Slashdot Mirror
End In Sight For Alpha
minektur writes "news.com has an article stating that DEC ... I mean Compaq .... Uh, I mean HP has decided to EOL the once mighty Alpha architecture. Let's all take a moment of silence." I was lucky enough to have access to a 533 MHz Alpha back when the fastest Pentiums were only around 200 MHz, and the Alpha architecture earned a special place in my heart. It will be missed.
Screw Itanium. An architecture that requires a highly-tuned compiler to optimize software well enough that it doesn't stutter from huge branch prediction misses is not a worthy successor to the Alpha. Excellent firmware, elegant architecture, good speed/MHz ratio...
I own three Alphas (a Personal Workstation 600, an Alphastation 255, and a homebuilt machine using a PC64+ motherboard), and they're great machines to use. I'm currently on the lookout for an ES40 - when I see one for below a couple of thousand dollars, I'm going to snap it up.
sure sounds the same
quality = more money, people. you are all free to drive Yugos if you please.
That aside, my father had an alpha for FreeBSD development, i believe. I used to play Larn on it. before i ever heard the word "final fantasy" Sorry to hear it got sucker punched. (Remember CEO's: If you're not number one, then buy number one and kill it. It's cheaper then investing in R&D.)
shucks.
CS majors know the time/space tradeoff, but they never get taught the 3rd, crucial, tradeoff of the set: comprehension!
They ran Linux quite nicely. Many Digital engineers had input into Linux
what about transmeta? But rather than running in (intel) emulation mode, how about some of you (or maybe me when I can afford the time and money) try and use the VLIW concept to the full. The idea of getting smallish code blocks to execute as single instructions must have some appeal to some fo you speed freaks.
The Singularity is closer than you think
Quant
I know they had the megahertz win in the early days against intel, but man I love my alphas. The architecture of the entire system that DEC built around them is just so nice. Maybe it's because I haven't played with any E10K or better hardware from Sun or whatever HP has to compete on that level, but I wouldn't give up my Compaq GS series hardware for anything. Not to mention Tru64 and TruCluster -- I swear Tru64 has the best man pages of any Unix - free or commercial. I often find myself going to the Tru64 pages for info on various standard syscalls.
I don't think it's a wise move for HP -- I wish Compaq had known what they were buying when they bought Digital.
VMS Engineers are well along in the port to Itaniac. The port will be relatively easy because of the portability built into VMS during the VAX->Alpha port and the fact that the memory management in Itaniac looks a lot like the VAX. This according to a VMS Engineer who spoke at a conference I attended in Nashua NH (where the VMS engineers are).
I think I had heard some rumors somewhere of a VMS/x86 port.
Won't this silly rumor ever die????
Well, I guess we can all move to Linux or Solaris.
Search first, ask questions later.
Wow, you have P4 Xeons that are 64-bit processors and can support 20gigs+ of memory? No? Then how can you call them more powerful than an EV6 Alpha?
I will miss the Alphas, they were doomed the moment that Intel got control of them.
Kindness is the language which the deaf can hear and the blind can see. - Mark Twain
i still have a alpha 533 myself. its a great piece of machinery. too bad i dont really have a use for it anymore and its just sitting in my closet collecting dust.
PS. if anyone wants to buy it off me email me at neonmatrix@homtail.com
So, why do they favour Itanium over proven hardware like PA-RISC and Alpha? Simple: software. Alpha's epitaph was written the day Microsoft decided to stop NT development for Alpha's. Without commercial software, what's good hardware worth?
My answer would be: A LOT! A few years ago I bought a (once expensive) 266MHz Alpha for about $300, without any software. It took a while to get Red Hat 6 running, but the machine really rocks! As most of us know, Linux per se does *not* require x86 hardware. I guess you could even go through the trouble of getting Wine to run Win32 binaries under Bochs, if performance is not your primary issue. However, in my daily usage I hardly ever need anything outside Linux. In those cases - when someone sends me a Word document - I use and old Toshiba laptop, running Mandrake.
So why is x86 hardware the de facto standard Linux hardware? Good: price/performance ratio. Why is x86 relatively cheap? Large sales volumes. Why so? Windows won't run on anything else. Why do people buy Windows? Because everyone does.
