Itanium allies to pool development efforts

TalkBack 22 of 23:

Next »

« Previous

• Thread View
• Flat View

Itanium had failures on several fronts.

1. Marketing failures: this chip needed to be sold for considerably less, even if at a substantial loss, to gain market share and therefore developer support. Itanium-based boxes have always been expensive and relatively slow, especially for the premium charged. I've ported workstation software (image processing - very parallel) to both Itanium and Opteron, and found that the Itania performed well, but just weren't worth the longer development time for tuning and the 4x price differential.

2. Design failure: The VLIW approach is valid, and isn't going away; it will probably be in future Intel CPUs under the hood, much as RISC lies under the P3 and P4. The technical flaw in Itanium was in eliminating out-of-order execution. The theory was the chip could clock faster and have a shorter pipeline without it. The reality is - Itanium clocks slower than Opteron, which HAS Out-of-order processing. Keeping the 3 dispatch slots full on an Itanium is difficult for compiler writers AND application programmers; on an out-of-order processor, the machine will just reorder the instructions on the fly to get better utilization of it's functional units. This same logic, applied to the Itanium, would give significant performance improvements and reduce the workload on the compilers. The arguments about "faster clocking" and all the rest are obviously bull, or we'd see 4 GHZ Itaniums, which we don't. For general-purpose software, the decision about which instructions can fire in parallel is often better made by the chip itself at runtime, based upon current load, than by trying to predict it at compile time. The latter is like trying to decide in advance which lights you're going to stop at when driving from NY to Florida; at best, you're going to stop at a few greens, and at worst you'll run reds and have collisions.

All this being said, Intel could still revive this chip. But fleeing to the high-end is just delaying a painful death. If the plan is long-term viability, sell it for the same or less than Xeon, using a low-cost supporting chipset. Then get MS to support XP on Itanium (as they did until recently, when the pulled support for all but the "server" OS). Most 32-bit software runs fine on a decently-clocked Itanium, and for apps compiled specifically for it, it can really shine. But not at $10K/box.

Posted by: tyezek   Posted on: 08/31/05 You are currently: a Guest | Members login | Terms of Use

Alert moderator to an offensive message

Subscribe to this discussion via Email or RSS

SponsoredWhite Papers, Webcasts, and Downloads

• Three Steps You Need to Know to Stop Data Loss Varonis Sensitive data exposed to misuse or loss... it is the stuff of nightmares ... Download Now
• Effective Performance Testing in Java Environments Quest Software When applications perform poorly so does your business. With performance ... Download Now
• Five Steps to Determine When to Virtualize YourServers VMware Server virtualization isn't just for big companies. Entry-level ... Download Now