Slashdot Mirror
Compaq expands Linux line
Richard Finney sent us story to Compaq's latest Alpha announcement. They will be rolling out some new Alpha-based models, trying to use the popularity of Linux to boost sales of this chip. Interestingly, Bob Young, of Red Hat is saying that the # of units of the Alpha are too small to make a huge difference.
If Compaq cares about Linux, why are they letting em86 (the program that runs x86 binaries on Alpha) flounder? I notice that they recently removed it from their Alpha Migration Tools page (http://www.digital.com/amt/), too. Because of this, em86 has been languishing. It can't run a modern Netscape without crashing quickly, it can't run libc6 programs very well, etc. Whereas their FX!32 program for NT is being kept very well up-to-date (and always was better than em86 anyway).
Alpha chips that have been maximized for Linux? Actually, I think they mean the servers have been maximized for Linux. Still, why do you need to maximize something so that Linux will run on it? Modify? Maybe. Minimize, certainly. Is there some kind of impression out there in the media that Linux is some sort of "special needs" OS, or that it needs ultra-fast-cool hardware to run?
Sounds like more FUD to me.
No sig.
Please elaborate. Any technology starts out as being pretty much useless. Even though the design may be good, the initial concept demo is generally very limitted. It only needs to show the concept w/o the expense of making it robust.
What is it about ANDF that makes it unviable? The key limitation I've seen is that they attempt to provide an abstraction to the OS. That would be solved by using a subset of the capabilities; abstract the instruction set alone. Thus applications could be distributed for "Linux", not "Linux on i386".
Then again, I've only seen ANDF at a 10,000 ft
view. Please give details.
The barrier to entry is that software only runs on Intel. If all the other big names got together and supported ANDF, this wouldn't be an issue any more.
ANDF is a technology for compiling programs into an intermediate format that can then be targetted to specific architectures. It's better than Java byte code in that it retains the program's structure.
If they'd drop the cross-OS considerations, and instead just focussed on using ANDF to ensure that application software can run on any processor that has a Linux kernel, then the market would be wide open....
Hmmm.. nah, Intel wouldn't like that. Never mind.
The Alpha processor won't make a difference, eh? /his/ kernel? Why goll-lee, it's an Alpha system.
Hey Bob, ever wonder what Mr. Torvalds uses at home for hacking
Sorry, but I think Linus has more experience with this Linux thing than you do.
... start selling cheap, low-end Alpha chips. I'd gladly buy a K7 motherboard (once it's out), and a $300 Alpha CPU. That's about where they'd reach mass market. The chips wouldn't have to be *that* fast, just faster than Intel.
Marketshare isn't everything. This is a good deal for Red Hat. It gives them the claim of scalability. "Buy Red Hat Linux for x86. It's scalable; if you ever need more power, step up to an Alpha [, Sun or SGI]."
/* MAGIC THEATRE
ENTRANCE NOT FOR EVERYBODY
MADMEN ONLY */
Do you have more information about this? The Math library release was a great start. It made a significant impact on the performance of our Alpha Linux cluster. We really need the Dec Uni...er Compaq Fortran compiler suite for our development work. I think I could scrape up a couple $K to put a copy of that on the head node in our cluster.
my understanding was that gcc performed badly on an alpha. digital, er, compaq's compiler produced far better code. any word on whether compaq was going to help the gcc/egcs team out?
also, anyone try building an alpha/linux distribution with the compaq compiler? (sorry rms, just think of it as goal setting)
US Citizen living abroad? Register to vote!
see this
You make valid points. We have 20 Alpha Linux systems that we use to crunch numbers 24/7 using our own parallel codes. For this purpose, the Alphas make perfect sense--porting our codes was pretty easy. We bought the first several systems complete with monitors and tried to use them as workstations. But I quickly concluded that it was too much trouble to deal with the little bugs in Alpha Linux applications (especially X), so I pulled the good video cards/monitors and other useful parts and used them to build Celeron front-end systems. The alphas are quite contentedly running headless on the shelf, and we just access them from the Intels. The last batch of alphas I bought doesn't even have video cards.
I have an alpha running linux it is rather old a 21164 running at 500 Mhz but it sure runs some applications much faster than any intel machine.
I payed $7000 for this machine but today it is less than $3000 (with descent amount of memory and disk). The problem is that a 21264 is still about $7000
Pros: Floating point performance is great (most people do not care but for my MD simulations I NEED it)
Cons: Xfree86 is unstable.