It's just the everlasting circle that won't be broken anytime soon. Not by better hardware (Alpha, PowerPC, MIPS, StrongARM) and not by better software (Linux, BSDs other Unices). It's so depressing...
--
Programming is like sex... make one mistake, and support it the rest of your life
"Can you effeciently farm out quake/UT/Morrowind/Maple/Matlab/etc to multiple processors?"
The answer is yes... sortof...
Quake/UT are 3d First Person Shooter games. This means that as far as the client side goes (Joe sixpack's computer) there is no immediate advantage. Most processing power is done on rendering in real time, which can't be done on more than one host (however, its still possible to do it on more than one video card, or multiple CPU systems in SMP mode).
However, on the Quake/UT server side, everything changes. Most of the multiplayer games are limited by 1) network bandwidth and 2) cpu power when it comes to scaling (limits usually around ~32 users). Battlefield 1942 (kindof like RTCW) is an excellent example of a game that could dramatically imporve by distributed computing. For one thing, that game's server is dramatically CPU limited. You can't get more than ~32 clients(out of a 64 person max) out of a server with acceptable results, even on a 100Mbit switched LAN. However, if you could distribute out the server, have maybe 2, 3, 20 servers, each in charge of 1/2, 1/3, 1/20th of the users in the game (sortof like IRC chat) then you could successfully scale to any # of users without having any CPU scaling issues. I think everquest might have a hint of this type of technology.
I don't know the answer about things like Maple or Matlab, but I'm pretty sure they could at least take advantage of distributed computing a little. (again, depends on how you use it)
Two infinite things: your stupidity and mine. But I'm not sure about the latter. If my sig offends you, I'm sorry.
The Alpha 21264 actually does beat the crap of any P4 in any real world app. Plus they do also have SMT (Simultaneous Multi-Threading), and they actually support 64 bit addressing, until then you can keep your P4 to run your latest winmark benchmark... I will keep my 21264 to actually do some real work.
Thank you very much... the don't call'em FeeCee's for nothing.
Looks like HP is changing from an engineering company to a bunch of vacuum cleaner salesmen, like say, Dell. They spun off their measurement systems stuff (Agilent), killing PA-RISC and alpha, what do they have left? Reselling Intel boxes and perhaps some consulting.
On the other hand, chip development costs seem to grow exponentially, so keeping on developing two high-end architectures for a very small market doesn't really make sense economically.
I started my PhD on an alpha 233 MHz, using jed to edit my c sources. Today, five years latter, while waiting to defend the thesis I'm installing debian on a au500 system, again using jed to edit the sources.
Now this news... It seems like the end of an era.
You are over reacting. 64-bit computing is useful to a very small percentage of the computing world, and those who need it are already have one. Furthermore, the applications that need more power than current top-of-the-line processors are well served from multiprocessor systems, clusters, etc. For example, scientific applications like nuclear reaction simulation, fluid dynamics computation and other similar things run in super computers. The Alpha CPU is a capable chip, do doubt, but it does not make a big difference in those applications that need such a big horsepower.
You are referring to advantages of the Alpha like SMT and 64-bit addressing. Xeons and Pentiums 4 HT already have SMT (hyper threading), and the Hammer from AMD is in the right track for a commercially viable 64-bit x86 chip.
There are also other 64-bit chips, like the Ultrasparc. Personally, as a user, small office/home office and developer of defense applications certainly don't need 64-bits.
I don't really see any advantage of the Alpha on the market. Sure, it is a good chip, and a technological marvel, but a dead one. In other words, it is not marketable any more(i.e., no one really needs it).
Now is the time for HP to show that it's not just going to waste all the time and effort everyone put into the Alpha.
The now have the opporunitity to publish all information regarding the Alpha to the community so that anyone who wants to can continue to provide support. Or, if anyone wanted to, produce their own Alpha based chips.
By allowing the continued use of the Alpha they could extend the life of these systems instead of killing them off in favour of newer systems. I know that they probably will not want to, but hey, it's a nice guesture to make.
