Slashdot Mirror
1.6 GHz Alpha With Transputer Features Coming?
GFD writes "The Register has a story about a 1.6ghz alpha with 8 paralell rambus channels (8oomhz) and a transputer like channel to 4 other cpus (10ghz), integrated memory controller and l2 cache. Science fiction? Maybe, but oh my God what specs!!"
How does the source of news affect whether a computer is a tool or not? They seem to be two completely debatable yet unrelated subjects.
I'm a big fan of the register, and in my experience, they seem to hit the nail on the head so much more often than just about any other news site. Problem is, they generally speculate about events that haven't been announced, so it sometimes takes 3 or 6 months for stories they've reported on to actually make sense. Which they do, 75 or 80% of the time.
So far as Alpha goes, ESPECIALLY if they're reporting it, I'll believe it to be a distinct possibility. If MSFT had said the same thing, I'd say it's marketing schpeel... But they're (MSFT) aren't exactly un-biased... (guess the register isn't either, though they're biased against liars and spin-meisters)
21264 Processor with 2MB or cache
64 Megs of RAM
10GB HD
Dual 10/100BT Network card
High performance memory system (1.6GB/s)
all this goes foryou guessed it $3500.This is the AlphaServerDS10 at www.dcginc.comI don't know if you can still link them together with the inmos links, so the parallelism might be out the window. The low power and good code density make them useful for embedded applications, though.
Registering accounts later than some other chrisb since 1997
Transputer had problems too, sure the hardware side was really neat for their times, but the software sucked.
OCCAMM was a good language but parallel programming is hard.
And their C compiler was buggy according to a friend of mine who had to program these things.
Goo hardware, poor software --> useless.
And then, they took a loooonnng time to build a new generation of CPU (I don't remenber the number)...
Poor C compiler + vaporware CPU equal bankrupcy, no matter how good the architecture is.
FX32 is or will be useless for Win32 apps... remember - MSFT and Compaq pulled the plug on future NT development for the Alpha... Remember?
Now, I wonder if Compaq will (if it's their's to do so with) release FX32 source? That'd make the idea of an Alpha or PowerPC box running linux much more palatable.
Of course, even accounting for cache, the prices are artificially high, just because that's the price people are willing to pay. Profit margins on those chips are significantly higher than on commodity CPUs.
"The invisible and the non-existent look very much alike." -- Delos B. McKown
Interesting to see transputers mentioned, I cut my parallel processing teeth on them programming in Occam. Peter Welch was my professor back at University of Kent and I have been a fan of the processor ever since. Funnily enough, I am now a fan of Alphas since I got into 64 bit microparallelism, and equally a fan of Linux (currently running a 600MHz 21164 LX system at home which is very slick). Nice to see the whole lot coming full circle so to speak. I had to laugh back when Thinking Machines claimed to have invented MIMD with the CM-5 since I had been using a MEiKO computing surface with 80 transputers just a couple of years before and that was way more MIMD than the CM-5 ever was. No wonder TM when under, filling their box with sparc processors, pah! Mind you, cray filled theirs with Alphas and didn't do much better. Occam was (and still is) a fantastic language for writing pure parallel apps, the parallelism in implementations of C were always grafted on, Occam had it built in from the ground up. The US seemed to be going its own way by trying to write compilers which would derive their parallelism from serial code, never going to work. Peter was adamant the only way to do real MIMD was to ditch the "dusty decks" of old fortran code and re-implement using a real parallel programming language. Last I heard Peter was still at University of Kent in Canterbury and still doing some great work. -- Shane Sturrock, Edinburgh Biocomputing Software
1.6 Gh mmmm that is getting a bit dangerous, hope they shield it well, or you either be cooked from the inside (du to the microwave radiation) or become impotent cause them radio waves will kill ya sperms
Admittedly I was in the last year to do the MSc Parallel Systems and therefore the end of the transputer as a teaching aid in university, but I was called into a couple of jobs where they use them. Ive seen some T4's being used in a brand of video phones to do some image work. I also like alot of the features offered by OCCAM, specially the way code replication occurred for the "worker" processors. Very scalable. But christ I hope TDS doesnt come back ever! Transputers aint that dead, and hopefully their good features will resurface and I can get back to some decent parallel work. Brad
T9 I think. Thing melted. Could only be run at a few MHZ "stably" I was told.