What compaq should do is to release their compilers for linux, and preferably release a $5000 21264 machine then I would by 16 of this and stop using the SP2
Copyright 1998 arne Verbatim copying and distribution is permited as long as this message is preserved
Alphas come in models of up to, AFAIK, 600 MHz, with the 21364 being a > 1 GHz Merced-killer in the NTDF.
Yea right, ANDF sucked major do-do's.
It was broken as designed, and only worked for simple test cases. Most real world apps do more than write "Hello World" to the screen.
I built my 164LX @ 533MHz/ 128MB sdram, 4Gb HD for about $1600USD (with tax, etc..)
;^P
I think people overlook that fact that Alpha's are for the most part marketed at the cutting edge performance curve.
That's why I bought an older model, just like I bought my Pentium after the PII came out. I'm not obsesed(sp?) with speed.. but my (close to 2 year old model) alpha still cracks rc5 keys 10 times faster than my computer at work (PII 350).
Roger
I said no... but I missed and it came out yes.
First, I'm not sure Bob Young's statements were entirely in context. If you read the actual quotes, it seems to me like all he's really saying is that the Alpha is not a driving factor in their business right now, due to lower market share. Most of the derrogatory spin seems to applied by the reporter.
But, I must say that I can understand why they might not take Compaq too seriously yet. Purchasing a Compaq branded alpha with any kind of enterprise-level features will cost you tens of thousands of dollars (at least as of a couple months ago), and even then Compaq won't support you. We had a Microway Alpha in here for about a month as a server, and it had tons of problems (configuration out of the box, shipped by Microway). I spoke to a lot of people at the time, and the consensus seemed to be that if you have time to port & clean-up individual applications, then Alpha Linux was great for running those apps, but for general purpose serving, it just wasn't a quality product yet.
The Alpha port requires a lot of work for a small number of users, and since these users are probably even more technical than the Intel base it probably is one of the lowest income platforms they support. I can understand why RedHat doesn't see it as a compelling platform right now.
I actually think other distributions are a good thing. Just having redhat on alpha sux. I'm not saying redhat sux. Just having debian, or slackware, or suse, or anything on alpha is bad. Choice is good.
(I need sleep)
-matt
That is exactly correct. The "inner loop" of the RC5 algorithm is very rotate-intensive. Processors without rotate instructions (ie, almost all non-x86's) have to use a copy, two shifts, and an OR -- and the older Alphas (21064, 21164) have only one shifter in their ALU, which sends its ILP straight to hell.. The 21264 is better, with two shifters in different pipelines, but the x86 implements rotate as a one-cycle native instruction.
The Alpha rocks at just about any other crypto algorithm, though; 64 bit ALU width, a high clock frequency, and highly effective branch prediction (the Alpha uses the simple heuristic of always predicting to follow a branch to a *lower* address, so it's trivial for a compiler to build a loop such that it correctly predicts all but the last iteration) all goes a long way, and the Alpha's biggest weakness (memory subsystem) doesn't hurt it because in crypto the data set is small enough to fit in L1.
I heard a rumor of 800MHz 21164PC's for $300 by the year's end -- anyone have some hard info? www.alpha-processor.com remains silent about the latest process shrink.
They have everything that could advance certain states of computing, but do not have a management clue on how to put it all together.
Sounds to me like you're drawing an awful lot of conclusions with very little actual information. Your comments amount to anti-Compaq FUD.
"Optimizing for Linux" could be as simple as choosing hardware that has really good Linux drivers, or that won't require special configuration steps under Linux.
It could also include little things like picking a video adapter that had good XFree86 support, and even bundling a three-button mouse instead of the usual 2-button Logitech Firstmouse. (BTW, Logitech makes a three-button OEM version of that mouse. Hewlett Packard bundles them, they're not bad. I'm looking for a consumer version.)
Optimizing for Linux could also involve staying away from any chipsets that had known problems with Linux.
In short, optimizing for Linux is a very good thing.
Compaq released binaries for libcpml, a port of their math libraries to Alpha/Linux. That was a big help to those running fp-intensive applications.
There have been strong hints that we can expect more from Compaq. They are being far more aggresive than Digital ever was at supporting Linux. With Digital I always had the impression the Alpha/Linux project was a basement project, it was hardly even mentioned anywhere on their web site. So far with Compaq, we have seen many announcements, press releases, and some actual software released for Linux.
One possibility is a commercial port of C++ compiler to Linux. The compiler may not be free, but if the binaries can be redistributed it will be a big help.
