Intel: x86 won't encroach on Itanium

TalkBack 3 of 36:

Next »

« Previous

• Thread View
• Flat View

Intel's problem is

that it is or appears to be irresolute at this stage. They clearly contemplated a successor to their's x86 (IA32) ISA with the IA64. However it seems that Intel is still siting at the crossroads and it is hesitating to make firm commitments one way or another.

IBM, SUN, HP (and previously DEC) new well that a good high-end processor does not make a good high-end system. Itanium-2 is decent but it is lacking all the necessary infrastructure h/w to allow it to become the key part of a coherent architecture. Only SGI managed to pull this off: they designed the memory, interconnect and I/O subsystems that can leverage Itanium. And they have been pretty successful at that, even considering that they were relying on RHEL IA64 Linux to run the system.

IA64 shortcommings include:
1) EPIC: radical departure from well understood / well accepted CISC or RISC technology. EPICs require very sophisticated compilers to leverage the inherent ILP. Applications written for the 32-bit x86 need considerable effort to be ported to IA64. Operating Systems need even more effort.

2) Itanium has no specification for scalable systems architecture. Although SGI managed to glue 1000s of Itaniums together, it is using its own cache coherence protocols and inteconnect designs. If IA64 gets an 'upgrade' specification, SGIs design may have to be scrapped.

3) Itanium is a mem bandwidth hog. It demands very wide data buses to keep feeding it otherwise it will be just stalling most of the time.

4) Itanium is not multithread friendly. Processors of this complexity need to execute multiple thread contexts (SMT). Imagine a server receing 2,000-10,000 interrupts / sec (not uncommon in servers) which means that the processor will be just saving restoring context to service interrupts.

5) Itanium consumes too much power and generates too much heat.



In short, Itanium-2 is a decent processor but it is an intermediate step not the final refined product. I guess Intel didn't realize that to succeed needs a long term commitment to R/D. IBM, Cray and SUN know this too well.

However since the x86 ISA is just so hard to improve upon, Intel needs to sit down and think what they should plan for the next 10 years.

What they need is an intermediate 'sematic' stage between IA32 and IA64 and a plan to make Itanium more SMT and more scalable.

-m

Posted by: michael-t   Posted on: 03/08/05 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
• Finally, an easier way for Small and Mid-Sized Companies to Run Their Business Applications: IBM Smart Business IBM From the PC to the Internet to every piece of hardware and software in ... Download Now
• Five Steps to Determine When to Virtualize YourServers VMware Server virtualization isn't just for big companies. Entry-level ... Download Now