Combjelly, I was there, and I listened to Wen-Mei Hwu's presentation. It was very well done, and it exposed a lot of the faults in the first generation IA-64 architecture (of course, the fact that there are faults have not been a mystery up to this point).
What the article fails to really get into is that Wen-Mei had a lot of optimism for future IA-64 compilers, and in new generational enhancements that would extract a lot more performance from the architecture.
Many have shared the opinion that IA-64 compilers have already matured, and that there wouldn't be a lot more performance to extract from an Itanium processor. Wen-Mei exposes that myth as well. If IA-64 really can fix the majority of micro-architectural performance issues in just 2 or 3 generations, then it can easily become the world's most powerful microprocessor in the next few years, and that would certainly be worth the long development and high costs to get to this point.
Sure, it's a lot of "ifs", but McKinley is already promising 70% more performance for only a 25% increase in clock frequency, and it's only on .18u. Compared to other competitive processors, like the 1GHz Mako, which is on .13u SOI, McKinley has a lot of headroom. The .13u Madison core might take frequency up to 1.5GHz or greater, and with the extra cache and micro-architectural improvements, it will already soon be approaching and surpassing the rest of the high end lineup.
wanna_bmw |