I seem to recall something about "open" processors before, such as an open sparc or something, so this wouldn't be the first time it was done, just the first time that a big corporation allowed it to be done with their 'redundant' interlectual property. I also think that this would be good for preservation purposes and to have more information about micro-processors of our era for future generations. Just look at the mess that NASA have been in before when older components obsolete.
You've got to have a target audience first. Digital had been slipping for quite a while some due to its outragious support costs. When I was living in Norhtern Virgina in the 80's I had to get a DEC engineer on site for some hardware work and it cost me $5,000.00 just to have him dispatched! I'm a DEC user from the 780 cluster days who moved on to Burroughs, then Unisys, then HP/UX, then Sun back to HP andcurrently am on Sun. Software is always the determining factor on servers.
Oh ok this is a great idea--
Parallelize Joe Sixpack's mundane everyday computing tasks.
That way IE can format every hard drive in the company when it catches a malicious activeX control!
It will make your desktop support monkey's job more efficient and enjoyable, because every PC in a given environment will have the same crash at the same time every time!
Cole's Law: Thinly sliced cabbage
HP has been 'getting rid' of PA-RISC for ages. I remember attending an seminar on the virtues of Merced in the 1996-1997 timeperiod. At that time they planned phased out PA-RISC CPUs and were going to do PA emulation on Itanium by 2000 at the latest, to allow older HPUX installations to make a smooth transition.
In other words.. they'be been singing this song for at least 5 years and the Merced/cKinley delays have royally screwed their plans. They did have other plans on the horizon (though at the time I believe the roadmap only went to the 8500 or 8600 chips.. the 8700+ processors were probably a mad scramble when they realized it was going to be even longer.
-'fester
It kind of makes you wonder what would have happened if DEC had accepted Apple's offer to base PowerMacs on the alpha. Back when Apple was getting ready to leave the venerable M68040 series of chips, it approached DEC and said it wanted to make a deal with them to produce alphas for Macs. The CEO of DEC said no, because he wanted to focus the companies efforts on extending the life of VAX/VMS for one more generation, and getting involved with Apple would be a distraction. Of course, he was fired shortly there after for being a knucklehead, but by then it was too late. Apple had teamed up with Motorola and IBM to develop the PowerPC architecture. Still, you gotta wonder what would have happened if he had had a clue and played with Apple.
I'm glad I got my Alpha when I had the chance, a 21066 chip on a Digital AXPpci 33 motherboard. Sure, it's not one of the fastest ones out there, but I paid $150 for it and it works fine with RH 6.something.
One of these days, I want to snatch up one of several Alpha's on eBay and they have some really nice ones for not much money at all.
For example at this auction, with 5 hours left (at the time I'm writing this) you can get a dual 533MHz Alpha with everything you need for $520, install an OS and you're ready to go. I'd only want to exchange the 6 4.3GB drives for bigger capacities.
It's a pity DEC was even worse at marketing than IBM is. My assembly language book from college talks about how DEC introduced their 16 bit machine in 1976 and how a 32 bit machine may be created one day in the near future but that they'd be prohibitively expensive and never gain widespread popularity due to the price. DEC was comfortably ahead of everyone else for high performance mini-and-desktop computing, and they blew it. Their software always stuck me as very well thought out and easy to work with. There was only one problem -- all the managers were buying IBM. Oh well...
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
I mean it was really early. The environment was that we cross compiled/linked the image on a VAX, copy the image to dual mounted disk, dismount the disk, boot Alpha and mount the disk and run the image. No debugger. When the image crashed, I got register dump. Not even stack trace. Network to the box was Digital's LAT (Local Area Transport), so I used Xlib over that transport.
I think I spend a couple weeks there. In 2nd week, we got debugger, version X.0001 or something. When I finally got our library to start running some simple rendering test, the picture didn't look right. A squre cube looked very distorted. Run a quick test of trigonometry functions. Hmmm, sin() returning value bigger than 1.0 didn't look right. Was told that I was a first one to excersize floating point on their chip. It was fixed shortly and we got nice pictures drawn.
