安装Montvale芯片的主机已经面市。Montvale是Itanium架构开启多核时代之后的第一次升级产品，其制程没有改变。主要变化在于：

1. 相比Montecito略微提高了主频，性能最多可提升19%，浮点计算差不多提升了12%。

2. 在可靠性方面做到内核级锁步。内核级锁步实现了两个锁步内核作为一个逻辑CPU内核，通过消除内核中未检测到的错误来提高核内数据的完整性。这项技术与现有的插槽级锁步技术相结合，可确保计算结果在内核与插槽间保持一致。

3. 按需配电技术。服务器在很多情况下都处于利用率不高的状态，这项技术可以在处理器未处于高负荷运转时，自动降低主频，从而实现动态降低处理器功耗，最终降低能耗和散热成本。有统计数字显示，通过这样的技术，用户节省下来的电费最高可达24%。

Tukwila作为2008年要面市的芯片，Intel公布了一些设计规格。Tukwila将是Itanium架构首款4核处理器，使用65nm制程工艺制造，性能约是Montvale的两倍，但不会明显提高主频。Tukwila采用了大型片上高速缓存、集成内存控制器及全新的高速互连设计。 集成的内存控制器直接采用全缓冲技术,同时采用多条点对点串行总线跟外界通信，会将处理器对内存的带宽以及处理器对外通信的带宽提升到一个新高度。除了将采用与未来英特尔至强处理器平台相同的芯片组之外，Tukwila还将引入多项RAS特性和虚拟化功能。

Tukwial的架构示意图如下：

Poulson作为Tukwila的下一代，初露端倪，其将采用4个以上内核的安腾处理器，并且利用32nm的制程工艺来制造。这是自2002年英特尔安腾2处理器以来的又一款全新架构，其并行指令度会再次提升。凭借更多内核、线程及每周期指令，Poulson在可扩充性、性能、可靠性及灵活性方面得到很大提升的“概率”相当高。采用全新架构的Pousoln的示意图如下：

    10月9日，英特尔Itanium Processor Architect首席工程师Cameron McNairy在Gelato Itanium Conference and Expo上表示，“我们没有放弃提高单核心性能，在推出每一代Itanium处理器时都提高了单核心的性能。”

    在谈到 Itanium芯片由Itanium2 9M 发展到双核心的Itanium2时，McNairy表示，“尽管核心基本相同，但我们进行了一些变化。例如，我们将指令快取和资快取分开，简化数据快取冲突的处理，并增加核心的资源。”

    McNairy表示，当Itanium处理器由Montecito发展到Tukwila时，每个核心的性能仍然会有所提高。他还说，英特尔还在考虑提高多线程技术的效率，但没有披露更进一步的详细资料。

     McNairy指出，尽管提高单个核心的性能旨在提高整个处理器的性能，但这种方法并非适用于所有负载。

    英特尔的产品发布计划显示，它将于今年晚些时候发布Montecito的后续产品Montvale。Tukwila则将于2008-2009年间上市销售，Poulson的上市销售时间为2010-2011年间。

HP Completes Montecito Itanium Rollout into Integrity Servers

Server maker Hewlett-Packard Co today completes the revamping of its Itanium-based Integrity server line with the rollout of new machines that employ a home-grown chipset and Intel Corp's

The Montecito chips were announced in mid-July, and bring the Itanium processor line to more or less performance parity with other RISC chips, thanks in large part to the move to a dual-core architecture. HP has been in sore need of the Montecitos for several years to compete with IBM Corp.

The Arches servers use the new dual-core Montecito chips or the single-core Madison Itaniums, but it is hard to imagine why anyone would buy a new Arches machine equipped with the Madisons. Back in March, however, some customers that were hard-pressed for larger main memory or higher I/O bandwidth might have done so, to be sure, since the switch from the "Pinnacles" sx1000 chipset to Arches chipset--using the same Madison processors--could boost performance by as much as 30%.

