Alpha Lives! But Who Will Market It?

chriton writes "The Inquirer is running articles about HP's and new "Marvel" server which will arrive Tuesday, Jan 14th and the expectation that HP will try to keep it's performance quiet. Not because it's bad like Itanic I, but because it's too good! It's built on Alpha EV78 processors connected by a switched fabric and promises blazing performance. "Marvel has, apparently some rollickingly good benchmarks that HP wants to underplay, just in case people start comparing the performance of the Alpha Marvel architecture with the Itanium 2 it also sells, and perhaps more importantly, the SuperDome machines." Alpha offers the kind of choice and competition the processor market will sorely miss when it goes. The FTC was sleeping when they allowed HP to acquire it."

Intel. not HP

[Jah-Wren Ryel]

Intel acquired Alpha, not HP. Now it was clearly in order to prep decpaq for acquisition by HP, but that's beside the point.

The Alpha engineers were given the choice to work on Itanic for Intel or to hit the road. Kind of like their worst nightmare...

Re:Intel. not HP

[Tuzanor]

no, compaq sold alpha technology to intel, but they still had it for themselves and were still selling alphas. When HP bought compaq, they inherited the alpha line.

We use some alphas at work

[SirTwitchALot]

and while they're great machines that perform well, they're very limited. It's difficult for us to get many of the applications that we use for the Alpha, and if the app is available, the vendor usually provides poor support for it. Sure you can compile OS software on the alpha, but the commercial world overwhelmingly uses traditional closed software. HP decided to stop production of the Alpha because they had a competing product (pa/risc) that was in higher demand. They even plan to eventually lose PA/RISC in favor of itanium, as the article mentions. As far as price goes, one of our clients purchased a wildfire gs320 because of the low price. They found that while it offers acceptable performance, it's very difficult and expensive to find the expertise needed to properly maintain this equipment. We run a primarily Sun shop not because it's necessarily the best, but because it's what everyone else runs, and thus easier to maintain and cheaper in the long run.

Re:how can hp use alpha tech?

[uncleFester]

Compaq sold it to intel before the merger.

marvel was already in the works before the HpaQ merger, and it would really make little sense to take a chip all the way to fab w/o at least running SOME of them to try and recoup some cost.

Plus it will probably give Intel a good idea of which components of Marvel to rape for the next gen of the (t)Itanic.

I was a very short-lived DecpaQ Tru64 admin, but have to admit I fell in lust for the OS and architechure. Our alphas ran superb for their age and the obscene obese demands our Oracle DBA inflicted upon them. Nary a whimper. I still think it's mildly criminal Compaq threw away the horsepower farm simply because they were too stupid to market the things properly.

Re:Intel. not HP

[Burdell]

Incorrect. Intel licensed some of the Alpha technology (and bought the fab that Compaq used for their Alpha CPUs), but Digital, then Compaq, and now HP owns the Alpha. This is not an unexpected release; all along Compaq and HP have said they were committed to one more full generation of Alpha CPUs (the EV7 generation). Supposedly, the third generation Itanium will incorporate some of the Intel licensed Alpha technology, and then it is supposed to "catch up" with the Alpha (so there would be no EV8 generation).

IIRC, some of the associated technologies like the switching architecture and some of the NUMA features were not licensed but held by Compaq for their Itanium servers (to give them an edge).

Re:We use some alphas at work

[vondo]

It seemed like there was only a short period of time where Alpha was cost effective compared to Intel, the rest of the time up until maybe three years ago, Alpha was often simply a heck of a lot faster and that performance was only needed in niche markets compared to today's desktop market.

For us it was waiting for Linux to reach a certain maturity and then realizing that it had. There was also the issue of g77 versus DEC's f77 which exagerated the performance difference.

Re:We use some alphas at work

[jpetts]

It seemed like there was only a short period of time where Alpha was cost effective compared to Intel, the rest of the time up until maybe three years ago, Alpha was often simply a heck of a lot faster and that performance was only needed in niche markets compared to today's desktop market.