T9 I think. Thing melted. Could only be run at a few MHZ "stably" I was told. Brad
Is speed all that matters to you right now? How about those nice transputer like links between cpu's. A great old idea, dumped in favour of crap shared bus's. The speed would be nice, but Im more excited a potentially better architecture might be re awakened. Brad
I use my tool for entertainment all the time ;-) :)
The use of my tool is certainly one of my longest hobbies.
Some people say that I am addicted to playing with my tool.
They're probably right; but who cares?
I better be careful with my tool; i wouldnt want to pick up any viruses through its use.
And in thirty years, my tool will be obsolete
I think the only difference between one 'tool' and another is that with computers, smaller IS better!
Three Step Plan:
1. Take over the world.
2. Get a lot of cookies.
3. Eat the cookies.
Well a decent Intel Mobo with features comparable to an Alpha MB (64-bit PCI, UW SCSI) and SMP support (to keep it's performance comparable to single processor Alpha System) will probably run you $400 - $600 dollars, then you pay x2 or x2.5 for the chip itself, but only one rather than two PIII 600's, and you're pretty much even in terms of cost, but you've got a much faster system in the Alpha - supposing the apps you want to use are available (I know, it's just a recompile, but I swear - plenty of programs exist that are I deem essential that are also commercial, proprietary, closed, whatever you want to call it)
You really get what you pay for. If you don't need it, don't shell out the money. If you need it, then you can probably justify the outlay
Your post is approaching the annoyance level of "beowulf - drool!" posts.
The artical said it was going to use the EV7 bus, and I belive the k7 uses the EV6 bus. so, no they won't be compatable anymore then a pentium and a pentium II
"Subtle mind control? Why do all these HTML buttons say 'Submit' ?"
ReadThe ReflectionEngine, a cyberpunk style n
pleese get one
"Subtle mind control? Why do all these HTML buttons say 'Submit' ?"
ReadThe ReflectionEngine, a cyberpunk style n
Transputer boards were available for many different types of personal computer, including IBM-PC and compatibles, Macs, Amiga, and Atari. Most transputers ended up embedded in things like laser printers though, allegedly, and quite a few went into fancy research machines, hundreds and thousands at a time.
:-)
I've got one in my junk box somewhere. I should embed it in acrylic and turn it into a topical paperweight.
"The question of whether machines can think is no more interesting than [] whether submarines can swim" - Dijkstra
Do you realize how much this beast is going to cost? No doubt it will be priced well out of the reach of those who would _want_ to play Quake.
Even with all that speed, I guarantee that OpenGL performance will be lacking. Reason? No hardware accelerated OpenGL drivers for Alpha that I know of. And the glx driver for TNT/Matrox depends on x86-compatible assembly instructions, so that won't work.
Glückwünsche, haben Sie Slashdot ermordet, indem Sie zum korporativen Druck beugten und Subskriptionen einlei
Alpha 21364: A Scalable Single-chip SMP
Alpha OEM
--The more you know, the less you know.
But we didn't get CmdrTaco saying "I want one"
Something is just missing
But will it come in colors? :)
I'd run Tru64 on it if I had one, its MUCH faster than linux on alpha..... (sorry)
If linux runs on them, AND you don't have to sell your home to get one, then who needs intel/NT/MS, etc? If the two above conditions are met, I bet you one of these machines, that they will get 50+% of the market before you can say 'What the heck...'. nvon
Well, the computer scientists I worked with at Oxford (you may have heard of it) Computer Laboratory were hardly struggling. And some very surprising things turn out to be more easily paralellised than at first seems likely. Just adding two numbers together seems hard at first. The carry from the least significant digits seems to be necessary before adding the next digits together. But it is not so. I have seen the PixelFusion chip in action in a graphics context, doing real time shading, texture mapping and bump mapping. Put a few of these in a compute engine, and see it go!