The Arches chipset supports 2 GB DDR2 DIMM main memory, which doubles the memory capacity of the machines and speeds memory access while lowering energy consumption, and next year, when HP supports 4 GB DDR2 DIMMs, the memory capacity will double again on the Arches boxes.

Still, the Arches design is really balanced for the dual-core Montecito chips, which offer double the performance of the Madisons, clock for clock, and now that HP has certified these chips on the Arches machines, only customers that have the most stringent testing and certification requirements and that have not yet been able to test the Arches-Montecito combination will buy Madisons for these machines.

While HP is happy to tell customers that they can finally get Montecito chips in their Arches machines, the stars of the launch event today are the rx3600 and rx6600, two new midrange boxes that employ the "Titan" zx2 chipset and the Montecito processors.

The "Pluto" zx1 chipset was originally created for workstations and entry servers using Itanium processors. But the advent of dual-core processors means that all server makers can push the envelope with entry server chipsets and make a very cost-effective and powerful midrange box. This is precisely what IBM has done with the p5 550, which is a two socket server that uses dual-core Power5 and Power5+, and the p5 570, which can glue together up to four of these boxes with NUMA-style clustering to create a machine that spans from 2 to 16 cores.

While the new rx3600 and rx6600 do not span quite as far, the overlap in performance and, as it turns out, bang for the buck is quite close from 1 to 8 cores. That is no accident, of course. That was HP's design objective.

According to Tim Danielson, worldwide product line manager for the Integrity line at HP, the company is figuring that customers are going to find the revamped Integrity line to be pretty attractive when the Titan and Arches chipsets are combined with the Montecito Itanium 9000 processors.

"We think the change to Montecito will be pretty quick," says Danielson. "All of the economics really encourage the move." He says that HP has revamped the line to offer twice the performance or more at the same or lower price point as the equivalent Madison-based products.

And, significantly because HP has such a large installed base of midrange HP-UX and OpenVMS machines out there in the world, the new rx3600 and rx6600 machine will be particularly appealing.

"We look at four-core and eight-core as being the sweet spot in the market," says Danielson. "And if you look at the space where the rx6600 plays, we are now the leader in performance and price/performance."

He says that the rx6600 will best the equivalent IBM p5 system by 20% or so in terms of raw OLTP performance and do so with 40% better bang for the buck. Larger rx8640 machines with Montecito offer about 30% better value than IBM's Unix boxes, according to Danielson.

Helping that price/performance along with the rx3600 and rx6600 is the fact that HP is finally using components from the high-volume ProLiant line in the Integrity machines. The two use the same SAS disk drives and the same DDR2 memory cards, although HP does some extra testing on the cards that go into Integrities.

Using common parts means HP can charge less money for an Integrity box. For the past several years, if you based a comparison of Unix servers on list prices, the Integrities did not fare so well against the IBM pSeries and p5 lines. And with Sun catching up with the dual-core UltraSparc-IV+ chips and aggressively selling its "Galaxy" Opteron machines against entry and midrange HP-UX and OpenVMS machines, HP could not charge the kinds of premiums it got away with for the past three years.

Back in July, Computerwire previewed some of the salient characteristics of the rx3600 and rx6600 machines, and the rumor mill had them pegged just about right.

The rx3600 is a two-socket, 4U rack-mounted server, while the rx6600 is a four-socket, 7U rack-mounted machine. Both use the zx2 chipset, which has three components: a memory and I/O controller, an I/O adapter, and a scalable memory expander. The memory and I/O controller links the Itanium processors to the processor bus and contains a main memory and a cache memory controller.

The scalable memory expander is an adjunct chip that HP created to increase both the memory bandwidth and the memory capacity of the machines that use the zx2 chipset. The I/O adapter part of the chipset is used to create the PCI, PCI-X, and PCI-Express buses in a system, which are used to link peripheral adapter cards into the processor complex.

While HP could have supported Montecito chips in the zx1 chipset, the use of dual-core chips, a large L3 cache for each core (up to 12 MB), and the addition of VT hardware-assisted virtualization. HT multithreading, and Intel Cache Safe Technology (formerly known at Intel as "Pellston") really necessitated a new chipset.

