It’s a whale of a chip. Intel is announcing today the 10th generation of its Itanium chip, which is a 64-bit microprocessor aimed at the very high end of the corporate computer market.
The chip is a modern miracle in many respects, with more than 3 billion basic components known as transistors. It is one of the highest achievements in technological design, but it’s also a relic of an age when power consumption didn’t matter. As such, the chip is confined to a small, high-end segment of the computing market at the high end.
Not that the high end isn’t lucrative. Itanium is attacking the reduced instruction set computing (RISC) market, which generates about $15 billion in revenues a year.
But Itanium launches, while technically impressive, are always tinged with regrets about what might have been. When Intel began working on its 64-bit chip architecture with Hewlett-Packard in 1994, the goal was to create a chip that would become the future of all computers. It was supposed to debut in 1999, but the first chip, code-named Merced, came out in 2001, just in time for the 9/11-instigated recession.
The first Itanium was power-hungry and slow at running 32-bit applications compared to Intel’s 32-bit Pentium 4. Software makers had to recompile their programs — a lot of work — to get them to take advantage of 64-bit computing.
Intel continued to invest, even though the potential market was diminished. It did so as a challenge to RISC chip vendors such as Sun Microsystems, Silicon Graphics, and IBM. Some of the rivals capitulated rather than invest billions in their own chips. But Intel’s biggest competition remains its still-successful 32-bit Xeon family of server chips, which are made in very high volumes.
The code-named Poulson chip that Intel is unveiling at the International Solid State Circuits Conference today is a big leap forward from Tukwila (pictured at top), which was released as the Itanium 9300 in early 2010. Tukwila was very late, but Poulson has stayed on schedule, said Rory McInerney, vice president of the Intel Architecture Group, in a conference call with reporters. He isn’t saying yet exactly when Poulson will be available for sale.
“We are targeting mission critical applications,” McInerney said.
The chip has eight computing cores, or brains, and can issue 12 instructions at a time, compared to four cores and 6 instructions for the previous generation. Poulson has 54 megabytes of cache memory directly on the chip for storing frequently accessed data. It can move data along its internal freeways 33 percent faster than before.
The chip is built with Intel’s 32-nanometer manufacturing process, compared to 65-nm for Tukwila. That means it’s easier to miniaturize circuits on the chip and pack more of them closer together. All of that results in a chip that is more power efficient, smaller, and cheaper to make than the previous generation. While Poulson has 3.1 billion transistors, Tukwila had 2 billion transistors. Power efficiency is anywhere from 30 to 70 percent better for Poulson.
Intel remains committed to Itanium, as it is working on another design code-named Kittson, which will ship after Poulson. These chips are used in huge corporate servers and supercomputers. They are meant to run enterprise-wide resource planning applications, giant corporate databases, and data warehouses.
One of Poulson’s features is a much higher reliability, or RAS technology. It also has error correction, which ensures that there are no errors in crunching vast amounts of data.
Intel competes with rivals such as Advanced Micro Devices, Oracle and IBM in the RISC market. Intel notes that market researcher IDC reported Itanium-based computer revenues were bigger in 2009 than AMD’s Opteron server chip business.