Slashdot Mirror
Ixnay WinNT on Alpha
Thanks to Jason Perlow for sending us the
story that Compaq has laid off roughly 100 engineers responsible for WinNT on the Alpha platform, and will be not be doing more development on it. It's an interesting development, especially taken in light of Compaq's recent push with Linux, True64, and OpenVMS as the OSes (OSi?) of choice with the Alpha platform.
Yea, I think he works directly as a Linux International promoter... But I bet he still has contacts ;-) and I can't think of anyone else who could get a couple 21264's to the gcc guys, can you? If so, feel free to beg em ;-) couldn't hurt.
But, the only thing we found was for Tru64, so the "chiefs" decided to get some more SGI's, and keep the commercial UNIX problems all with one company. :-( I do know for a fact though, that if the preformance (espically FP w/ Fortran) would have been there from a commercial Linux compiler, not only would there be a Alpha Linux box here, but a couple PIII's running Linux too. For pure hardware preformance, there would have been a lot of nice hardware we could have got for the price of the Origin and Octane we got insted.
In the end, brand loyalty won, because it came down to a risk on something unknown, with questionable preformance from unknow compilers, compared to knowing how SGI would screw us already. As for Linux developed with a commercial compiler I say YES! I have long been a believer that people who don't optimize thier compiles, but go out and spend $700 more for a CPU that runs 20% faster are total idiots! It depends much on if the distribution was put together with the compiler, or if the compiler is included, and the price tag. Actually, I have thought about this for over a year now, and I think it was last summer some time I came to this conclusion:
"If someone is selling a C compiler for Linux, and they claim it is better than GCC, why in the world don't they bundle it with a whole Linux distribution that was compiled with thier compiler?"
If in fact the preformance was better, you may well be able to spend say $100 to $300 on a commercial compiler, and get a system that was 35% faster overall. I mean, even GCC can make your system up to 30% faster if you use the right flags (see my tests and the claims of Mandrake Linux). People are always looking for the latest greatest hardware boosts for an edge, and it's very short sighted to think about spending all that money on faster hardware when you have the source code right in front of you, and your runing unoptimized binaries.
If Intel REALLY wanted to support Linux, they would port thier compiler to Linux (which is an OUTSTANDING C compiler for x86), and let VAResearch sell Red Hat recompiled optimized for each specific CPU using Intel's compiler. They would probably see a preformance boost that they can't get out of hardware alone.
Along those lines, I also believe that AMD should really start backing the GCC project by donating half a dozen of thier new Athlon processors to the top developers in GCC. If they did that, and shiped instruction set specs and details, there could be a -march=athlon flag that could potential put thier preformance WAY ahead of PIII.
I'm also hoping that IBM pays some close attention to GCC now that they are supporting the Linux community, with the G3 and the older Cyrix based stuff they still own, they could really make Linux on thier hardware "wake people up."
Don't get me wrong, I think GCC is doing pretty darn well. But I do think that GCC development and support is much more important than elevating Linus to the level of a god.
You know, is there any way that Digital could make the alpha-nt source available to the public? I know, it's a crazy idea, and I'm almost positive that MS would have fits over it.. but is it possible they made an oversight and Digital could release it?
This is the reason free software can be so beneficial - alpha-NT would have continued unhindered if it had a developer base independent of the company. Now it's just going to sink for no reason other than economics. :/
--
We'd had an annoying time with Compaq's Intel boxes
First, I've never had a problem with anything made by Compaq (except for their consumer-level stuff, which is about the same as anybody else's.) I am writing this on a (circa) 1995 Deskpro that has never had any hardware problems at all.
and their crappy dealer-based sales system
As opposed to Digital's dealer-based sales system?
(yes, Digital had dealers, just like Compaq.. and the only thing required to become one was a PO from a customer worth $15000 or more.)
I'm sorry to hear about your problems with Compaq, but in truth, my experiences with Digital (pre-Compaq) are just as bad - how about a hard drive that failed, and Digital doesn't have any to replace it - so they said "we've got that on back-order right now, we expect stock in 3 to 6 weeks." _FIVE MONTHS LATER_ our hard drive arrived.
I for one welcomed the purchase by Compaq.