I was told that we were the first external customer to run code on Alpha. And of course, we were doing all that work on the only ture operating systerm on Alpha: VMS :-)
Another interesting but far less practical project I got involved later was to try out Digital's binary translator which translated DECstation (MIPS) Ultrix's binary image into Digital Unix (or was it still OSF/1) Alpha binary image. It was pretty impressive. It took our image, which was more than 40M on Ultrix (about 6million lines of PL/I, C and Fortran), and created image more than 80M of size. It was still maintaining whole MIPS image inside it because it has to interpret the MIPS binary in some complicated situation. I think it was for something like exception handling which our PL/I code heavily used. After they fixed the last problem regarding this exception handling, the translated image actually passed through our basic regression test suite. I was not involved but there was also VAX/VMS to Alpha/VMS binary translator, which we played with too. If I remember correctly, some VMS softwares on early Alpha/VMS were actually binary translated images. We never shipped anything using those translators (it is pretty much impossible to debug the translated image), but it was a interesting excersize.
Hiroto
Architecture-wise.
...and once somebody builds a cluster for the Intel, the "ideal" thing advantage is pretty much gone - if you choose to use a cheap Linux cluster framework (e.g. beowulf!) and don't insist on reinventing the wheel.
Making a distributed framework were easier on the Alphas than the Intels....but most the time you only do it once - you can say it is a one time overhead.
Alpha processors were more expensive, and suffers from most of the disadvantages a less popular product generally suffers (less support, less programmers, etc) compared to Intel and AMD.
That's why when the x86's finally catches up with parallelization, Alpha is destined to go away...
I can't say I miss them, cuz they're still here! DEC (Sorry, they perfer to be called Digital Equipment Corperation or Digital I think.) really did they're job too well in developing these guys.
However, HP's being smart. Since the research is being done with other architechtures, it's best to follow suite.
I fell in love with these boxes upon the first boot...Having "more" and "cat" in the bios just rocked my world. I wish others would have taken the hint!
-=fshalor
As proud owner of several alphastations (200 4/166 (FreeBSD, NAT), 255 4/300 (Linux, MTA), 500/333 (OpenBSD, Firewall)) and servers (As 800 5/500 (FreeBSD, Web Server), As 2100A 5/375 (FreeBSD, Web Server), i cannot help but lament the fate of my trusty friend. Apart from an excellent processor, the systems it was built into were (mostly) masterpieces of engineering, with well balanced memory buses, quality peripherals, good chassises, etc. My way of saying goodbye will be one more purchase (Used ES40)
Besides, did you know that the firmware had an x86 emulator just to be able to boot up x86 PCI cards, i used to have a Creative Labs Permedia2 video card that, when booting, showed an animated logo. Accurately enough, the logo also appeared on the alphastation, as the card was booted of the firmware emulator.
Regards
Roberto at spock dot cl
Because HP is already committed to IPF (IA-64, Itanium) and thinks it will become better than what Alpha could be. That's the official story. The unofficial is that they gave up competing on products a long time ago, and now just want to do services. Presumably that's a mixture of dumb MBAs who repeat the "services are key" mantra without understanding you must have a product to service, and if you didn't manufactured it, its manufacturer is more likely to service it than you; and of realising that people want Wintel or Lintel "just because they are used to it", and because of volume driving down prices.
In other words, the RISC market was too fragmented, and instead of coordinating the RISCs by phasing out PA-RISC and going all the way Alpha they decided to instead try to sell Intel on their VLIW, and thus IPF was born. To be fair, the first blunder was Digital's when they failed to win Apple and Novell as Alpha users, then consequently to make Alpha a volume architecture with several licensees, OEMs, foundries, notebook versions and all that you need to go heads on against a monopoly.
First, they do believe in IPF, or so it seems.
Second, is being in the top 500 supercomputers list important at all? I guess they'd rather be lucrative. I think the way they chose to be lucrative is mistaken, but that's probably their rationale.
Actually the PA-RISC has a nice position. Telcos tend to use predominantly the SuperDomes, due to HP's relationship with Amdocs. PA-RISCs are actually nice systems, and HP builds some nice systems around them. HP-UX isn't GNU/Linux or Solaris, but still it's Unix, so you can't throw it away. Too bad for them that Unisys will sell IPF machines that will be as nice as HP's, and so will other vendors, and some of them will have GNU/Linux or Unix to run on them.
There isn't, see below.
