Please consider a donation to the Higher Intellect project. See https://preterhuman.net/donate.php or the Donate to Higher Intellect page for more info. |
Itanium
The Itanium was a microprocessor produced by Intel in 2001–2002. It was the original implementation of the IA-64 architecture jointly developed by Hewlett-Packard and Intel. It was superseded by the Itanium 2. The original Itanium was not commercially successful, and only a few thousand were deployed. It served primarily as a software development platform for commercial IA-64 software for the Itanium 2.
History
HP and Intel first collaborated on developing a new processor architecture for servers and workstations in 1994, based on research done at HP starting in 1989. By the second half of the 1990s, HP sought a next generation replacement for its successful PA-RISC line of server and workstation CPUs, and wanted to tap Intel's volume and expertise in chip design and manufacturing.
IA-64 was based on Explicitly Parallel Instruction Computing (EPIC), in which the compiler lines up instructions for parallel execution. Itanium was designed to ensure compatibility with both Intel x86 and HP UNIX applications. During development, it was widely expected that IA-64 would become the dominant processor architecture for servers, workstations, and perhaps desktops, superseding the ubiquitous x86 architecture and providing an industry-standard architecture across an unprecedented range of computing platforms.
The original Itanium processor, code-named Merced, was released in 2001. Acceptance in the marketplace fell far short of expectations for a number of reasons:
- delivery was two years later than originally expected
- performance issues.
- lack of optimized software.
- yield problem and high manufacturing cost.
- difficulty in obtaining the Itanium CPU. Most of the early batches of CPUs were used on Massively Parallel Processors (MPP) Supercomputers with several hundred CPUs in a single machine.
- poor relative performance when running "legacy" IA-32 applications.
- general market slowdown in server sales following the dot-com bubble.
The original Itanium was not commercially successful. Intel and HP continued their development efforts, which led to the first release of the Itanium 2 processor in 2002, at which time the original Itanium was no longer produced or sold.
Hardware support
Several server vendors developed servers using Itanium, including Groupe Bull, Fujitsu, Fujitsu-Siemens Computers, Hitachi, Ltd., HP, NEC, Silicon Graphics, Unisys, IBM, and Dell, but some vendors suspended active marketing of their Itanium systems when the processors failed to materialize on schedule. Only HP and IBM delivered systems in quantity. Some vendors migrated to the Itanium 2 when it became available.
Some of these vendors joined with Intel and a number of software vendors to form the Itanium Solutions Alliance, with the goal of promoting the architecture and collaborating to accelerate software porting.
Software support
Itanium currently supports a very broad range of operating systems from Windows Server 2003, multiple distributions of Linux including Red Hat and Novell SuSE, HP-UX and OpenVMS from HP, and mainframe environments GCOS from Groupe Bull and NCOS from NEC. Further efforts are underway by Platform Solutins, Inc. to add IBM's mainframe operating systems OS/390, MVS and z/OS. Lastly, Transitive Corporation is in beta release of a translator to run Solaris/SPARC binaries on Linux/Itanium based servers.
At last count, over 10,000 applications are in production on Itanium based systems with the number having doubled each year from 2003 through 2006. In comparison, IBM's POWER-based systems offer approximately 6,000 applications across IBM's AIX and Linux.
Early Technical Challenges
The Itanium processor, code named Merced, shipped in June 2001. Manufactured in a 180 nm process, it was offered at speeds of 733 and 800 MHz, with a choice of 2 MiB or 4 MiB off-die L3 cache. In IA-64 mode, it was the fastest floating point processor available in the marketplace. However, on integer calculations it performed only slightly better than an equivalently clocked x86 design, and when running legacy x86 code, performance was only about 1/8th that of a similarly clocked x86 processor. Software emulation (as used later on the Itanium 2) would have been faster.
The main performance limitation with the Itanium processor was the high latency of its level three cache. Intel's engineers had evidently been hoping that the high bandwidth would offset this, but the latency was so high that the cache was not significantly faster than the main memory interface. With the faster first and second-level caches set relatively small (32 KiB and 96 KiB respectively), this further increased the load on the main system bus.
Compounding the performance impact of the lack of available cache bandwidth was the fact IA-64 code has a larger footprint than equivalent x86 code, so the number of instructions that could be contained in the cache was even smaller than the sizes alone would suggest. These effects might have been mitigated had Itanium been designed around a fast processor bus. However, at a mere 266 MT/s it was only equal to consumer Athlons of the period, and was a full 33% slower than first generation Pentium 4s. This was worse than it would appear, since Itaniums were intended for multiprocessor server systems where system bandwidth is paramount. Itanium clock speeds were also disappointing relative to the gigahertz speeds being delivered by the x86 architectures of the period.