Are you kidding? PixelFusion is a company making 3D graphics chips, not general purpose microprocessors.
Big difference. Parallel graphics logic circuits do not constitute a "processor." Many logic circuits computing in parallel is "parallel processing" but is not necessarily "hundreds of processors."
--LP
As an undergrad I worked in a computer vision lab in 94/95. They used custom designed boards and transputers for vision recognition - including controlling/driving a golf cart!
They were great at the time, but programming them was a bit weird- Parallel C or some such.
The way the lab designed the boards, you could rack mount them and just link 'em with cables. you could assign a task to a single board or a group of boards. Not fast enough? Slot in another board! of course it wasn't as simple as this all the time, but it wasn't far from it.
AFAIK Rolls Royce (aero engines, not cars) used a similar system to spot defects in x-rays of engine parts.
What killed it eventually I believe was the lack of development caused by mismanagemnt and repeated selling of the technology/company to other companies. Everyone knew it was a great product, but no-one knew what to do with it. Sad.
The thing will seriously rock for large (and very expensive) servers, but will probably not be of interest to the general linux community for a few years. First systems for this probably won't beat Merced to market, but will certainly not be far behind.
At any rate, unless you're an IT guy for someone who needs some serious horsepower, don't get your hopes up anytime soon.
It is fun to drool, though...
The 6.4 GB/sec memory bandwidth that feeds all this data into the CPU is 16x faster than Apple G4 systems' current 400 MB/sec and 8x faster than Apples "fast" 800 MB/sec motherboard coming out in a month or so. Fast memory bandwidth (and I/O bandwidth) is the difference between "supercomputer marketing" and a real supercomputer.
The G4's flops are totally bogus because you can't feed the CPUs at those rates, at least for reasonably sized datasets that overflow the L2 cache.
This is all "apples and oranges" comparisons of today's product versus futures though, and I'd agree that switching to Alpha would be a dumb move for Apple or Apple users.
--LP
http://www.totalimpact.com/ "IBM's reference design allows us the ability to integrate our multiprocessor technology into the motherboard design as well as add support for the G4 microprocessor" said Brad Nizdil, Total Impact's president. "We have the multiprocessor expertise and are currently running a 128 processor system under the MacOS"
Uhm, everyone saying "Apple would never do that" seems kind of, well weird, for Slashdot. Whatever happened to doing anything? Darwin is Open Source (despite you guys' bitching), so it wouldn't be impossible to port that yourself. Also, if I remember correctly, having a microkernel (Mach) would allow for very simple porting, most of the work (all?) would be porting the kernel.
As to why, why the hell does anyone port Linux to palmtops and BSD to the Dreamcast?
We'll call it the i-DEC :-)
I browse with my threshold at 2 so I can't read my own comments :-)
Don't believe a thing that The Register reports. They lie so much.
Can you say "Baywolf cluster?"
well, but since neither Compaq or M$ supports the Alpha-NT installation, who cares?
But Linux on Alpha hardware....
I have it: we'll call it the OMFG!_BOX
I mean, Alpha is rather sweet and all, but I wanna run a few games in my spare time and I'm not sure if that fx32 thing is gonna be speedy enough at 1.6 ghz to run MS Solitaire 2000.. but maybe a K7.
Uhm anyone seen my medicine
Athlon (K7) uses the EV6 bus, while 21364 CPUs and motherboards use EV7.
Calm down.
MacOS X (aka Rhapsody, Son Of NeXT/OpenStep) running on one of those puppies...
Comparatively, the 64-bit port to another RISC should be pretty easy... right...? Please say yes... please...
