Slashdot Mirror
HP, Intel Call it Quits on Itanium Partnership
envisionary writes "Hewlett-Packard Co. and Intel Corp. have ended their partnership to co-develop the Itanium 64-bit processor line, according to a report from Reuters. The move follows disappointing sales for servers based on the processor, according to the report. Intel and HP developed the processor about 10 years, but the chip has been a flop due to delays, cost overruns and lackluster demand."
Now that HP won't be selling the Itanium, will they bring back the alpha? I heard they never sold the last model because it would compete with the Itanium.
Actually, they had it twice... Alpha AND PA-RISC.
*sigh*
- IBM selling its PC business
- Cell workstations
- POWER5 amazing benchmark records
- IBM incents Linux on Power app development
- Launches a Power architecture coalition
- IBM and Red Hat begin certifying apps for Linux
- IBM ups its Desktop Linux push
I know it's tinfoil hat talk, but I must wonder if IBM isn't about to make an end run around Intel AND Microsoft for a new generation of desktop computing...Linux on POWER.HP getting out of bed with Intel could free it up from certain obligations it had to them and open them up to using the Power architecture.
I know, I know...it's just too crazy to think it's anything more than coincidence...
The supposed quote was about 640K, not 64K, and it's a myth
That's not my point. My point was the performance increase 64-bit processing gives was not enough to justify the increased cost of that chip. As another poster in this thread points out, if you have to use 64 bit processing for whatever reason, Sun's Sparq chips where even better. Then AMD released their 64bit CPU lines...
Any sufficiently advanced influence is indistinguishable from control.
I expect AMD to become bigger than Intel in the next 10 years
That is quite naive. AMD has close to 15,000 employees and is a $9 billion company. Intel has 85,000 employees and is a $140 billion company. This doesn't change overnight, and yes, 10 years is overnight when you consider companies of this size. Intel will have to make several more mistakes like Itanium. Plus AMD has a long ways to go to match the manufacturing capabilities of Intel.
I was being a bit lazy in pointing to my own comment [though it did contain the most current development ] on HP reducing its diverse hi performance options. Slashdot has covered this topic every step of the long, sad way: /. mention
HP pruning the OS proliferation, HP repeats: alpha is really going away! and alpha chip to be discontinued
oh, what a small world...just noticed HP's dropping TRU64 was a gain for Veritas which was the subject of a recent
SLASHDOT: news for people who can't concentrate on work or have no life at all and got tired of yelling back at the TV.
I'd love to see Itanic turning to an "open" architecture, instead of dying altogether. That probably isn't going to happen, so we can expect to see the pain going on for a long while, with Intel downplaying the long term importance of the chip to the company. It's going to go the way of SPARC - a dead chip walking, with only the manufacturer being interested in it anymore.
First, the Itanium processor line is going nowhere. Just the HP and Intel partnership which means that other vendors will have as good of pricing as HP, so it will be more "open".
Maybe I'm just ignorant and biased, but the Itanium based HP servers that I have worked with are the fastest and best built 64bit servers that I have worked with to include Alpha, Sparc, and IBMs power chips. The Itanium chips appear 10 times in the top 100 supercomputers including numbers 2 and 5 on the list.
Intel recently released their low voltage offerings of the Itanium processor with speeds up to 1.4GHz.
There have been issues since the Itanium's release. The first chips, Itanium1's, did not perform well. The current Itanium2's are power hungry and require tons of cooling. Now they are excellent performing and have lower power and cooling requirements.
And people are interested in the architecture. For number crunching or a fast database, there are few better offerings. The Opteron or current IBM chips appear to be the closest contenders. If Intel were to lower the price of the Itaniums and continue improving their performance and reducing their power needs, I think these things are going to be around for a while.
It's too bad HP didn't already have a long term successful 64 bit chip, all the engineers that designed it from the ground up, and 10 years of history with something like the DEC Alpha chip.
