentions X11 Terminals. My wiki-fu shows the X Windowing System came about on or around 1983, while Cray-1 was 1970s vintage. I assume that was an upgrade at some later point.
X Window Release 3 (X11R3) was introduced on Cray into UNICOS (a UNIX variant of Cray OS, COS) in late 1989 using ported 64-bit Xlib. But it was not widely used within small Cray community.
But MIT cooked up X11 "PROTOCOL" of Xlib in late 1985 to 1986 on Univac and Unix in C with many other X libraries written in Common Lisp.
X10R3 mostly stabilized the Xlib around a few platforms and CPU architecture (DDX) in a"long" preparation for X11R1 in September 1987.
https://www.x.org/wiki/X11R1/?utm_source=chatgpt.com
Thank you. I didn’t realize that progress was still being made for Cray-1 machines a decade later.
It was fat finger memory. it was X10R3 or something similar, which I had previously used in UCL on Ultrix machines in the 80s. I don't think it was R5, I don't think much got upgraded in that space but .. it was a long time ago.
This was a SECOND HAND cray. It was a tax contra made in the 90s when Boeing sold a lot of stuff to the Australian Defence forces, and to avoid a massive tax burden donated science trash to the uni.
Since this was a 2nd hand machine having the upgrades for X11 doesn't seem unlikely.
Cray old-timer here; 'nert was definitely Cray-2 and later. In the Cray-3 it was turned into mist.
Hello Bandit!
I knew a guy who worked at one of the national labs that had its own Cray supercomputer, in a computer room with a big observation window that visitors could admire it through, of course (all Crays required observations windows to show them off).
Just before a tour group came by, he hid inside the Cray, and waited for them to arrive. Then he casually strolled out from the back of the Cray, pulling up the zipper of his jeans, with a sheepish relieved expression on his face, looked up and saw the tour group, acted startled, then scurried away.
The Cray 1-ish machines I had access to at Shell and Chevron were most definitely tucked away in rooms with no visibility into them. In fact the Chevron machine room had pretty stern "no photography" placards, which I took seriously and is sadly why I don't have a photo of sitting on the loveseat of their machine.
Getting access took just short an act of God and I was a sysadmin in the central support group! They didn't want us puttering on the machines, so as far as I could tell it mostly sat idle.
> a photo of sitting on the loveseat of their machine.
Unsurprisingly, I would say, the Cray is literally the prime example in "A Gallery of Computers as Furniture".[†]
[†] https://news.ycombinator.com/item?id=46601097
I was part of a group from the University of Minnesota that traveled to Chippewa Falls to tour the Cray plant. I expected to see exactly what you describe, the computer behind a big observation window. But instead they took us to a machine that was undergoing final testing before delivery, I think it was serial #5. They were extremely proud of their cooling and invited us to put our hands on the panels in the opening of the C to see how cool it was. I still freak out over thinking about what would have happened if someone had tripped and fallen into all those wires on the inside.
I was a heavy user of the Cray-1, Cray-XMP and Cray-2 at the magnetic fusion computing center at Lawrence Livermore National Lab. One of the important things to remember was that the Crays of this generation were vector processors, that is they carried out operations across multiple elements of arrays at once. If you didn't structure your code properly, you didn't see these extraordinarily high instruction rates.
And, yes, quite some time ago I noticed that my cell phone had surpassed the capabilities of these early Crays :)
The CRAY-1 was so ridiculously ahead of its time that it took until the Pentium MMX (1997) for “ordinary” computers to catch up to its raw performance.
That’s 20 years or about 10,000X the available VLSI transistors via Moore’s Law.
Seymour Cray never designed a silicon microprocessor. I wonder what he would have created had that opportunity presented itself.
https://en.wikipedia.org/wiki/Seymour_Cray
In his memory some have said his name are the last two characters in RISC and CISC.
I wonder how many times faster my iPhone 17 Pro Max is?
Sometimes I like to remind myself we are living in the future. A future that seemed like SciFi when I was a kid in the 70s!
