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.
The Alpha was always one of the better processors. It was fast and powerful and way ahead of its time. It is a shame that a truly great processor was killed by the economy and mergers galore. It will be missed.
I think this goes to show that it's not just about building a better mousetrap. You have to build a better mousetrap and then show everyone that it's SO much better than what is out there that it is worth the transition costs. It's something they teach in engineering 101, and it's the same problem microsoft has been bumping into for years now (and basically arm-twisting everyone to upgrade)
To make laws that man cannot, and will not obey, serves to bring all law into contempt.
--E.C. Stanton
I still have a 125Mhz prototype and two 150Mhz ones. It's sad to see such a good architecture end.
Distributed computing is going to be the trend...if I can stack together a few cheap chips to rival a single high performance chip, what would I do?
Given the exponential relationship of price to performance (i.e. a marginal performance increase will cost you a LOT more) associated with processors, I'll take the cheaper approach.
Granted, many apps don't fully use distributed processing power, but the ones that need most CPU probably do.
the high end server market is notorious for being slow to adopt change. Now HP is trying to get rid
of not one but TWO legacy architecture in favor of the unproven itanium.. Though both PA-RISC and
Alpha were niche players, they were highly regarded in their market. Maybe I'm just being a cynic, but
somehow I got a feeling Carly is pulling a SGI and migrating to a platform because everybody ELSE
thought it was a good idea..
Though I'm a big Sun box fan, I still have to give the proper respect for those two well regarded
chips. RIP PA-RISC, Alpha..
-- I have enough stupid gadgets to know that I can do without -- http://www.modestneeds.org
Samsung, but maybe they've stopped by now.
Granted, many apps don't fully use distributed processing power, but the ones that need most CPU probably do.
Even after considering the extra money spent to develop an app to scale well with parallel processing, the savings from using multiple "cheap" processors compared to one expensive high performance processor will still make it worth it. Not only that, but then you have an app that you can scale up as needed (assuming it was designed well) without having to purchase a whole new set of hardware, but rather just by adding a couple more processors to your current cluster.
you come here for new joke? All you could find here are different versions of same beowulf/southpark/russia/simpson/star war jokes. :)
HP will live to regret this. They're retiring a mature, stable, established and best-of-breed architecture (Alpha) for an unproven, late, incompatible, expensive, clumsy one (itanium). Their competitors must be laughing all the way to the bank. Just what is HP doing? Why do large corporations make such crazy decisions?
Stick Men
HP-UX already runs on Itanium, they sell IA-64 HPUX boxes.
Memory bandwidth varied, depending on the model. There were some with decent bandwidth, if you paid for it. Just like if you pay for it with, say, Sun. The cheaper Alphas were pretty badly crippled with both a slow bus and small L2, though...
Ultrix was also MIPS. And only MIPS. It ran on the DEC RISCStations, etc. but was never ported to the Alpha. Unix on the Alpha was OSF/1, Digital Unix or Tru64, depending on week, direction of wind, etc. And it's very different from the MACH-based Tru64/whatever-its-called-this-week.
Ultrix itself was an odd mish-mash of SysV and BSD. It wasn't actually all that bad.
As for VMS, people love it because it's rock-solid, very secure and has excellent documentation both online and in dead-tree format. It makes unix look like a toy OS in many respects, quite frankly. And I don't particularly like it - too bloody verbose - but I can certainly see why it's popular in certain circles.
As for the instruction size, it was fixed at 32bits. A far cry from 256 bits. Int being 64bits makes sense on a 64bit processor. And it's not like you can't define your own types if it really bothers you...
As for nobody buying them... um... there's a hell of a lot of them out there. I own one. I've used many. They were just another victim of the desktop PC getting powerful enough to handle work that until then had been the realm of very expensive workstations. So were SGI. So were Sun, to a lesser degree.
Digital bet their shirt on 64bit computing being their big selling point, but only those who actually needed it were willing to pay the premiums. And very few need it.
Few apps have specifically been optimized for it yet...make apps more multithreaded
This is a reccurring story in the development of parrallelism. It would be great in any form if people just developed for it, but even multithreading is quite tricky to implement compared to a single stream execution environment. And in most apps you just can't get the fine-grained parrallelism that would yield really good speed improvements.
This is a software problem, and no amount of hardware will make a significant difference.
Reliable, Great Value Hosting: $7.95/mo 2.4G/120G
It seems that HP caught compaqague. A terrible and deadly virus that destroys all the brain zones related to innovation, risk, calculation and self-estimation. Earlier we saw several companies being caught by this epidemics, the most notable, DEC, where the virus spread with such furor that in a question of months a once well-known company turned into another department in the corner of the company.
The fact that HP dropped a lot of its support for open source, closed the production of the Alpha architecture and seems to scale down other important sectors are a clear show that the desease got deeply into the corporation ranks. Soon we probably will see turning from blue to red and naming itself Compaq.
Depressing is right. Can you imagine what kind of chips we'd be seeing if Intel's ungodly amount of financial and engineering resources were being poured into something like the Alpha rather than kludging and hacking the x86 generation after generation?
It is one of my peeves that CPU architectural superiority means little in a world where x86 is the "default", and the negative feedback loop (Intel is cheap -> people buy it -> Intel is cheaper) seems to have no end in sight.