*drool* *drool*
Heh, and then I woke up... If it was ever going to happen, though, I might sell my car to get my greedy hands on one...
-- "so let us not talk falsely now / the hour is getting late"
I run off of a dual 21264 (500 MHz? I forget might be 600), right now and it flies. It would be great to see some more smoke from these cpu's. But I it would be great to get some faster, cheaper hard drives to keep up with the demands of these processors. My jobs generate lots of data, and I find that writing it takes away from the cpu's a bit.
-- Moondog
Whatever a "Baywolf" Cluster is... heheheh
In Capitalist America, bank robs you!
The current Alphas dissipate enough heat to cook food. Will this thing produce enough to start nuclear fusion? Will I need a liquid nitrogen cooling system?
You don't exist. Go away. --SysVinit Halt
hot it should be very hot!!!
PixelFusion are about to bring out a chip that should make all these single processor things a bit more obsolete. Arrays of hundreds of processors, embedded within large tracts of RAM.
The original Doctor Dark.
I've always thought the alphas were "neat". Never had a chance to invest in them or really work on them, but I've always been very impressed. Which is why I'm a bit sad to see this news.
... and from what this humble /.er can see (granted I have been before and I could be wrong), that doesn't seem to be changing.
1.6gHz? Great. I really mean that. But with NT support dropped, how much of a mainstream market is there going to be?Or, more precisely, how much of a market does the architecture have now (AXPlinux , Digital Unix, NT), and how much do you think this will really change it?
They may be on the "cutting edge" (or maybe not), but without the right PR, they're going to continue to mire in obscurity.
Now, the alphazealots out there will rise up and shout that it doesn't mire in obscurity as it is,etc etc... Don't get me wrong, I'd love to see it take a bigger share of the intel market, but it just hasn't happened up until now
Yes, not only that - but when the added an FPU to the transputer, they wrote the thing in Occam first and essentially proved the design.
Of course, I alway thought that one of the barriers to entry in the US Market was that initially one had to program them in Occam, a language containing parallel processing verbs, etc. [ Occam, as a recall, didn't have dynamic memory allocation, either... best to look around the house and find those Transputer/Occam manuals
I squirrelled away. ] Of course, Inmos eventually offered a C compiler, but I suspect it was too late.
Sidenote:
I remember meeting Welch at the 3rd Conference on Hypercube Concurrent Computers in 1989; he
did a couple of presentations on Occam and Transputers. I wonder what he's up to?
Additional:
At the above conference, the "Show Floor" had a number of vendors showing off hardware. One vendor had one of the first PC board add-in cards with 4 Transputers on it. My partner-in-crime and I were checking out the demo - the (now tired) generation of the Mandlebrot set. We had generated the same using an early model of the Intel Hypercube - 10 minutes per run.
We dutifully watched the screen fill with the image and then asked the magic question, "So, how long does it take you guys to generate that image?" They looked confused, and then, realizing our question, replied, "We just did." I.e. the image (set) was being generated in about 5 seconds or so.
That was one of only two times I have ever been truly shocked at the speed of a machine, the other being when I first watched a 486/66 boot Unix.
The Norton Anthology of English Literature, 4th Ed., Vol 2
If we simplify thing's a bit (ho hum), they where like lego, if you needed more power, you just added another chip (and a little bit of memory) They had a serial interface to each other that allowed them to exchange information between each other AUTOMATICALLY, ie you did not have to take that into account when programming (at the least not under occam or TAOS, but I seem to remember that helios was different in those respects). They where popular in the Robotics and other embedded industrys that needed lots of computing power (the T800 was in it's time the most powerful microchip esp. in floating point, where Intel and motorola lagged). The fact that if a problem needed more processing power than originally thougt, you just added another card of 4 transputers or so to the frame, simplified controller design a lot. I remember a sheep shearing robot that used transputers :=) SHARC's where inspired by Transputers, for one thing. TAOS, possibly the most interesting OS of them all, was developed for the transputer, but talking about advanced technology like that to slashdot.borg's that think a low graniluarity OS, built on 60' technology like linux is cool seems like a wase of time. P.S. Helios was a CDOS derivative like the Amiga OS, not a unix clone like many thought, it did have Unix compability libs and programs that where used a lot, but these where slow and big.