Sadly I don’t think we will ever see Warp Drives, Time Travel or World Peace. But we might get Jet Packs!
At the risk of sounding cliche i'll point out that ios probably uses several times the capacity of a cray 1 just to get the keyboard to work.
Yes, but the impressive part is that the iPhone does it on a few mW instead of a few MW
Recently I've found myself wanting a tricorder type device.
With the visual analysis of LLMs, throw in some extra ingredients with a camera, microphone, speaker, and display, and were getting close. Add Bluetooth for biosensors and... Iterate a couple of generations...
We're getting there!
So...a cell phone?
:-)
Need local "offline mode" somehow.
From the Wikipedia article on the Cray 1:
"The 160 MFLOPS Cray-1 was succeeded in 1982 by the 800 MFLOPS Cray X-MP, the first Cray multi-processing computer. In 1985, the very advanced Cray-2, capable of 1.9 GFLOPS peak performance
...
By comparison, the processor in a typical 2013 smart device, such as a Google Nexus 10 or HTC One, performs at roughly 1 GFLOPS,[6] while the A13 processor in a 2019 iPhone 11 performs at 154.9 GFLOPS,[7] a mark supercomputers succeeding the Cray-1 would not reach until 1994."
These flops are not the same. The 2013 phone flops are fp32, the A13 flops look to be fp32 as well (not entirely sure), while the Cray numbers (like the rest of the HPC industry) are fp64 (Cray 1 predates what would become IEEE 754 binary64 though, so not same exact arithmetic but similar in dynamic range and precision).
A modern Nvidia GB200 only does about 40 tflop/s in fp64 for instance. You can emulate higher precision/dynamic range arithmetic with multiple passes and manipulations of lower precision/dynamic range arithmetic but without an insane number of instructions it won't meet all the IEEE 754 guarantees for instance.
Certainly if Nvidia wanted to dedicate much more chip area to fp64 they could get a lot higher, but fp64 FMA units alone would be likely >30 times larger than their fp16 cousins and probably 100s of times larger than fp4 versions.
> while the A13 processor in a 2019 iPhone 11 performs at 154.9 GFLOPS,[
Sustained ? Or just for some ms when the thermals kick in ?
>how many times faster my iPhone 17 Pro Max is..
Sadly most of that power is not working for you, most of the time, but working against you, by spying, tracking and manipulating you.
Bet that was not include in your sci-fi dreams in 70s..
Oh but we had The Forbin Project, its sequel Colossus, and later Wargames. Not to mention Star Trek episodes with malignant computers. And I have No Mouth But I Must Scream.
In the 70s, science fiction fed me Cold War fears of world-controlling mainframes.
Colossus: The Forbin Project is simply a renamed release of The Forbin Project, a few months after the later had a poor opening. Didn’t help the box office much. I liked it, back when it was easy to dismiss as an impossible dystopia.
Oh. Well the sequel in print was named Colossus. It is about continuing life under the reign of supercomputer.
I'd love to see you substantiate that - bet you can't.
I certainly can't. I'm running GrapheneOS.
Not terribly impressive considering an average 20 year old super computer c. 2005 is still about 100x as fast as today's best consumer cpus
Well, yeah, Dennard scaling ended around 20 years ago!
> The CRAY-1 was so ridiculously ahead of its time that it took until the Pentium MMX ...
You'd need a different comparison to show how the Cray-1 was special. If the comparison is to single commodity CPUs, like the Pentium MMX, you could make much the same comparison for many mainframes and supercomputers. Several supercomputers in the 1980s exceeded 1 GFLOP, for example, and it wasn't until the 2000s that you could get commodity CPUs with that performance.
If these kinds of machines interest you I highly recommend the book "The Supermen" by Charles Murray. It has all the details you would ever want on Seymour Cray and others in the business.
I was working at a geophysical company in the 80's and we lusted after a Cray-1. Best we could afford where array processors (CSPI) connected to VAX systems.
