Intel has estimated that Itanium 2 is some 150 to 200 percent faster than the first Itanium hardware. Later this week Intel is due to release new benchmarks for the Itanium 2 chip. In the meantime, we have produced this report using results from a range of systems, and have compiled results from an Itanium 2 by assuming the 200 percent figure is correct.

To produce this report we used results from the Standard Performance Evaluation Corporation (Spec), which publishes a comprehensive set of benchmark results on its Web site. Since the 32bit estimates for Itanium 2 are unimpressive, we selected 64bit floating-point and integer results for systems having one CPU clocked at the highest speed available, and aggressively optimised to give the best performance.

The floating-point benchmarks show just how competitive Intel's 32bit 2.2GHz Xeon processor is when compared with Itanium. Extrapolating Intel's own application results with C++ and Fortran datasets gives Itanium 2 a 10 percent advantage, compared with IBM's 1.3GHz Power4 processor. One application for floating-point calculations is for encryption algorithms, which would be fundamental to e-commerce servers.

IBM's Power4 processor, Intel's 2.2GHz Xeon and AMD's 1.733GHz Athlon XP 2100+ chips are all set to beat the Itanium 2 on integer performance, the Power4 processor expected to be 17 percent faster. An interesting point is that HP's Unix compiler trails Windows-based compilers in the floating point results, a situation reversed with integer performance.

Some cynics say that raw integer and floating-point performance is not always a useful measure for servers and that Intel's new architecture was primarily designed for users needing to address more than 4GB in online transaction processing (OLTP) scenarios. They can support this argument by pointing to the benchmark results from Sun's UltraSparc systems, which lead the enterprise database market but score poorly in benchmarks. However, these results are a useful indicator of workstation performance.

While the speed increase of around 200 percent claimed by Intel for the Itanium 2 is impressive, it should be remembered that the original Itanium was marketed as a test and development platform. Much of the claimed 200 percent increase comes from moving the L3 cache onto the same die as the processor, from adding new execution units to the chip, and from speeding bus bandwidth from 2.1GB/s - less than that of the Pentium 4 - to 6.4GB/s, which is comparable to the buses used by competitors.

In fact, doubling the width of the system bus to 128bits and increasing the bus speed by a factor of 1.5, triples the system bus bandwidth. Similarly putting the L3 cache on-die increases the cache bandwidth by a factor of 2.5.

The final point about Itanium 2's architecture is that it requires software to be created with an optimised compiler in order to achieve optimal performance, and there will be further performance gains to be had by improving the way compilers set up code for Itanium 2.

Have your say: contact IT Week