I doubt. Intel bought the patents and the documents, but most engineers left. Intel has lousy employee relationship, so they wouldn't be able to reproduce the in-house expertise Digital, Silicon Graphics, HP (before merge) had and that IBM, Sun now have. Also, they are already forcing customers to change the architecture. Would they risk it again, knowing each change in architecture is a chance of jumping ship to someone else with a better story to tell, like IBM or Sun?
First, there is no one else involved, only Samsung.
Second, Samsung can't compete. It does not have neither the focus, nor the ISVs, nor the customers, nor the applications, nor the systems, nor nothing needed to compete. Sun & IBM do, HP, Digital and Silicon Graphics had.
I doubt. Technically yes, but where are the volumes, the customers, the profits? Anyway they already jumped ship. They are now SiPackets, former API Networks, selling the HyperTransport stuff to AMD and the like.
Forget it. Licenses are not available for the asking, even if you had loads of money. And you would have to get the engineers, and find the customers. Do you think anyone would, after the .com bubble?
Sun has already stated SPARC for them is binary compatibility and a viable future, not performance only. Their going Alpha would hurt more than help. They hope to get UltraSPARC to be competitive with POWER and IPF, and that's it.
There is a difference. There was never MS Office running on the Alpha, only MS Word and Excel, and these are gone now. There was never an Alpha notebook. Alphas and Macs were never in the same price bracket.
Leandro Guimarães Faria Corcete DUTRA
DA, DBA, SysAdmin, Data Modeller
GNU Project, Debian GNU/Lin
"I'd think HP would keep it around just so IBM doesn't take over the top spots for supercomputers."
After studying a number of processors indepthly and doing a research project with the Alpha, I came to the realization that the Alpha engineers were truly the best in the world. (My next favorites were the highly-scalable SPARC by Sun, and the PowerPC, another very smart RISC processor by IBM, Motorola, and Apple.) I had also read a book about the pains the Alpha engineers went through to design a processor so far ahead of its time. For some reason, however, DEC couldn't stay afloat and the company exchanged hands a number of times, and, presummably, lost a lot of their engineers. Compaq inherits the company, and merges with HP. Funny thing is HP was already in bed with Intel and helped design the 64-bit Itanium. Now, even though Alpha had gone 64-bit since around 1995, it doesn't make sense for HP to compete with Intel using the Alpha, after all its efforts to help Intel create the Itanium and monopolize the 64-bit market.
When something like this happens, there is always guaranteed to be a fall guy. Before you know it, Alpha, the most outstanding processor in the world, is history.
Intel bought the technology, so if their new 64-bit processor (which shatters compatibility anyhow) doesn't perform well enough, they could just start making Alphas and call them their own.
I thought that Intel bought the manufacturing arm and licensed some patented technologies (whatever DEC sued Intel over to begin with) and the right to manufacture Alphas, but not the whole intellectual rights to the Alpha architecture itself (which Compaq got)? So while Intel could certainly churn out Alphas, they could only churn out existing versions and not create new ones?
Or perhaps some other party might pick up the torch. Sun would be a good candidate, since they're in a tight competition with IBM, and the Alpha seems to be the only thing to top IBM's Power3 (and is doing so with half the number of processors!!!).
Egads, Sun would never abandon Sparc. They have spent billions on just simply developing the name in the marketplace, and to suddenly switch gears and drop Sparc to sell Alpha would be suicide. Most people who purchase Sun don't do it because their stuff is faster than anyone elses (because in general they are not), they buy it for the stability of the hardware and OS. Sun has been able to thrive even they've always been in the role of the lessor performer. Note that there aren't too many Sun's in the Top500, Sun just isn't that interested in that market.
Come on HP. The Alpha has just as loyal a following as Apple... It's a big mistake not to start improving it and seeing what it can really do for you.
No, it's a big mistake to try to sell computers using three different architectures (four if you count the overlap between PA-RISC and Itanium). It makes no sense at all to keep Alpha around (as much as I like Alpha). They've already bought into Itanium and PARISC still has legs while they wait for Itanium to mature. Now they can surely integrate more concepts from Alpha into future chips, but Alpha as an independent entity has no useful purpose in the HP landscape.
Maybe Transmeta will buy the rights and finally get a little oomph into those chips of theirs.