I really doubt however if the Linux kernel will ever be compiled with anything but gcc. There's simply too much code in there that is gcc-specific. But distributing an X server built with Compaq's optimized compiler is a real possibility.
I wonder how much of the Alpha's popularity problem stems from the fact that Compaq/Digital sells it primarily in premium servers and workstations. Sure, it means that the people who get the chip get really nice machines, but it also means that the pricing that most people see is, well, a bit on the high-end.
Just once, I'd like to see an announcement on the Alpha that says, 'This chip is shipping for $??? in quantities of 1000.' They do sell them that way to third-party distributors, after all.
Maybe once Alpha Processor, Inc., finally gets going...
Right... I think I see where you're coming from.
:-/ )
However, RedHat still does sell 5.2 for Alpha - and Sparc, too.
I can't find the boxed set, but I suspect that this is simply because of sound business reasons.
Equally, I also suspect that Compaq are barking up the wrong tree... An OS like Linux doesn't make a CPU. They simply have to lower the price, or at least produce boxes that match up to what the UltraSparc hardware can do - which probably means a departure from PC style hardware.
But really, there's no need to get paranoid. Just keep pushing your favourite CPU make and model if you beleive in it.
(Currently running NT of all things...
Hey,
I think that it will be nice to see this happening. Hopefully, they can reduce the price somewhat on them though; all I'd need would be a single CPU 21264 system. Unfortunately, a lot of people are right - just buy 5 Intel based systems and with distributed processing, you have the same horsepower or more. Of course, you also have the added overhead of maintaining a distributed system, and the added space and electricity usage of each additional system...
Oh yeah - and to the 466MHz, Intel has finally caught up poster... Please keep in mind that MHz is mostly an irrelevant number when comparing chips from two different families. As an example - say I could overclock a 286 to 2000MHz - do you think it would run things faster than a Pentium II at 300MHz? I didn't think so. When dealing with chips of vastly different architectures, comparing MHz to MHz is mostly an excersize in futility.
I say this, because a 466MHz 21264 will beat a P-II at that speed any day of the week, even on Sundays. Heck - even looking at the PPC 750 - at similar clock speeds they beat the P-II. Essentially both the Alpha and PPC chips make much better use of their clock cycles than just about any x86 chip on the market.
Oh - and I don't think Bob Young is being terribly negative in his comments about the Alpha. But, the article did make one major faux pas - indicating that software would be hard to get compiled for the Alpha platform... Uh - OpenSource, and lots of Linux Distros come with lots of it - there are more than enough OpenSource utilities and apps on most distros that if one cannot compile up a good set of them on a Linux box running on an Alpha - I'd be suprised.
- Porter Woodward
True, and I heard that Apple invented to atom bomb and started the holocaust. You know the ignorant folks spell Mac in all caps like it's an acronym. MAC, heh.
It's far easier to forgive your enemy after you get even with him.
It's far easier to forgive your enemy after you get even with him.
I agree that alphas rock everything else on the planet, well maybe not, but i love them. Tho I do have a problem with the rc5 claim, my alpha didn't do anything all that impressive with the rc5-64 cracking, I read that it was due to some instructions present in the x86 processors but that need 2 or 3 instructions to do the same thing on the alpha. I could be very wrong....but thats how i remember it.
I have been a fan of the alpha for a while, I have spent maybe the past 40 minutes looking for a place that would actually list a price for a motherboard with a chip so I could get an idea of what it would cost to build a new system. It is things like this that, in my humble opinion, prevent some people from going to an alpha. I always build my own systems, it may just be a stange fetish of mine, but I like my systems only the way I can build them, and not being able to find a place that will let me get a quick idea of how much the chip and m/b are going to run is very annoying. When talking about the spread of linux to the alpha and all that this is probably more relevant than it first looks. It would be interesting to see how many people build thier own systems, and how many people are turned off by not having that same ease and flexibility on the alpha platform
I'm not going to disagree that Digital really messed this up from the beginning. The Alpha processor is simply way ahead of its time in performance and speed. But, we come back to two problems. First, in the Unix world, it was tied to the $2000/license DEC Unix in the beginning which is fine and dandy for researchers with lots of cash (considering the machines debuted at $20K, I guess an additional $2K isn't much). These days, I think most will agree that for day-to-day operations, OS's are pretty much commodity items, with Linux, *BSD, Windows, MacOS, and BeOS all available for less than $100. Now that the Alpha processors have dropped in price (check out www.microway.com, where you can get a good machine for under $3000), you can't justify the $2000 price tag for the operating system. And I think Compaq is making a good move with embracing linux. What they really need to do (ask any researcher) is port the DEC compilers to linux so we can run all our simulations and do our development with solid compilers but without the DEC Unix license overhead.