Blew my mind age 4. Then found out about the imos transputer. And robotics magazine. 70's were popping. Ponging?
The Transputer was ahead of it's time, maybe due to anticipating the end of Moore's law a bit too early!
The Transputer's inter-chip channel connections remind a bit of Nvidia's NVLink or AMD's Infinity Fabric.
Inmos not imos, if my memory cells serve ne correctly. I lived overseas at the time, so I did not hear about the Cray till like 1980/81. My friend (we were like 12) had an idea to write a simulator for digital circuits, and I was puzzled as to why you would want to simulate a circuit when you can build it and test it. He was way ahead of his time.
XMOS is still keeping the Inmos dream alive, more or less!
> The aesthetics of the machine have not been neglected. The CPU is attractively housed in a cylindrical cabinet. The chassis are arranged two per each of the twelve wedge-shaped columns. At the base are the twelve power supplies. The power supply cabinets, which extend outward from the base are vinyl padded to provide seating for computer personnel.
Answers my question.. I always thought they can't possibly want people sitting on something so expensive.
It looks really nice - but never understood the seating. Who came up with that?
It ensures the Cray is placed as the centerpiece of the room, not as mere equipment shoved on the wall
Probably they could achieve the same with circular construction but without the seats.
They needed someplace to put the power supplies anyway. The seat cushions just added a touch of whimsy. Could you imagine IBM putting seats around one of their computers?
[dead]
for a much more in-depth description of its predecessor, see https://archive.computerhistory.org/resources/text/CDC/cdc.6...
i don't think there is a comparable book about the cray-1?
There are several versions of the Cray 1 Hardware Reference manual online. They're not as descriptive as Design of a Computer, but still informative.
thank you! that is very informative.
My first time on Cray, I thought I had a core dump, my program ended so quickly. No, the job had finished.
Same with "go test". :)
According to this link https://www.datacenterdynamics.com/en/news/cray-1-supercompu... these things retailed for $8 million (in 1975 dollars!). That's almost $50,000,000 today - I can't even imagine how much work went into justifying a purchase like that.
I was able to pick up recycled flourinert, which I used for an immersed dual celeron setup. It was mind boggling to see the submerged motherboard chugging away and silence beyond the the soft whir/gurgle of the water pumps. My first CRAYon machine was so messy. I always hoped that it was coolant from our U of MN's Cray.
Compare this to modern system: we could fully simulate this kind of slow computer in real time, I would guess, with all the wheels, clockwork and mechanics as well as logic circuits with full electricity. Deciding the simulation level would be a bit challenging - is this an atomic, an electron or more common sense simulation -; still I think the simulation would work as well as this kind of document.
Absolutely not on the electrical level. ISA emulation, probably, maybe, though these Crays are not IEEE machines so you have to actually fully emulate the floating point operations, instead of punting them to the FPU. That may make getting hundreds of MFLOPs on difficult.
Why the thin black curtain on the window?
Thoughts:
1. To block some sunlight from getting in.
2. It’s a secure facility and wanted to prevent people from looking in.
3. To not have to look at something outside.
4. It’s a secure facility and wanted to prevent the chance of taking a picture of someone or something outside that could compromise the location of the computer or someone’s identity; sometimes the first place a photogenic computer was built was at a customer site.
I suspect the curtains are closed to avoid causing exposure issues etc. in the photos being taken.
As for windows in a computer room, seems a bit unusual, but a nicer working environment than the usual windowless box I'd guess...
Many computer rooms had windows. Computers were expensive things that companies often wanted to show off - but they needed special rooms with raised floors, HVAC (which also make them look even more exclusive/impressive and worth showing off), and the cost (millions of dollars when senior operators might be making $50k/year) meant you wanted to be careful who could touch them. Thus a window so that you could bring outsiders by the room on the way to a meeting and they could look in and see that shiny equipment.