The rx3600 is currently available with the 1.4 GHz/12 MB version of the Montecito chip (that's 6 MB of L3 per core), but sometime in the fourth quarter, HP will add support for the more powerful 1.6 GHz/18 MB Montecito (that's 9 MB per core, which is exactly the same clock speed and cache available on the Madison chip).

The rx3600 spans from 2 GB to 96 GB of main memory, and has eight hot-swap PCI slots. It ships today with PCI-X support, and will be upgraded to PCI-Express support in the first quarter of 2007, presumably with a microcode upgrade and not a motherboard swap. The system comes with an eight-port SCSI controller with RAID 1 mirroring, which is used to support HP-UX Unix and OpenVMS. Customers that want to use Windows or Linux on the boxes have to add in a SmartArray controller from the ProLiant line, which has RAID 1, 5, and 6 support.

The rx3600 has two Gigabit Ethernet ports and room for eight hot-plug, 2.5-inch SAS disk drives. HP is shipping 73 GB SAS disks today in the rx3600, and will begin shipping 146 GB units in November. The rx3600 also has redundant power and fans (both swappable), and will be available as a tower server in the first quarter of 2007 for midrange customers that do not want racks. The base rx3600, which comes with a single 1.4 GHz Montecito chip and 2 GB of memory, costs $10,531.

The rx6600 server, with 7U of space, has a lot of room for expansion. This is something that midrange customers always want, because they do not like the disruption that comes from swapping boxes all the time. The rx6600 supports three different Montecito variants: the top-bin 1.6 GHz/24 MB part, as well as the 1.6 GHz/18 MB and 1.4 GHz/12 MB versions.

The rx6600 supports from 2 GB to 192 GB of main memory, and has the same eight I/O slots and two Gigabit Ethernet ports used in the rx3600. It also comes with the same SCSI RAID 1 controller for HP-UX and OpenVMS and requires the same upgrade to the SmartArray controller for Windows and Linux users. However, because the chassis is much larger, the rx6600 can support up to 16 hot-swap 2.5-inch SAS drives. A base rx6600 with one 1.4 GHz Montecito chip and 2 GB of memory costs $14,771.

Both the rx3600 and the rx6600 support HP-UX 11i v2 with the September 2006 update from HP. Red Hat Inc's Linux 4 Update 3 is also supported, and support for Novell Inc's new SUSE Linux Enterprise Server 10 is expected in the fourth quarter.

Microsoft Corp's Windows Server 2003 Enterprise Edition can also be installed on these boxes, as can the new OpenVMS 8.3 release from HP. The good news is that applications written for OpenVMS 8.2 are automatically certified for OpenVMS 8.3. Whatever changes HP made to the new OpenVMS are transparent to applications.

In addition to launching the servers based on the Titan chipset and getting Montecito processors into the machines with the Arches chipset, HP also announced today that customers can drop Montecito chips into its entry rx2620 servers and its midrange rx4640 machines, which use the older zx1 chipset. A base rx2620 costs $4,884, while a base rx4640 costs $15,614.

A base rx7640, incidentally, costs $33,058, a base rx8640 costs $74,725, and a base Superdome costs $209,389. The exact feeds and speeds of these machines were not available at press time.

HP says that it will have a processor upgrade to the next generation of the Itanium processors--the "Montvale" kicker to Montecito--about a year from now. Montvale is expected to have a faster front side bus, even larger caches, and higher clock speeds thanks to improvements in the 90 nanometer chip technologies used to make the Montecito chip.

Montecito supports a 400 MHz or 533 MHz front side bus, and Montvale is expected to get a 667 MHz bus. Montecito was originally supposed to have a 667 MHz bus and clock close to 2 GHz, but for some reason Intel backed off on the speed. 

The top 10 dead (or dying) computer skills
Are your skills in need of upgrading?