But the cost effectiveness which you are talking about doesn't appear to factor in stability. Alpha machines running OpenVMS were rock, absolutely rock solid. We had a machine running Oracle on OpenVMS/Alpha that was not rebooted for three years, and never once showed ua a single problem. It just ran and ran and rad, and it ran FAST, too. I for one will miss the low admin burden of those Alpha/OpenVMS/Oracle boxes...

OS problems

[Anonymous Coward]

The problem with Alphas is not the CPU/hardware, it's the OS. Since the begining, the OS was badly designed and tested. Things were so bad around 1994/95 that you can't reliably restore a tar backup. The OS crash at last once a week. At that time we switched to another vendor (IBM/AIX, less performance, more costly, very reliably on both hard/soft). We are still using AIX, and paying about three times more for the same SPECbench than Sun/HP, and we are happy. No crashes.

Re:We use some alphas at work

[cballowe]

The Wildfires have not given me any reliability problems. The serial console subsystem is a glorious thing. ESC-ESC-scm sometime -- it's a lower level than the system console.

Beyond that, if you don't need an 8way or bigger box, the ES45 (or watch for the ES47 based on the EV7 series processor in the coming weeks) is probably a better bang for the buck proposition. Also the ES80 will be an 8-way ES series -- the notes I have suggest that it will be significantly cheaper. It's also been suggested that if one is spec'ing new systems now, budgeting for the GS80 and acquiring an ES80 when they're released is probably a better option.

Re:Intel. not HP

[Anonymous Coward]

Intel bought the rights to the Alpha processor, but the FTC saw this move to stomp out competition in the processor market. The FTC required the processor to be licensed to a few other companies, including (now HP), and Samsung among others (I don't remember the full list anymore).

Re:Dec Alpha

[Dahan]

I love my DEC Alpha 500PC164. Too bad the only thing that will run on it is Debian.

Hmm, what's wrong with it? Why won't it run OpenVMS [montagar.com], Tru64 [compaq.com], or NetBSD [netbsd.org]?

A brief history of alphas (and thier troubles)

[pantherace]

I have been a supporter of alphas for a while. When I got the first alpha I could play on it was highly superior to anything else. (The University had for it's high performance computing a 16-way mips) and several persons got 533MHz alphas, when they were the first chip >500MHz. Originally NT ran on them. It was stable. (Primarily because DEC rewrote most of the important parts themselves for alpha) and fx86! was wonderful-ran x86 code the fastest at the time.

Then DEC kind of died. They didn't seem to market. The tech was good, but no marketing, and some issues. Before the compaq merger, DEC sold StrongARM, and all it's fabs (aging) to intel, in return Intel was supposed to fab the next generation alpha chips, and was prevented by the FCC/court (or a combination) from aquiring alphas (due to anti-trust, not that that mattered to the DOJ when they did...) Intel did not fab the next generation (21164@smaller process and 21264s) of alphas. They claimed that they couldn't because the chip was too complex. (There is no evidence that they ever did, and this was just before the compaq merger)

Compaq acquires DEC. It takes it's time, but releases 21264s (fabed mostly by Samsung, and some supposedly by IBM), They branch off the alpha tech to API (Alpha Processor Incorperated) which sets the EV7 (21364 = (21264 core w/improvements + RAMBUS controller) development back. (additionally, MANY alpha engineers were hired by AMD when DEC was merging, and the EV6 bus (and many other tech goodies) were licenced to AMD (Slot A was originally an Alpha slot) for inclusion in Athlons (who still run on an Alpha bus)

Compaq decides to inhouse the developers again and sets EV7 back more. (EV8 is reportedly mostly on schedule) Then Compaq decides to sell the alpha tech (or much of it) to Intel (DOJ apparently doesn't care about anti-trust at this point) and cancels EV8 (which was reported to include Hyper-threading, multiple cores on a die, Onboard Memory controller (like the ev7) (pretty much every "cool" thing Power4, Intel, and AMD were planning on having.)...and was due out this year) EV7 is phenominally behind schedule. Finally EV7 makes an appearance, or will (asuming the article is accurate) as basically an EV6 core with tweaks and a RAMBUS memory controller onboard the chip (256-bit dual channel, so it actually isn't a POS like many RAMBUS inmplentations. (for comparison some RDRAM implementations have been 16-bit, many currently are still that or 32-bit)

Now, Alpha is slowly slipping, but currently (aside from Power4) the only chip holding it's own against Intel/AMD. (based on a several year old core at that) The EV7 will be the last generation of alphas, without all of the features the EV8 would have had (and probably the performance crown for a LONG time)

(compiled from memory, it is 1:30, and post errors/debates as responses)

Re:Reminds me of compaq....