That and the fact that Intel can use its x86 cash cow to keep funding the Itanium whether or not it has any real merit. Not saying that it doesn't (EPIC IS a cool idea), but in a level playing field, do you think they can get away with just throwing so many transistors at the problem?
As various promising architectures die off (Alpha, PA-RISC, who's next? POWER?), in the end, was the computing community better served by the dominance of one architecture designed for the lowest common denominator? It's all speculation, sure, but I think not.
-- "This world is a comedy to those who think, a tragedy to those who feel."
What does a 533 MHz Alpha have to do with a 200 MHz Pentium? You are comparing clock speed of totally different architectures. What one can do in one clock step can be much more (or much less) than what the other can. I'm not saying that Pentium are, or where, faster than Alphas. I've heard that they were really good but, you can't compare CPU performance just by comparing the clock speed. Actually I have a couple of questions. The Alphas are RISC CPUs so, what clock speed do they have to run at to top the Pentium at 3GHz? Will it "fry" the computer?
A better strategy for Alpha might have been to do whatever was necessary to price it not much higher than a corresponding Pentium-based system at the time and get lots of market share and software support quickly. But this would have required deep pockets over several years, and pretty much only Intel can afford to do that.
Now, of course, we are getting a worse mouse trap: Itanium is just a horrendous architecture. It should never have seen the light of day. But Intel will manage to push it on us, whether we want it or not, because pretty much all the alternatives are effectively gone. Only AMD's 64bit chip holds out some promise because you can switch to it without changing over your entire hardware and software infrastructure.
Distributed computing is going to be the trend...if I can stack together a few cheap chips to rival a single high performance chip, what would I do?
Well, Be thought the same thing, and look what happened to them. Turned out one processor per person was enough after all, for the vast majority of users. Or should I say one general-purpose processor per person, a modern graphics card is more powerful than the CPU for its specialized task. And don't forget you won't just have to buy more processors, but the motherboard to support them - compare the prices of single, dual and quad hardware.
Granted, many apps don't fully use distributed processing power, but the ones that need most CPU probably do.
I think you are confused between distributed computing and SMP. They are different design approaches to different problems. A task that executes well (quickly + cheaply) on one won't necessarily execute well on another, even if the CPUs on both are identical.
DEC's strength was always engineering, not marketing, but they were killed by the commoditization of IT due to the twin forces of IT marketing giants (Compaq, Microsoft, Oracle) and open software (mainly Linux). It's clear today that any advantages the Alpha and/or OpenVMS give are completely wiped out by the cheapness of mass produced solutions.
HP is not taking a big risk betting on Itanium because the CPU is almost entirely irrelevant in today's market. My notebook runs 2-3 times faster than the front-end Alpha's used by our tour operator client, and it's only the lack of decent software such as the multithreading ACMS clients we wrote (able to handle 500+ terminals on a modest Alpha) that prevents us using Linux instead, on whatever box happens to be lying around. (And yes, we'll do a port of ACMS and the multithreaded clients so that our client can switch away from his Alpha/OpenVMS clusters).
Anyhow, the demise of Digital and all their technology was clear from the day Dave Custer and his team went to work for Microsoft on NT.
Sig for sale or rent. One previous user. Inquire within.
Quote from former HP employee I know:
"HP will be a printer company in two years"
Well, I think Be failed for some really different reasons. Just like you can't say, "hey, Mozart is a music genius and see what happened to him - he's dead!"
For PC, yes, one CPU per person is enough - I'd extend that further - one person doesn't even need one CPU. If you think this way, then it's pretty clear why we stack processors together - use 10 CPU configuration to serve 30 users on dumb terminals! Isn't that cheap? I think so...that's what a lot of people do when they're short of $$$.
I think you are confused between distributed computing and SMP. They are different design approaches to different problems. A task that executes well (quickly + cheaply) on one won't necessarily execute well on another, even if the CPUs on both are identical.
No, I'm not confused - just putting them together in order to avoid confusion for the people who read it
On the other hand, multiple CPU on a cluster let's say, is more difficult. There is, I think, 1 or 2 good OS that would help you do the job, but it's not trivial and require participation on your part. For example, LAM/MPI, a very common and 'easy' approach, is still requires pretty explicit communication code in your programs.
... was DEC's adoption of exclusionary tactics. Smack dab in the middle of the interoperability wars that bestowed fleeting fame on AT&T UNIX, DEC decided to build walls around every product they made. OSF/1 may have been a superior O/S but it wasn't SVR4 and it wasn't FIPS compliant, and for a while we couldn't purchase it. And a just a little while later, it didn't matter.
Ken Olsen was spot on with his snake oil pronouncement, but it helped kill the company.
This is only to be expected, and I was actually expecting it to be announced about six months ago.
Carly Fiorina has made it clear that HP is no longer a technology company, but a sales company. They are no longer willing to take risks, they are no longer willing to develop new ideas and different architectures, and very ironcally, they are no longer willing to invent. If you need proof, just look at their nearly-dead calculator division.
The Alpha is dead. RPN is nearly dead. The spirits of Hewlett and Packard are dead, and Carly is going to make a very successful printer sales company by killing them. Unfortunately.
"People who do stupid things with hazardous materials often die." -- Jim Davidson on alt.folklore.urban