Remember in 1980 the modern PC wasn't even invented (Apple II was) yes. Most people didn't have access to any terminal as part of their job, only specialized positions did. Typewriters where was everyone had on their desk. As such a computer was something a company could wow visitors with. Times have changed.
It was the 70s, and that was simply the style at the time.
Or, you know: Sometimes, a window shade is just a window shade.
In college I had an account on our ACM chapter's DEC Alpha, which I used primarily for mudding. Its DNS name was cray-ymp.xxx.xxx.edu, which resulted in more than a few moments of shock/consternation from mods. "You're mudding from a CRAY Y-MP???"
who was the industrial designer? Cray-1 fits so well it looks straight out of 2001: A space Odyssey
somewhat related:
"Seymour said he thought it was odd that Apple bought a Cray to design Macs because he was using Macs to design Crays. He sent me his designs for the Cray 3 in MacDraw on a floppy.” reports KentK.
https://cray-history.net/2021/07/16/apple-computer-and-cray-...
Cray himself. Here's a talk about designing it. My favorite part is his description of the aluminum machining they had to invent in order to move the freon through the frame to keep the machine from melting. It's a great talk.
https://www.youtube.com/watch?v=vtOA1vuoDgQ
In the past few years whenever I re-watch 2001 when Dave is shutting down HAL, I see a spaceship capable data center. And HAL sings "Daisy.." finally at the foundational, bare metal layer.
> "The compact mainframe occupies a MERE 70 sq. ft. of floor space."
(emphasis mine)
I remember doing a report on this in high school in the late 80s. I'd love to do an order of magnitude comparison to a modern M4 Mac... Amazing how far we've come.
I just did a BOTE calculation for my iPhone (A17 Pro chip; GPU rated at 4 Tflops). According to the sales blurbage in TFA, the Cray 1 performed at 80 Mflops. (Yes, that is OBVIOUSLY not comparing apples to Apples -- pun intended). Unless I've dropped a decimal point, my iPhone is (capable of) 50,000 times the floating point speed of a Cray 1.
In my back pocket. To watch cat videos.
> I'd love to do an order of magnitude comparison to a modern M4 Mac... Amazing how far we've come.
Yes, would be nice to compare the capabilities (multiuser, multitasking, security, RCE). Did we get _so_ far ? How many users can a Mac sustain ?
Cray 1 - 160 MFLOPS
M4 CPU - 280 GFLOPS
M4 GPU - 2900 GFLOPS
I remember computer magazines of the time talking about "a Cray on a chip" in their April 1st jokes.
Well... we're there. Far past, in fact. We live in the future that then was so far out of reach that people could only joke about it, not consider it a realistic possibility.
In one lifetime.
I had the cover of this pinned up on my wall! Supercomputer porn.
I quickly skimmed the instruction set and did not see anything resembling a sub-routine call or branch and link instruction.
Did I miss it?
The "R exp" is subroutine call (which saves return address to register B00), and I believe "J Bjk" is the subroutine return.
The Cray-1 didn't have a hardware stack, so subroutine call is basically just jump there and back, using a register for the return address rather than pushing/popping it to/from the stack.
Another oddity of the instruction set that stands out (since I'm in process of defining a VM ISA for a hobby project) is that the branch instructions test a register (A0 or S0) rather than look at status flags. In a modern CPU a conditional branch, if (x < y), is implemented by compare then branch where the compare instruction sets flags as if it had done a subtraction, but doesn't actually modify the accumulator. In the Cray this is evidentially done by doing an actual subtraction, leaving the result in A0, then branching by looking at the value of A0 (vs looking at flags set by CMP).
Gemini explains this as being to help pipelining.
I recall when reading TAOCP that Knuth's MIX assembly supported subroutines by requiring the caller to modify the RET call to it's own address (obviously not re-entrant!). This sort of thing was common when Knuth started in the early 60's, may have still been around by the time of the Cray.