From someone who left HP R&D last year (not WFR'd, got tired of my job and moved to a different company):
They had better than that. The folks that just joined Intel designed PA-RISC processors before moving to Itanium (sometimes refered to as PA-RISC 3). Those folks are top notch designers that have shipped succesful microprocessor products for 20 years. They already saved Intel's ass a few times in the Itanium collaboration. They designed Itanium2/McKinley entirely, and the upcoming Montecito is mostly a Fort-Collins design that replaces yet another Intel project failure (the codename and some of the most unpleasant parts of the design is all that remains), similarly it is rumored that Intel's Tukwila design (from the "famed" Alpha folks) is being ditched and will be replaced by yet another rescue design from Fort-Collins.
One of their managers was fond of saying that those folks could create a Sparc processor that would top the performance charts (Sparc has been performing pretty poorly for the last 10 years). I believe that.
Back when Alpha was in competition for the best performing microprocessor, its only real competition was PA-RISC. It was a leapfrog game between the two architectures. Since PA-RISC was never 'sexy' (few HP products are), the public only remembers Alpha, but reality was different.
As to keeping Alpha, remember that PA-RISC had a marketshare about 5x the one of Alpha (~30% of the Unix volume, both in volume and revenue; Alpha was stuck around 5-6%). And PA was already on the way out when HP acquired Compaq. Not only that, but most Alpha folks had already left Compaq by then (mostly to Intel as a group, and individual to other companies). So Alpha was never an option for HP. So please stop spreading this myth that HP killed Alpha. DEC/Compaq killed Alpha before the merger.
Regarding run by a complete loser of a woman with the sole intent of systematically destroying the company, I had the pleasure of working with some former engineers from "the old DEC". Some of them are excellent people. Overall, however, I saw a lot of bad attitudes that could sink a tech company. NIH, no concern for deadlines, shipping a real product or customer experience. Some of those folks are so wrapped in the memory of the golden years when DEC R&D was perceived as the best in the industry (while the real top talent has already moved on or retired), they don't realize what makes a company tick (pleasing customers).
I am sorry to say that, while DEC/Compaq (and now HP) management might not have helped, I believe that the Alpha/Tru64 R&D folks played a good part in the killing of their products, by being a bit too much convinced of their own greatness and failing to see that this technical greatness did not help their customers. Their toys did not win in the marketplace, outsold by less sexy widgets built by less arrogant folks (including the PA-RISC and HP-UX teams). I do not like Carly much but she and her team are probably saving HP and what remains of DEC by keeping such bad attitudes in check (by cutting the teams that do not deliver). Being 'sexy' doesn't help much in the marketplace, and it doesn't seem to help much in HP anymore. Good, too much money was wasted on sexy things.
People are rooting for the underdog, cheering AMD and boohing Itanium, longing for the good old time of technically pure Alpha. Yet Itanium is a very clean design compared to x86 / x86-64, and people forget some of the crap associated with Alpha (lack of byte loads on first generation processors, WTF !?). Such selective blindness is ok for teenagers in their basements, unfortunately they seem to be held by more senior folks who should know better.
Our industry is in a pretty sad state. Perception, mindshare, hype and FUD matter a lot more than they should.
ARGH...
.nee. Alpha Processor Inc. The company that Samsung "ran".
Read my lips. Alpha is DEAD. Expired. Pining for the fjords.
Samsung has ZERO interest in investing in it any further. They'll keep making them only as long as HP buys them, not one day longer.
Jesus, it's over people. Almost all the Alpha developers are working in either Hudson, MA for Intel or Boxboro, MA for AMD.
How do I know? I worked at DEC for 18 years. Then at API Networks
If Intel had just gotten over the Not Invented Here syndrome and ponied up enough money, Alpha would still be going. But they didn't.
Repeat after me, Alpha is Dead.
Sheesh.
What's my Karma Mr. Burns? "Excellent"
You forgot to include Itanium.