The technical specifications indicate an original 1998–99 target launch date. The repeated and lengthy project delays meant the processor was out of date before it had began shipping. Hence, the Itanium was not a competitive product when launched, although it would have been two years earlier.
Despite these initial shortcomings, Intel and HP continued their development work. Merced was succeeded by the first Itanium 2 (code-named McKinley) in June 2002.
Itanic nickname
Itanic is a derisive nickname for the Itanium which refers to Titanic, the infamous ocean liner which sank in 1912. It was coined in 1999[1] and has often been used since then by The Register[2] and Scott McNealy. It alludes to the belief that the IA-64 architecture is a white elephant, which cost Intel and partner Hewlett-Packard many billions of dollars while failing to achieve expected sales in the originally projected timeframe.
Itanium has been the subject of a great deal of both hype and derision since its inception. Perhaps this is best understood within the context of its importance in the marketplace. The Itanium Solutions Alliance estimates the total value of the mission-critical market at approximately $28 billion, which is considerably greater than the total market for x86 solutions. In addition, the organization has publicly declared its goal of making Itanium the dominant architecture within that market. [3] Given how much is at stake for both proponents and detractors, the level of discourse is hardly surprising.
Competition
Itanium competed at the low-end (primarily 4-way and smaller systems) with servers based on Intel IA-32, and at the high end with IBM's POWER architecture and Sun Microsystems' SPARC architecture.
Itanium was touted as representing a departure from business-as-usual in the high-end computing market. The market leaders in that space, particularly IBM and Sun, traditionally built their solutions from the ground up, all the way from silicon and platforms, to software and support. Itanium represented an approach taken in the past by companies such as Sun, in which an open architecture allowed hardware and software products from a broader community of vendors to be integrated into each system. Over 70 vendors now offer Itanium based systems ranging from global vendors such as HP, SGI, NEC, Fujitsu and Unisys to more regionally focused suppiers such as Lenova of China, Kraftway of Russia and Groupe Bull of France. However, Intel retained exclusive control of the IA-64 design, manufacture and sale. Attempts have been made to portray competing proprietary RISC architectures as open. For example, Sun offered up the vendor independent SPARC specification, but only Fujitsu of Japan seriously pursued investments in SPARC processor design and manufacture. Similarly, IBM's creation of an industry consortium for POWER processors failed to attract commitment beyond Hitachi of Japan and longstanding OEM Groupe Bull of France.
By 2001, the “horizontal” approach was most successfully represented by the dominance of the x86 architecture, which as of 2002 accounted for more than 90 percent (by unit volume) of all servers sold. High-end systems remained relatively free from this horizontal mode of competition, largely because of the tight integration required to meet the highest performance, scalability and availability requirements. Over the years, that level of integration was most easily provided by a single vendor.
By 2001 the market needed a more horizontal approach. In particular, the cost of developing all the components of a comprehensive computing platform was enormous, and became a burden that even the largest companies were no longer equipped to bear. (As one example, the cost of a next-generation silicon fab had risen to between 3.5 and 5 billion US dollars.) Standardizing on a single architecture would allow hardware development and manufacturing costs to be amortized across a broader vendor community and product line. It would also increase the level of innovation and collaboration. Of course, it also creates additional challenges, as diverse vendors work to coordinate development and product delivery, simultaneously competing for a common customer base.
Intel positioned Itanium to break into high-end business and technical computing, in the hopes of duplicating x86's successful "horizontal" market. It's success to date was limited to a replacement for PA-RISC and Alpha in HP's "vertical" systems and MIPS in SGI's HPC focused servers. Adding to its failure to unseat POWER and SPARC was another complication: the x86 architecture's growth into the enterprise space. Given its existing economies of scale fueled by its enormous installed base, x86 was the preeminent "horizontal" architecture in enterprise computing.
Supercomputers
The fastest supercomputer based on Intel Itanium 1 processors was IBM's Titan NOW Cluster at National Center for Supercomputing Applications. CPUs: 320 Intel Itanium (800 MHz). Rmax: 677.9 Gigaflops. It was listed on the TOP500 list from November 2001 (position #34) until June 2003 (position #111).
External links
- Itanium: A Cautionary Tale Intel takes over the world - with the lowly x86
- Itanium-Is there light at the end of the tunnel?
- Analyst firm offers rosy view of Itanium
- Customer Perceptions of the Future of Itanium
- End-Users' Feedback: Transform IT and Increase Business Performance Through Itanium-Based Standardization
- Itanium Solutions Alliance
- Intel Itanium opened cartridge processor images at cpu-collection.de
- Intel Itanium technical specifications
- Intel Itanium developers' manuals
- Intel® Itanium® Architecture (Volume 3: Instruction Set Reference)