Two, in the case of NT, you've got the ever-present chicken and egg problem. Well, no one wants to port until the base is popular, but the base never grows until applications are ported. I'm suprised that Microsoft hasn't jumped on this yet. Ever since Intel signed agreements with HP to port Linux to Merced, I'm suprised that Microsoft hasn't gotten pissed off enough to find a competing chip - Alpha - to really build its 64-bit strategy around. Now I know everyone in the audience will boo and hiss, but face it - Microsoft made the PC market for Intel, and they could do the same for Alpha. So, reluctantly, I have to cheer on Microsoft in this area.
The pricing on Alphas has dropped significantly. For our research group, our first 266-MHz 21164 Alpha was over $10K, but the latest 633 MHz 21164 Alpha was about $7K. Now we're drooling over the 21264's, waiting patiently for Microway to offer one in the sub-$8K range. And you really can't beat these machines for computationally intensive work.
I get this from CINT95 Results Looks like the Pentium can at least keep up in (I'm guessing) integer arithmetic. This isn't addressing "memory bandwidth"; but doesn't this mean the instruction scheduling is as good?
on linux an alpha is like a cheetah on speed. Now that more and more distros are coming out with ports for the alpha hopefully we can move away from redhat too (please, that wasn't meant to start a flame war, I just prefer other distributions)
I think you can still buy the UX motherboards. They include the 21164 CPU at speeds up to 600MHz or so.
Many users do build their own, and before Digital Semiconductor was dissolved their OEM resellers were selling SX and LX boards for this purpose. (The UX board was developed by Deskstation, and is now supported by API I think...)
You can't buy a barbones 21264 package yet because Compaq hasn't provided an OEM channel for it. Rumors are that IBM will be reselling 21264 boards soon...
Here is the problem though. I see the 21164 @533Mhz listed as rating with int95=16.1 and fp95=18.8. The cheapest i have found this machine is $1995.
A PII 400 rates at int95=16.9 fp95=12.8 and a nearly identical system can be built for $1030. Add $180 if you wish to use a PII 450 with int95-18.5 and fp95=13.3. It does not figure to purchase an Alpha at those costs. Expecialy when you can factor in the price of AMD and Cyrix chips.
Hell, at those costs you use an SMP mommy on the 400Mhz PII systems and double your CPU power at $1500, or use 450's to bring it to $1860. Im not even going to get into the 21264's @500Mhz that Ive seen for $6950 with a int95=26 and fp95=49. A dual 450 will outperform that machine in integer, although still be 1/2 the floating point. Currently who needs the floating point? Not the home user, and definutaly not the web server market either. At current, their price is their greatest threat. Its almost as if they are trying to compete with Intel boxes using Sun pricing.
If you can find better prices, Im sure that the masses will follow.
The people who have built Linux/Alpha clusters have been asking for a few simple things for two years now. The message from above has always been "We're thinking about it." I'll take Compaq's commitment to Linux seriously when I see a continuous contribution to Linux software technology. The release of the math library was a giant step in the right direction. Let's see more statements like that, not stuff that was cooked up in the marketing department.
You can never have too much RAM or too much disk space. --Ancient American Proverb, circa 1980
Does anyone know how much these Alpha boxes will cost? I would certainly be interested in buying one, but really don't want to pay more than I would for a good PC... IMO that money would be better invested in a small Beowulf-ish system -- figure you can build a PC using generic parts (Celeron 300A, two 128 MB DIMMs, 100bt NIC, etc.) for under $800, then $4000 means a 5-node micro-cluster with a switch in the middle. For tasks that can be distributed, you're probably getting more CPU for your dollar.
The counter argument is that an Alpha workstation would have less latency. I was playing around with an Alpha 21164 at 600MHz running NT the other day... find *.dll took about 1.5 seconds to return 750 items... and the stupid Windows cursor was actually changing for each widget it went across in the explorer window (if you pay attention, it normally doesn't on an Intel, but on an Alpha you really see just how noisy the Windows UI is -- it was actually visually distracting). I can only imagine what Linux would be like on one of these....
So I guess my decision comes down to: do I need a fast workstation or something that is fast for batch jobs? And my answer is: I don't have enough money anyways, so never mind. *sigh*
Morning. Need coffee.
Looks like pentium has caught up with Alpha, finally.