Also, one of the barriers to the US market, was that the DOD refused to use products based around transputer technology, since it was british. The DOD may not matter much now, but it was very important to the chip industry in the 80's.
For all you guys complaining about heat, (or just making jokes). Try running a PII 300 at 600Mhz and see how much power it burns.! Why? because the PII 300 and the Alpha 21264 are made in the same process .35u. Even an alpha 21164 at 667Mhz only burns like 50W. thats about the same as a 650Mhz Athlon.But the Alpha is made on a .35u process while the athlon is on a .25u.So while the current mass wisdom is that Alpha's are power hogs, if you look closely you will notice that they burn about the same amount of power as other cpu's made in the same process. And better yet they are 64bits. while the rest are 32bit. Which means more transistors for the alpha and hence it should naturally burn more power.
Get yourself some solid-state hard drives. Clears that problem right up.
If these do get very popular.. also expect prices to drop in Alpha prices...
-
ping -f 255.255.255.255 # if only
PixelFusion is making general purpose chips and will be using them on graphics cards. The chips contain an array of processors, each with some memory they can work on. The individual processors were described to me as "about as complex as a Z80, but running at about 100MHz". Do you think they tricked me?
The original Doctor Dark.
Awesome!
What is so great about transputers? I thought they were just a kind of microproccesors.
Is that something that happens when a sports bra is on too tight?
--
Well, according to "Great Microprocessors of the Past and Present" this is just the beginning. The Alpha was designed with this in mind from the start, "Alpha was designed for the future for a 1000-fold eventual increase in performance (10X by clock rate, 10X by superscalar execution, and 10X by multiprocessing)." Check the Alpha section: http://www.cs.uregina.ca/~bayko/cpu.html#Sec5Part4 I'd recommend reading the page in general, LOTS of great CPU info: http://www.cs.uregina.ca/~bayko/cpu.html
Check out the old "Amazing Computing"'s. Amgia was duing transputing back in 86
http://www.microprocessors.co.uk/roadmap.pdf
21364 and 21464 Infohttp://www.microprocessors.co.uk/futures.zip
Register Article with the two links above.http://www.theregister.co.uk/990321-000011.html
The futures zip file is hard to download but worth it.Check out the Alpha site.
You can read the slides for yourself, but all that needs to be said is an estimated SPECint of 75 and a SPECfp of 120 (the best now is like, 30 and 60).
At those rates, who needs hardware acceleration for Quake?!
Glückwünsche, haben Sie Slashdot ermordet, indem Sie zum korporativen Druck beugten und Subskriptionen einlei
What has the world come to. When a machine with those specs carries a comment like it should make a great quake platform. All I have to say to that is. Pack up your tool kits. Magnetize your work. The world isnt worth developing for if QUAKE is all that the work comes to. The architecture is nice. The Target Audience is misguided.
Transputers had at least 3 features that made them so far ahead of their time that they died through lack of applications. Or maybe it was just because the Brits and the French are lousy at bringing things to market. If only Intel had bought out Inmos ...
Anyway, getting back to the features:
- A process scheduler implemented in hardware (an outer loop outside the usual inner instruction fetch and execute loop), which allowed transputers to implement concurrency with very fine granularity because the context switch time was exceptionally low. (And the process scheduler was directly driven by I/O events at the transputer links, below.)
- Four high-speed serial "Inmos links" on-chip through which the transputer could be linked directly to other transputers and to other peripherals without further glue logic, so that building multiprocessors was very inexpensive and scaled linearly. Furthermore, these links ran not only extremely quickly (for their time) despite being serial, but more importantly they ran under DMA power all simultaneously and at the same time as the processor was doing its own thing independently.
