quaint little rutted bucket

Saturday, August 07, 2004

* warning: technical blabber ahead. if you find the below almost alien-sounding, I'm sorry, the inner geek in me just can't help it. :)

Geeky Op-Ed:
"Bumps in the Road: Intel's Recent Delays in Perspective"


Joel Hruska of Sudhian.com fame outlines the recent problems over at Santa Clara in that famed semicon corporation, Intel. He also cites the problems not only plaguing Intel, but more importantly, the entire industry with regards to the 0.09 micron process. Remember that even Big Blue, who has been developing their 90nm process with Sunnyvale's Advanced Micro Devices (AMD), also have had problems scaling the PowerPC G5 processor on the latest dual-CPU PowerMacs, requiring Apple to design perhaps the first fully commercial and mass-produced workstation equipped with water cooling on a processor that hardly runs near 3.0GHz. Talk about stretching things to their absolute limits.

The problem is not entirely with the process itself. Intel's Prescott (a 90nm design) Pentium 4, which dissipates much higher heat levels than AMD's once-infamous 0.18 micron process CPUs, while properly scaling with the right amount of cooling, still suffers more-than-acceptable levels of power leakage and heat dissipation, despite Intel's claims to have developed necessary technology to reduce them for at least another process shrink or two. Intel's roadmap page, available as an easy search on Google, is essentially frozen on a single platform until first quarter of next year, largely due to the problems of scaling. AMD's not in a hurry either to increase the clock-speeds on its flagship Athlon64 FX and mainstream vanilla Athlon 64s. Of particular note was Joel Hruska's observation about Intel [finally] downplaying clockspeed as a raw measure of potential system performance; at least, Intel's latest product launches of Land Grid Array 775 (Socket 775), the 925 and 915 chipsets focused more on the newer technologies (HighDef Audio, DDR2, PCI Express) the platform brings into market rather than the overall performance of the platform itself. Proof in itself of the changing landscape in the semiconductor industry.

It is in this light then that we are starting to see designs incorporating dual-cores than clock-speed scaling--AMD's K9 is
physically a fused K8 design, while Intel's 2H plans (at time of writing) calls for a dual-core NetBurst design--essentially a dual-cored Pentium 4. This brings in SMP (symmetric multiprocessing) as a crucial feature software slated for 2005 release should have.

Despite all of these problems plaguing the industry, I am confident that things will go forward. In one way or another. It will be interesting to see how things will turn out a year from now.


Post a Comment

<< Home