Itanium: the beat goes on

Intel Developer Forum Steve Smith, who heads up the Itanium-Merced programme for Intel, was in expansive mood about the future of the platform when we talked to him last night. But he's not yet ready to give us a prototype Merced so we can use it as a paperweight, even the dummy one we photographed two IDF's back. Nor was he prepared to talk about the exact nature of the metal alloy which acts as a dispersal heatsink for the device, despite the close interest several metallurgists had in this question when we last raised it. He did, however, say that the sink, which spreads the heat straight from the die, uses a technology called vapour phase. That might give a clue to someone but boffins at The Reg are pretty thin on the ground. He confirmed that the prototype systems running at the forum were clocking at around 500-600MHz, and said that there was, theoretically no limit to the number of CPUs that can be clustered together. SGI, for example, will support up to 512 processors using local memory and broadband interconnect, and we know from our own conversations with HP that it has similar plans for its platform. Typically, however, the first Itania will come in four, eight and 16-way configurations, he suggested. Smith said that the Itanium platform had a different set of requirements than those which IA-32 platforms offer. He said that Intel was aiming to leverage the chip as a suitable platform for Internet servers, given that there is a 20 per cent anticipated growth over the next four to five years. He said that for these type of machines, Intel was ensuring that the whole industry support structure was there. He said: "We don't expect anyone to be running a 32-bit operating system on Itanium," although he added that Quake for Linux had run successfully on the platform. He said: "As of today we have disclosed the Itanium system architecture." There will be a more aggressive implementation of Streaming SIMD on Itanium which will deliver something like ten times the security of current Pentium IIIs, and "several times" more than Willamette/Foster. He said that, Itanium's 64-bitness will allow it to employ similar but more advanced techniques for moving code around, in a similar fashion to that which engineers have done with Willamette. CERN, he said, has already ported a program called Cactus to the IA-64 platform, a simulation which shows black holes colliding. The US National Security Agency (NSA), has also shown "a very strong interest" in the IA-64 technology. He was also prepared to dilate on Intel's relationship with Sun and the 64-bit Solaris operating system. "What we haven't seen from Sun is its willingness to bring the ports on," he said. "It's not just the technical reality of porting the OS, but also the business reality." He compared the Sun approach to that of HP. "HP has made a commitment to make it happen," he said. "We don’t see it as wise to pour our energies into something when it's not reciprocated." (As a sidenote to this, Paul Otellini this morning confirmed that the Sun-Solaris relationship would end with the Itanium processor). Most of the other details Smith outlined, we already know. For example, the size of the on-die cache, a cache which has a direct connect to the processor. Prices of the first Itania, said Smith, would be comparable to those of the Pentium III Xeon. Smith, who was responsible for the creation of the 386SX processor, said he felt a sense of satisfaction that he had brought this project to life. ® Intel Developer Forum: Spring 2000 Full coverage g