By Mary Brandel

Those in search of eternal life need look no further than the computer industry. Here, last gasps are rarely taken, as aging systems crank away in back rooms across the U.S., not unlike 1970s reruns on Nickelodeon's TV Land. So while it may not be exactly easy for Novell NetWare engineers and OS/2 administrators to find employers who require their services, it's very difficult to declare these skills -- or any computer skill, really -- dead. (Readers have their own views on dead and dying skills. Others offer their own suggestions for the pyre.)

In fact, the harder you try to declare a technology dead, it seems, the more you turn up evidence of its continuing existence. Nevertheless, after speaking with several industry stalwarts, we've compiled a list of skills and technologies that, while not dead, can perhaps be said to be in the process of dying. Or as Stewart Padveen, Internet entrepreneur and currently founder of AdPickles Inc., says, "Obsolescence is a relative -- not absolute -- term in the world of technology."

1. Cobol
Y2k was like a second gold rush for Cobol programmers who were seeing dwindling need for their skills. But six-and-a-half years later, there's no savior in sight for this fading language. At the same time, while there's little curriculum coverage anymore at universities teaching computer science, "when you talk to practitioners, they'll say there are applications in thousands of organizations that have to be maintained," says Heikki Topi, chair of computer information services at Bentley College in Waltham, Mass., and a member of the education board for the Association for Computing Machinery.

And for those who want to help do that, you can actually learn Cobol at Southern Illinois University at Edwardsville, which according to Mary Sumner, a professor there, still offers a Cobol course. "Two of the major employers in the area still use Cobol, and for many of their entry-level jobs, they want to see that on the transcript," she says. "Until that changes, we'd be doing the students a disservice by not offering it." (see also: "Cobol Coders: Going, Going, Gone? ")

2. Nonrelational DBMS
In the 1980s, there were two major database management systems approaches: hierarchical systems, such as IBM's IMS and SAS Institute Inc.'s System 2000, and network DBMS, such as CA's IDMS and Oracle Corp.'s DBMS, formerly the VAX DBMS. Today, however, both have been replaced by the relational DBMS approach, embodied by SQL databases such as DB2, Oracle and Microsoft SQL Server, says Topi. "The others are rarely covered anymore in database curricula," he says.

3. Non-IP networks
TCP/IP has largely taken over the networking world, and as a result, there's less demand than ever for IBM Systems Network Architecture (SNA) skills. "It's worth virtually nothing on the market," says David Foote, president of Foote Partners LLC in New Canaan, Conn. Foote tracks market pay for individual IT skills, which companies usually pay as a lump sum or a percentage of workers' base pay, either as a bonus or an adjustment to their base salary. SNA, Foote says, commands less than 1% premium pay. "It's like a penny from 1922 -- there has to be someone who wants to buy it."

Despite the fact that many banks, insurance firms and other companies still have large investments in SNA networks, the educational offerings in this area are also rare, according to Topi. "The dominant model of protocols is TCP/IP and the Internet technologies," he says.

4. cc:Mail
This store-and-forward LAN-based e-mail system from the 1980s was once used by about 20 million people. However, as e-mail was integrated into more-complex systems such as Lotus Notes and Microsoft Exchange, its popularity waned, and in 2000, it was withdrawn from the market. According to Foote, "cc:Mail is a bygone era. Now e-mail is tied into everything else, and cc:Mail didn't make that leap." Just the same, the product continues to be commercially supported by Global System Services Corp. in Mountain View, Calif.

5. ColdFusion
This once-popular Web programming language -- released in the mid-1990s by Allaire Corp. (which was later purchased by Macromedia Inc., which itself was acquired by Adobe Systems Inc.) -- has since been superseded by other development platforms, including Microsoft Corp.'s Active Server Pages and .Net, as well as Java, Ruby on Rails, Python, PHP and other open-source languages.

Debates continue over whether ColdFusion is as robust and scalable as its competitors, but nevertheless, premiums paid for ColdFusion programmers have dropped way off, according to Foote. "It was really popular at one time, but the market is now crowded with other products," he says.