- The above two features made the transputer exceptional for multiprocessing, but I think its instruction set was also far, far ahead of its time: not only ultra-RISC, but highly extensible too. For example, numeric literals in instructions were only as long as needed, because an extension bit would (if present) indicate that more bits were to follow if needed. This made code *extremely* tight. The scheme also allowed extensions to the instruction set to be made in a fully backward-compatible manner.
The transputer was ultra-cool, and the world hasn't seen anything like it since. No doubt somebody will reinvent this approach some day, but probably in a US or Japanese lab, as usual, and they'll take the credit for exceptional design ideas made in an earlier age. Sigh.
"The question of whether machines can think is no more interesting than [] whether submarines can swim" - Dijkstra
I am starting to get very pissed off with all these "20 billion terrahertz quad mega-processor" crap. I have not seen any of this cool stuff, and until it's out and I see one, as far as I care, it doesn't exist. ARRGG. (Score:0, Redundant)
Your Momma's so fat she makes emacs look like nano!
Does anyone know if the current Alpha Linux will run on it? I might be tempted to hold my hardware upgrades off a little while longer...
Ummm, that's 1.6GHz, 800MHz and 10GB/s.
Agreed though, what specs!
Why bother?
The peak rate for this future 1600 MHz chip is only 6.4 gigaflops with its "new SIMD 3D instructions." The current 500 MHz G4 chip peaks at 4 gigaflops with single-precision Altivec SIMD operations. That means that this future Alpha only has a 8:5 gigaflop advantage over the current PPC rather than the 16:5 advantage you would expect from the higher chip clock rate alone, plus the higher memory bus clock rate than the PC-100 RAM that the PowerMac G4s use. If this is as good as it gets compared to now, then how will it compare to higher clocked PPCs also coming in the future?
While it'll beat the snot out any future x86 chips for awhile, it's a far sight from being worth porting away from PPC, which has a bright future ahead of it. Expect Apple to stick with the PPC. It is easily comparable in performance (though still edged out by Alpha on non-vectorized integer and FP operations), but it beats the heck out of Alpha on price, heat, and power consumption -- which is key for notebooks and low-end consumer machines. Also, the hassles of binary incompatibility makes it by far not worth the trouble.
If it's for-profit but free, you're not the customer -- you're the product (e.g., the Slashdot Beta's "audience").
The problem with PixelFusion and the other massively parallel computers like the FPGA based reprogrammable designs is that not all (or in fact most) computer problems can easily by computed in a parallel fashion.
This is a classic of computer science. Step B requires step A to finish. Step C requires step B to finish, and step D requires step A to finish. While steps B & D can be executed in parallel since they only require step A to be finished before being done, A-B-C MUST be executed in sequence. These designs, which are much slower at sequential computation than conventional computer designs would not handle these problems as well. This is why PixelFuzion's promises are not going to hold up well. Computer scientists have struggled with the problem of parallelization of code for decades with little radical progress.
If it's for-profit but free, you're not the customer -- you're the product (e.g., the Slashdot Beta's "audience").
As Homer Simpson would say
"Mmmmmmmm...1.6 Gigahertz."
Apple would never do that in a million years. Independent developers w/ Darwin OTOH....
... neither MacOS (8/9/X) nor WinNT will run on these CPUs.
Like Jobs would *ever* allow MacOS X to run on anything _BUT_ (oh-so-superior-PPC) Apple hardware anyways... :-P
~AC and Linux/Alpha 21164A user (and ex-Mac user)
Well, according to slashdot folklore: I wonder how this thing would perform in a beowulf cluster....
Yeah, but how much will they cost (with motherboard)? Sure it's nice to have a CPU that's 3X as fast as an Intel top-of-the-line one and it costs about the same, but those Alphas (along with *almost* all the rest of the RISC CPUs [e.g., PA-RISC, UltraSPARC, MIPS) are RIDICULOSLY OVERPRICED!!! And the frickin' motherboards are $500 minimum!