EDN Senior Technical Editor Brian Dipert exposes, analyzes and
opines on diverse topics in technology.
Apr 27 2005 12:15PM | Permalink | Email this | Comments (0) |
Blog This! using: Blogger.com | LiveJournal |
Digg This | Slashdot This | add to Del.icio.us
Whenever I need some benchmarking data and don't have the hardware (my AMD system's reportedly enroute) or time to get it myself, I feel pretty confident that I'll find the unbiased and thorough examinations I crave at AnandTech (and ExtremeTech). Anand Lal Shimpi and his team came through once again with their recently-published analysis of latest-generation AMD and Intel dual-core products, along with benchmarking of:
When comparing performance results, keep in mind that each core in the AMD dual-core CPUs runs at a significant clock speed deficit versus the Intel alternatives. This deficit results in lower power consumption that enables AMD's dual-core chips to drop right into most single-core-designed motherboards. And the AMD chips still generally hang right in there with their Intel competitors. Like I said yesterday, AMD's CPUs' strengths are indisputable.
AMD's desire to assure socket backwards-compatibility is understandable when you remember that the company doesn't build its own motherboards and is dependent on third-party partners for chipset support. AMD does design its own chipsets on occasion, but only to seed the market, not for high volume production; keep in mind, too, that AMD's CPUs have integrated much of the functionality that'd be in an Intel-based system design's 'northbridge' core logic chip. The backwards-compatible strategy does force the two cores in AMD's latest CPUs to contend for access to external DRAM over a shared single memory bus, although they talk to each other over a dedicated HyperTransport link.
I suspect that at AMD's 65 nm process node, the higher transistor budget, coupled with a higher pincount next-generation package, will deliver single-die, dual independent CPU-core-plus-DRAM-controller combo chips, each communicating with its own dedicated memory array and with each other over HyperTransport. 'HyperTransport' is a key word; even as it stands, AMD's approach is still technically superior to Intel's in that in the Intel case, the front-side bus is employed not only for access to external memory and other system peripherals, but also for core-to-core communication. Beware the Tragedy of the Commons.
Intel's reportedly got a cost-slimming trick up its sleeve, though. The company's first-generation dual-core chips are in effect two single-core CPUs stuck together on one die. If one core fails testing, the die gets physically chopped in two, with the remaining good core dropped into a package that's compatible with existing single-core motherboard designs. However, the resultant chip, presumably in both HyperThreading-inclusive and –exclusive variants, will exhibit an on-chip cache deficit versus existing P4s. Celeron, anyone?