6. C programming
As the Web takes over, C languages are also becoming less relevant, according to Padveen. "C++ and C Sharp are still alive and kicking, but try to find a basic C-only programmer today, and you'll likely find a guy that's unemployed and/or training for a new skill," he says. (see also: "Hot Skills, Cold Skills ")

7. PowerBuilder
Recruiters that have been around since the 1990s, such as David Hayes, president of HireMinds LLC in Cambridge, Mass., remember when PowerBuilder programmers were "hot, hot, hot," as he says. Developed by Powersoft Inc., this client/server development tool in 1994 was bought by Sybase Inc., which was once a strong Oracle competitor.

Today, PowerBuilder developers are at the very bottom of the list of in-demand application development and platform skills, with pay about equal to Cobol programmers, according to Foote. Nevertheless, the product keeps on trucking, with PowerBuilder 11 expected this year, which has the ability to generate .Net code. (see also: "35 Technologies that shaped the industry ")

8. Certified NetWare Engineers
In the early 1990s, it was all the rage to become a Certified NetWare Engineer, especially with Novell Inc. enjoying 90% market share for PC-based servers. Today, however, you don't have to look far to find CNEs retraining themselves with other skills to stay marketable. "It seems like it happened overnight," Hayes says. "Everyone had Novell, and within a two-year period, they'd all switched to NT." Novell says it will continue supporting NetWare 6.5 through at least 2015; however, it has also retired several of its NetWare certifications, including Master CNE and NetWare 5 CNE, and it plans to retire NetWare 6 CNE. "Companies are still paying skill premiums for CNEs, but they're losing value," Foote says.

9. PC network administrators
With the accelerating move to consolidate Windows servers, some see substantially less demand for PC network administrators. "You see the evidence for that in the demise of those programs at the technical and two-year schools and the loss of instructors," says Nate Viall, president of Nate Viall & Associates, an AS/400 (iSeries) recruiting company.

10. OS/2
A rough translation of OS/2 could be "wrong horse." Initially created by Microsoft and IBM and released with great fanfare in 1987, the collaboration soon unraveled, and after repeated rumors of its demise, IBM finally discontinued sales in 2005. OS/2 still has a dedicated community, however, and a company called Serenity Systems International still sells the operating system under the name eComStation. (see also: "IBM, Bankers at Odds Over OS/2 Migration Path ")

This Article Reprinted Courtesy of http://www.computerworld.com

http://www.globalknowledge.com/training/customgo.asp?find=P38_FEATURE&id=18488&pageid=29&country=United+States

http://www.techworld.com/opsys/features/index.cfm?featureID=3503

The great Itanium revival

Intel explains how the Itanium has a future -- honest!

By Manek Dubash, Techworld

The Itanium is great, it's got a future, and it's selling pretty well -- considering. This is Intel's latest war-cry, as it bids to resurrect the fortunes of a chip that many IT industry observers have all but written off.

The reality has been less striking, and the processor appeared to struggle to gain market traction against entrenched opposition from Sun's SPARC and IBM's Power chips. That view was reinforced when Dell and IBM dropped support for the Itanium, leaving co-developer HP as the Itanium vendor with the biggest sales, by a long way.

Then the first dual-core Itanium 2, codenamed Montecito, was late, and Montvale, the Montecito update with a faster clock speed, larger caches, and a faster front-side bus won't arrive until the fourth quarter of 2007.

Itanium set for revival?

However, all is not lost we hear, as Intel's Itanium engineering director Rory McInerney told your reporter that Itanium is in fact now back on track. He also revealed slightly more of the Itanium roadmap, and explained that, in fact, Itanium is actually selling rather well.

If you keep in mind that, unlike the volume server market where Xeon rules, there are two other chips out there, the fact that the Itanium has come from zero market share to take a claimed one-third of the market is quite impressive, if Intel CEO Paul Otellini is to be believed.