Our top programmer gave us a talk explaining that "NT" had originally stood for N10 ("N-ten")
Your programmer was wrong. NT stood for "New Technology" - as in "not Windows 3.x" - it came from a new codebase, whereas the old (3.x) code stemmed from what was just a graphical shell (no multitasking, networking, memory management, etc..)
Well, to be fair both MS and IBM thought DOS was going to be dead meat under 286+ machines and attempted to replace it. Of course, the replacement, OS/2, was a disaster until about 1992. Then someone at MS figured out that they could kluge a protected mode extender onto DOS and tell IBM to screw themselves, and most PCs are still running with that solution (Windows) to this day, making everyone's head hurt.
On the other hand, load up WP 4.2 and Lotus 1-2-3 2.x -- You can still get a lot of work done with 640K.
--
Business. Numbers. Money. People. Computer World.
Wasn't IBM's big mistake putting the 384K ROM space at the top of real memory, instead of the bottom? My understanding is that IBM did it this way only because that's how Apple did it (top 16K of the ]['s 64K.)
--
Business. Numbers. Money. People. Computer World.
People known to be working on getting GCC working better on Alpha are: Richard Henderson, David Mosberger, Catherine Moore. If you think any OS that uses GCC is something you would like to support, thank these people, and the others at GCC/EGCS, they are the ones that will give you the freedom to run something other than WinNT and Tru64 on your Alpha.
MS/IBM Lan Manager (Also resold by 3Com as 3Open or something) was an OS/2-based product which failed to gain much significant marketshare. It would be long forgotten, except that Windows NT uses the same networking layer.
It certainly works very differently from NetWare, for one it ran NetBEUI and NetBIOS-over-TCP/IP. If you wrote your DOS app on the LanMan APIs, you're pretty much stuck on either DOS, OS/2, or Windows. (These were the days before 'WinSock' - APIs were pretty much NOS-specific.)
I think the guy's concern is that they is no NetBEUI support in Linux, and the MS DOS TCP/IP software cuts pretty significantly into your 640K.
(One solution might be to look at the book "Unauthorized Windows 95" by Andrew Schulman . In the book he describes how to use Win95 to create a 32-bit version of DOS, complete with protected mode networking.)
--
Business. Numbers. Money. People. Computer World.
But, yes, the Microsoft C compiler shows well on NT/Alpha SPEC benchmarks. Problem is, most of the common binaries are FX!32 emulation. But, for people working with source, NT/Alpha is a decent choice. I only wish GCC was as good for Alpha.
So, for example the scientific community or number crunchers who write thier own code can benifit from NT/Alpha, but people who rely on commercial software for things like graphics, sound, video, database, etc, don't get that benifit, and NT/Alpha isn't that great a choice (unless, of course, they get NT/Alpha native commercial binaries).
Not so fast folks. Compaq have not made any statements about the end of NT or Alpha, and the reality of the layoffs is quite different from the press spin on it.
.18 micron shrink is in the works (possibly with copper interconnects) which will have two benefits. 1) a significant leap in MHz, way over the 1GHz milestone. 2) A significant reduction in price (smaller die = more per wafer). When Merced ships Alpha will be significantly faster, smaller and cheeper! Alpha still rules the 64 bit roost, and nothing else in the 64bit market looks like getting anywhere close to it in price/performance.
The engineers layed off worked on the NT4 port to Alpha. That is done and dusted, there is no more work to be done in that field now. The next Alpha NT move is to Win64, and that port is being developed and driven entirely in-house at Microsoft.
NT is not dead on Alpha. If Windows 2000 scales as well as Microsoft are predicting, and the Win64 Alpha version gets finished as planned, then a lot of companies are going to be very interested in combining this with the horsepower Compaqs imminent Wildfire (Alpha) platform provides. If anything it will put NT Alpha in more direct competition with big UNIX systems, an area of competition it's been denied so far due to NT's lack of scalability.
Alpha also has a very rosey future. A
This story about the end of Alpha NT is just that, a story. Pure press fiction and FUD.
Macka
So, why does Microsoft even bother making Win2000 run on 32-bit machines? Why not start out 64-bit? This is especially true if you are targeting the "enterprise" market and their high-end go-fast servers
It's important to note that Alpha has been a key propaganda weapon for Microsoft to argue that NT is competitive with the big unix iron. Many of their boasts (such as 10,000 POP3 users on Exchange, or TerraServer - the 'worlds largest' on-line database.) derive from running NT on some very high-end Alpha equipment. The bottom line is that I can't belive MS would drop Alpha development because doing so would make NT/2000 look half as fast as far as the marketing department is concerned.
