Comment by jws | Hacker News Reader

Comment by jws | Hacker News Reader

Parent Full thread

Itanium was a bet that seemed the way forward at the time. Making an x86 instruction set processor that did on the fly analysis and conversion to a superscalar computation core sounded really hard, so VLIW was the safe "we'll do the superscalar decomposition at compilation time and run that". Turned out to be a bad bet because a miracle¹ occurred on the other side.

I think of it as a peer to the 2005 proposition that electric cars didn't need to be golf carts and the technology was coming into place to make high performance, real cars that were purely electric. Could have been true. As they say "very dangerous, you go first." That one paid off.

I wonder if there is a list of large dollar, novel, engineering expeditions and how they turned out.

␄

¹ Where "miracle" means some really smart people worked really hard at it for a long time.

Itanium was a gigabuck and drove everybody out of the high-end processor market except IBM leaving Xeon for Intel to monopolize for a decade+. I'd qualify that as a business success even if Itanium was a technical failure.

And, it may be paying off after all. Graphics cards are effectively giant VLIW machines. And while Intel's new cards get slaughtered on DirectX 11 or older games, they are quite competitive when used on the DirectX 12+ stuff which basically wants a giant multi-core parallel processor.

I’m not sure the timeline really works for this argument. Pentium was the first x86 superscalar and that was 1993 so it predates Itanium.

Rather Intel wanted to leave the baggage of x86 behind, segment the market for higher profits and hoped to get better performance than superscalar x86. But they never achieved the latter and most people just wanted 64 bit and software compatibility.

> Making an x86 instruction set processor that did on the fly analysis and conversion to a superscalar computation core sounded really hard, so VLIW was the safe "we'll do the superscalar decomposition at compilation time and run that".

The first superscalar x86 processor was the Pentium, which came out in checks notes 1993 (Wikipedia claims design work started in 1989, hit working simulation in 1990 and taped out in 1992). Intel didn't start work on Itanium until 1994, after Pentiums were already shipping to customers, and Itanium itself wouldn't ship until 2001.

Reply on news.ycombinator.com