This Cray version isn't so bad - it just requires that if the callee itself calls other subroutines, then it has to save/restore this B00 return address register. You could still support re-entrant routines with this as long as you saved/restored the return address to a software stack rather than a fixed location, but I wonder if Cray compilers typically supported that?
Apparently the reason for using a register vs stack for return address was because memory access (stack) was so much slower.
I'm kinda tempted to use this for the VM I'm designing, which will run on the 6502 8-bit micro. The 6502 doesn't have any 16-bit operations, so pushing and popping using a software defined 16-bit stack pointer is slow, and saving return address to zero page would certainly be faster. It'd mean that rather than always pushing/popping the SP, you only do it in the callee if needed. It's an interesting idea!
Cray's design for Control Data before starting his own company was interesting. You were required to start each subroutine with a jump instruction, and the subroutine call instruction would modify the memory at that location to a jump back to the caller. To return from a subroutine you would just branch back to its beginning.
Subroutine calls are for the weak :)
There's some more detail here: https://ed-thelen.org/comp-hist/CRAY-1-HardRefMan/CRAY-1-HRM...
The following quote gives some sense of how "manual" this was:
> "On execution of the return jump instruction (007), register Boo is set to the next instruction parcel address (P) and a branch to an address specified by ijkm occurs. Upon receiving control, the called routine will conventionally save (Boo) so that the Boo register will be free for the called routine to initiate return jumps of its own. When a called routine wishes to return to its caller, it restores the saved address and executes a 005 instruction. This instruction, which is a branch to (Bjk), causes the address saved in Bjk to be entered into P as the address of the next instruction parcel to be executed."
Details were up to the compiler that produced the machine code.
Essentially, the B00 register is a Top Of (Return) Stack or TOS register. It’s great for leaf routines.
You have to push it to your preferred stack before the next operation. You do the cycle-counting to decide if it’s a good ISA for your implementation, or not.
Obviously, ISAs with a JSR that pushes to stack are always using an extra ALU cycle for the SP math, then a memory write.
Doing it with a (maybe costless) register transfer followed by (only sometimes) a stack PUSH can work out to the same number of cycles.
With improvements in memory speed or CPU speed, that decision can flip.
Consider that in this era, your ALU also had the job of incrementing the PC during an idle pipeline stage (maybe the instruction decode). Doing a SP increment for a PUSH might compete with that, so separating the two might make the pipeline more uniform. I don’t know any of the Cray ISAs so this is just a guess.
Am I right in thinking there are no working Cray-1s?
I know a couple of museums have them, but I don't think any software has ever surfaced, am I right?
The cooling fluids have been illegal to manufacture for a very long time. I wonder if there's still enough left in the world to keep one running?
No, both COS (but missing Cray Fortran) and Unicos have been located.
whenever i hear about a Cray, i'm always reminded of the scene in the movie Sneakers where they sit on one to have a conversation
I never realized; thanks for pointing that out. Sneakers is one of my favorite movies.
Documentation is fantastic.
How does a vector processor of yore compare to SIMD of the late 90s to the current day? Are they two words for the same thing?
Before IBM's deep blue, apparently there was a chess program called Cray Blitz, I'm sure the cray had the raw power(20MHz? But with 64 parallel operations) needed to beat the top humans, but the software just wasn't there yet.
Blitz was ported to C and continued development FOSS as Crafty, mainly by a U of Alabama professor, but to this day it can't beat top humans on modern cpu (topping at 2650 elo instead of 2850 of, say, carlsen.)
The latest version of Crafty has a significantly higher rating on CCRL than Fritz 10, the version that defeated Kramnik in 2006. He was the World Champion and was rated 2750 at the time. I do not know what source you used for Crafty’s rating but ratings from different lists are not comparable. It is highly probable that Crafty running on a Ryzen could defeat any human.
I am also of the opinion that with an optimised program the CRAY-1 would have been on par with Karpov and Fischer. I also think that Stockfish or some other strong program running on an original Pentium could be on par with Carlsen. I am not sure if Crafty’s licence would count as FOSS.
[dead]