If he doesn't want to support it, then fine. Linux is open source. Compaq should take the initiative to take the code and write their own drivers for the Alpha. That's really all that needs to be done!
Compaq Linux! If they do a good job and can offer affordable Linux solutions with lots of prebuilt software then people will buy the computer.
| Mac in all caps like it's an acronym
...
But *Macintosh* is an acronym
"Machine Always Crashes If Not The Operating System Hangs"
-- Rick
Looks like pentium has caught up with Alpha, finally.
Nonsense. The memory bandwidth of the 21264-based models is far superior to any Pentium design, let alone the instruction scheduler and four-way pipeline... check the SPEC results if you need evidence (http://www.spec.org/).
Microway does not seem to be the cheapest solution on the block. DCG is apparently currently selling XP1000s for around $7k. I expect .edu can get
the same kind of price from the Q, with digital Unix.
ObURLs:
http://www.dcginc.com
http://www.harddata.com
Huh? There is no 450MHz Alpha on that page... there are numbers for Compaq's DS20 and XP1000. (The DS20 is a dual-21264 at 500MHz, the XP1000 is a single 500MHz 21264.)
Even the workstation at 26.9 comes in far higher than any Pentium. And of course the fp scores are no contest...
ubbb, sorry.
One big hurdle to performance was the math libraries. Compaq recently released those -- they're available on their web page for download. The general compiler issue is one that hasn't escaped Compaq's notice. Things are being examined right now. Don't mistake silence for inactivity in this area.
There were good deals on Alphas back when Digital Semiconductor was producing OEM motherboards... you could get fully loaded 164LX systems for around $2000-2500.
Now that Digital Semiconductor is defunct, and Compaq has given up on OEM sales, the only 21264 Alphas for sale are Compaq branded (and expensive). Some 21164-based models are still available though, and very inexpensive.
But there is hope: Compaq is looking for a third-party Alpha OEM supplier, and rumors have it that IBM will pick up the business.
really? are you using the distributed.net clients? in my experience alphas are not so great at rc5... my 164SX is comparable to my 200MHz PII. this is partly because the alpha has no bit rotate instructions, iirc...
DES is another story... my 164SX runs circles around our pentiums on DES, and slightly faster than a UltraSPARC 350MHz (which also does well on DES).
Oh blood and ASHES! I hit preview and all my text is deleted. Argh..
Short story is Intel is funding companies and efforts to subtly "lave out" alternative CPU's. EXPECT this FUD effort to carry over into the Linux world, such as Rob Young's unflattering remarks about Linux on Alpha.
Gee, didn't Red Hat used to sell an Alpha Linux?
Didn't VA Research used to sell Alpha workstations?
Didn't BeOS used to support PowerPC?
Didn't MetaCreations promise a Mac port of the MC/Intel "MetaStream" web 3D project?
Isn't Intel deliberately withholding the specs to Indeo instead of producing a Mac version (Indeo is for wanna-be's, but still I've talked to people who can't consider it because it's not x-platform).
We're just about to get independence from Microsoft and people don't see we'll end up right in the grip of Intel. Watch for future "Linux efforts" by Intel designed to keep non-Intel Linux users off the cutting edge. No one can "own" Linux, true, but anyone can make a version so technically appealing that the masses won't give a rat's ass about "politically correct" licensing (look at all the people who didn't care about the *OLD* QT/KDE licensing...)
I would have liked an Alpha box, like the one Linus got from Bob. However, why don't they do an AMD; skip the profit for some time so they can get a larger consumer base?
I think 4000$ is an absolute upper limit for volume shipping.
I'd buy an Alpha in an instant if they were cheaper. That's the whole story.
I bought a 166Mhz Alpha box for 200 Bucks without SCSI2 HD, from www.cpumicromart.com. This is a very good box, performs really well. This let me learn to prog on a Alpha for a very cheap price. Now that I am comfortable with the Alpha I will be looking forward to see the new Alpha boxes.
Frankly, GCC has problems that will take more than incremental improvments to solve. IMO we should cut off everything after the parse tree pass (ie, starting with the tree->RTL pass) and either rewrite the back-end from scratch or mate the front-end to a different existing open-source compiler system (trimaran, perhaps) which has a better icode than RTL and which is better-suited to generating code for modern architectures. Many advances in optimization technology have been made in the past decade, but GCC cannot use many of them because they cannot be easily made to fit the legacy code.
Strong words coming from a MAC fanatic. The MAC is the most proprietary of all the computing backwaters, IMHO. Apple's arrogance led to the rise of Microsoft.