[Paul Komarek]

Last I heard, the important Alpha guys ended up at AMD or in small companies. I remember hearing that some or all of the DEC engineers working on StrongArm refused to move to Intel when Intel aquired that cpu, and that Intel had to move some people from the, geez, i810 or something similarly not StrongArm-ish into the StrongArm project.

At some point, I thought the important Alpha folks went to Nexgen, which was bought (directly or indirectly) by AMD.

I'd really like to have a more detailed and more accurate picture of who went where, but it's bedtime for me. Maybe someone else would like to elaborate.

-Paul Komarek

end of line already announced

[DZign]

If I remember correct about 2 months ago HP announced the end of life for the Alpha
(in about 2-3 years I believe).

So what does this news mean ? Just trying to sell
some more alphas ? How long will they support this system ?

Re:Dec Alpha

[Heavy Machinery]

Follow the link Dahan just gave you:
http://www.tru64unix.compaq.com/noncommercia l-unix /

for the hobbyist Tru64 licenses.

But, who needs Tru64 when there is VMS? If you want to try out a public access VMS account on the Internet, check out the "Deathrow" cluster (yup, VMS had true clustering decades before Micrsoft claimed to be able to do it) at:
http://deathrow.vistech.net/ [vistech.net]

-Chris

SPEC CPU2000

[Luminous Coward]

CINT2000 and CFP2000 results from the Standard Performance Evaluation Corporation website.

AlphaServer ES45 [hp.com]
1250 MHz
SPECint2000 = 928
SPECint_base2000 = 845
SPECfp2000 = 1365
SPECfp_base2000 = 1019
CINT2000 [specbench.org]
CFP2000 [specbench.org]

HP server rx5670 [hp.com] (Itanium 2)
1000 MHz
SPECint2000 = N/A
SPECint_base2000 = 807
SPECfp2000 = 1431
SPECfp_base2000 = 1431
CINT2000 [specbench.org]
CFP2000 [specbench.org]

Dell Precision WorkStation 530 [dell.com] (Xeon)
2800 MHz
SPECint2000 = 957
SPECint_base2000 = 921
SPECfp2000 = 887
SPECfp_base2000 = 878
CINT2000 [specbench.org]
CFP2000 [specbench.org]

Re:A good buy for some tech company?

[Luminous Coward]

CPU R&D is *really* expensive. That's probably why Intel doesn't seem to do much CPU R&D and sticks mainly to manufacturing [...]

Intel spends approximately 15% of their revenue [yahoo.com] on R&D (it is not clear how much is used specifically for CPU R&D). What makes you say that "Intel doesn't seem to do much CPU R&D"? On a side note, would you count compiler research as CPU R&D?

note: Itanic arch came mainly from HP [...]

The Itanic joke is getting old, Itanium 2 is now competitive [slashdot.org], especially in floating point applications. Do you have any evidence that HP was more involved in the design of IA-64 than Intel?

Re:Intel. not HP

[Anonymous Coward]

Correct to a point. Intel did buy the Alpha design team. All those engineers and developers are Intel employees now.

HP's involvement

[John Bayko]

Do you have any evidence that HP was more involved in the design of IA-64 than Intel?

This is what the Great Microprocessors List [sympatico.ca] has to say about it:

The design itself came from designers at HP who estimated in 1992 that complexity would prohibit more than 4-way issue PA-RISC designs. Also, HP had just bought Cydrome which had experience in designing VLIW systems, and engineers from VLIW producer Multiflow,. The decision was made that the PA-RISC would be replaced with a VLIW initially called SP-PA (Super Parallel Precision Architecture) or PA-WW (Precision Architecture-Wide Word). Intel, which had started fabricating PA-RISC CPUs for HP, was approached as a development partner to share the cost and increase its popularity.