As for "64-bit ready", it's not the software, but the hardware. If MS ever gets a 64-bit OS and applicaitons out the door, you get a free performance boost with your NT/Alpha box. Of course, it seems that Microsoft is munging the 32 to 64 bit transition as bad as they munged the 16 to 32 bit transition.
--
Business. Numbers. Money. People. Computer World.
People on the high end of Intel hardware ($3000 to $7000 range) stick with Intel because of OS flexability
I've never heard that before. Is this true? I thought they used it because of executable compatibility. Do people really buy $5000 x86 machines because they run more than one OS? Any case-studies out there?
I thought there was some speculation that 64-bit NT would ship on the Alpha sometime next year, before Merced can make it out the door. I wonder if NT/Alpha is really being killed, or Microsoft is just picking up all the development under some secret back-slapping deal.
Regardless, Compaq/DEC needs to start marketing the hardware better at people who buy large x86 servers.
As for Alpha applications, it seems that there are very few workstation apps, but server applications seem aplenty. Specifically, all of the Microsoft stuff (SQL, Exchange, etc) runs on Alpha.
--
Business. Numbers. Money. People. Computer World.
First off, the suffix "i" is the plural form of "us", as in the plural of "virus" is "virii" not "viruses". Second, even if "OS" ended in "us", it would need to be a Latin word to qualify for the "i" suffix. Unfortunately, I can't seem to find an word that ends in "us" that is not derived from Latin, but I'm sure there is at least one.
It could still turn out that way, unfortunately.
Dropping WinNT support for the Alpha will probably have a short term effect of decreasing Alpha sales. If Linux and other UNIXes catch on more, the Alpha sales could rebound, except this time with UNIXes instead of NT on them. However, if they don't, Compaq could just shift its focus to Wintel and drop the Alpha entirely.
So, this could be one of two things:
1) Compaq signalling they'd rather concentrate on UNIX than WinNT
2) Compaq signalling they'd rather concentrate on Intel-compatible chips than Alpha
10 PRINT CHR$(205.5+RND(1)); : GOTO 10
Golly Beav, why would Compaq, a company that is having some financial turbulence, want to concentrate on providing the resources they have available now instead of waiting on another company to provide a product (64 bit NT; stay with me here) that many people are saying they may never ship?
Gee Wally, I dunno; maybe there are some things we just aren't meant to know.
Do people really buy $5000 x86 machines because they run more than one OS?
That's not exactly what I meant about Intel Hardware. What I meant was they can run NT if they are use to Windows, and get full system preformance expectations from the hardware with NT. They can run Linux or FreeBSD, and get the preformance they expect.
With Alpha, you can only get the preformance you would expect from an alpha if you run Tru64. If you run Linux, or NT, or anything else, your preformance will not be all that it could be.
Alpha hardware has always been ahead, don't be fooled by the fact that people are saying how slow the Multia is, that came out when Intel only had 386's, and was dang fast for the time. Alphas are still fast, but you can't compare the old Alphas to the new Intels.
The Alpha market has problems because Tru64 UNIX costs soo much, and the compiler is another big $$$ on top of that. But, if you want to get the most out of your Alpha, you buy them.
If people out there really want to preserve Alpha as a choice in the CPU market, Linux could be an answer in the future, but it isn't now. Linux on Alpha is plagued with a few problems yet (or maybe it's just Red Hat Linux for Alpha?). In addition, gcc isn't bringing executable preformance to the levels of the commercial compilers on this hardware (more important that Linux itself).
Compaq would be wise to take some of the money they save cutting NT, and put a small fraction of that into patching up GCC for Alpha... Or, porting some commercial compilers to Linux/Alpha and selling them at a reasonable price (although this is a less preferable option).
If GCC can reach the level of efficency on Alpha hardware that it has reached on Intel (and now AMD hardware, see PGCC), Alpha will make a BIG come back. But, if Merced comes out before this happens, all bets are off. Alpha can beat Merced, but it will need the support of people NOW, not when Merced arrives. (Any Compaq guys out there? Hay, Mad Dog, how `bout getting the GCC guys a couple more 21264's, then we would all be happier!)