"We looked at the unit ramp of Power, Sparc and Itanium set to time zero. We're right on their curves from first shipments, and transitions are necessarily slow when it comes to large servers," Otellini said. "In the mainframe space, you're displacing 10- to 20-yearold architectures; it's a marathon."

In other words, for a mainframe-style chip, given that the replacement rate is dead slow -- you don't rush an buy new mainframes every three years -- Itanium is following the growth pattern of Sparc and Power.

McInerney fills in the details, saying that, in Europe, researcher IDC's numbers show Itanium taking share from Sparc and Power -- and that it's garnered almost one-third of the market. He says that yearly system sales will be 80 per cent of Sun's Sparc and 75 per cent of IBM Power by the end of 2007. If you look at Gartner's figures, European sales of Itanium are 80 per cent of Sparc sales, says McInerney. That's still not quite one-third of the market, but it's none too shabby either.

Further, adds McInerney, software support is growing, with applications such as Oracle and SAP adding weight to the case for Itanium. For example, Oracle has reiterating its support for Itanium and plans to certify the next major releases of database and Oracle Fusion middleware products for the processor.

"Our investment of $10 billion in the Itanium Software Alliance is paying off because we're now finding software such as utilities that are being developed," says McInerney. "There are 12,500 Itanium applications, 4.5k run on HP/UX, 800 on OpenVMS, and rest are either Windows or Linux."

The roadmap

McInerney buttresses his case for the future of Itanium by pulling back the curtain, just a little on the chip's future roadmap.

If there's a theme, it's more cores and bigger caches, along with faster interconnects between cores. And Intel will put the memory controller on the die, AMD-style. What's more the notion of the common Xeon/Itanium platform lives: McInerney reiterated Intel's commitment to the idea.

He admitted that one of the consequences of AMD's successful competition with the Opteron chip was that Intel was forced to speed up Xeon development. This meant engineering resources got diverted away from Itanium, one of the casualties being the common platform. But that's all back on track, we're assured.

"There's also space for a product in between Xeon and Itanium - it's good for time to market", he says. It's all about a common chipset that allows hardware vendors to build a single system that can house either a Xeon or an Itanium.

What does this all mean? Tukwila, the next rev proper of Itanium, is due late in 2008 or early 2009. It'll be based on 65nm technology, will have four cores and more, and integrated memory controllers to keep the cores coherent. It'll also include Hyper-Threading,, higher-speed interconnects and memory error correction system that can contain and mark two DRAM errors. It will support the common platform, and will include further virtualisation support, namely Intel VT for Directed I/O. About 50 per cent of the die area will be cache memory, said McInerney, although he still wouldn't spill on how big it will be. Tukwila will deliver twice the performance of Montecito, he promised.

After Tukwila comes Poulson. Due in 2010/11 and socket-compatible with Tukwila, it'll skip 45nm technology and be among the first chips to be based on a 32nm process. It'll get four or more cores -- McInerney remained coy on how many, though the word "twelve" did escape his lips when he was discussing it in a general fashion. He then said: "We may not throw a linear progression of cores at the problem, that is, there might not be binary growth in the numbers of cores." Poulson will include multi-threading enhancements, and new instructions to take advantage of parallelism, especially in virtualisation. The cache will be "massive".

"The idea is for this core to be the base for next one or two generations of product, to get it right. We'll be on a beat rate for cores," he says. It might even be called Itanium 3, although McInerney didn't say so.

Then comes Kittson, with an "enhanced multi-core architecture". McInerney said he couldn't provide further details as the feature set hadn't yet been decided upon. "We're only just entering definition now," said McInerney. At a guess, you'll see this chip in 2013-2015.

So Itanium is gaining share in a slow-growing market, and the roadmap is back on track, re the key messages that Intel was keen to put over. The sub-text, not so hidden beneath the surface, was that the chip vendor won't be putting the Itanium out to grass anytime soon, and that software vendors, hardware OEMs and -- most importantly -- end users needn't fear for the future of Itanium-based systems.

So if you're in the market for a big computer that's designed to last for the next five to 20 years, that'll be manna for your ears.