It isn't going away. HP will cease to co-develop it with Intel. They will continue to design them into their servers.
"What's the frequency Kenneth?"
1) The MIPS architecture dead-ended. It couldnt scale far enough to keep up with x86.
2) PC 3d hardware eventually surpassed anything SGI was capable of designing.
3) SGI had an opportunity to migrate to x86, but didn't take it seriously enoguh and blew it. Now they're paying the price.
4) For a while there, they couldn't decide if they wanted to be a server company, a workstation company, or a desktop x86 PC company.
5) They laid off most of the engineers of their proprietary hardware. A lot of it was undocumented, so nobody was left who knew how to make anything of SGI's work.
All of these spelled doom for SGI's traditional market. They've since been shoehorned into a tiny niche of supercomputers. They "still sell" workstations and servers, but nobody buys them for anything new anymore. It's simply legacy support for existing IRIX customers.
Well that cache miss comment is about the most false statement I've seen in a long time.
:)
:)
;) Registers aren't really that big a deal, either. Most compilers don't use all of the 32 most RISC cpus have, which is only the logical registers not the physical (of which there are many more, on par with the EPIC machines). Sure, the 8-ish of IA-32 is woefully inadequate, but that doesn't mean you needs gobs of them.
:)
/.
Woops! Right, that's a big gaff. I should have said that the problem is finding the misses, in part due to the stall-on-dependency (of course, it's in-order) and also loads bypassing stores with uncomputed addresses. I.e. you'd like to move your load as far forward in the execution stream as possible so you find the miss sooner, but you can't move it in front of a store because it might conflict. I have no excuse; I just recently talked with someone researching ways around this problem (and I may still be misrepresenting it). Oddly their funding (from HP) was canceled.
Unless of course you hit in the 1-cycle L0d cache
Of course because that's not a miss.
In-order execution certainly puts more burden on compilers, but it frees up a ton of area for registers, functional units, and cache.
Eh, in theory. First, area clearly isn't a big issue with Intel and Itanium.
And functional units aren't a big deal at all, because neither the compilers nor OoO schedulers are able to find enough ILP to use them. They're mostly idle anyway and adding more is just wasting area. The exception is highly parallelizable FP code, but the solution there is to add vector units (which gives you more bang per unit of fetch bandwidth than non-vector in-order scheduling).
As to cache, look at any server part. On Itanium, Xeon MP, and even to a lesser extent Opteron with only 1MB of L2, the die size is dominated by the cache. Getting rid of schedulers and reorder buffers isn't going to allow them to add a significant amount of L2/L3 cache. The size of the L1 cache is a tradeoff between miss rate and access time, not miss rate and area.
Fundamentally, when you're willing to make chips that are 600mm^2, the area benefits of in-order don't really buy you much.
The real benefit, as I understand, of in-order was supposed to be complexity. Remembering Merced, I'm not sure that ever panned out either.
Regardless, this is all a case of reality trumping theory. The original plan for Willamette (P4) had a 32K L1 cache running on the same (double pumped) clock as the ALUs, but strangely it couldn't work at that speed. If K8 had come out when it was originally planned at the speeds planned, we'd already be singing the eulogies for Itanium and Xeon. Also, if I'd have just been bitten by that radioactive lemur I'd be out fighting crime from the treetops instead of working a day job and posting on
God damn reality.
The enemies of Democracy are
The original Pentiumn was a separate generation (equivalent to 586). The Pentium Pro, Pentium-II, and Pentium-III were all based on the same core and so were another generation (686). The Pentium-4 was an entirely different core design and was yet another generation. If we were still numbering according to the original scheme, the P4 would be the 786.
Have you ever tested the Itanium chips? I have and they are significantly faster then the ALPHA and PA-RISC chips. (Sorry, can not publish numbers to back this up). However, that does not change the fact that HP and Intel made a major design mistake with itanium by not building in native ia32 support. They only have emulation of ia32 which is sloooowwwww.