Earlier today, I summarized the part of my discussion with Intel's enterprise marketing director Shannon Poulin that had to do with the market requirements for server chips that generate a minimal amount of heat with as little compromise to processing power as possible. Prowess and heat have always at odds with one another when producing chips. Higher performing chips run hotter and, outside of Moore's Law, which indirectly implies that the heat produced by a given amount of processing horsepower will decrease over time, the main way way to cut back on heat production for a server chip is to ratchet down the number of watts it takes for the chip to run. In a mobile scenario, you can cut back on heat production through techniques like sleep states when the user isn't as active. But sleep states and server applications don't exactly mix.
That discussion took place in the context of a longer briefing about the roadmap that Intel's Digital Enterprise Group is announcing: a roadmap that includes, as News.com's Michael Kanellos points out, an incredibly aggressive push towards predominantly dual core offerings. In the bigger picture, the announcements by Intel aren't just an acknowledgement of the strategies long pursued by arch-rival AMD; they're proof that Intel can turn its oil tankers (fabrication facilities) around a bit more like motorboats while going on the counterattack.
In the AMD acknowledgement department are two items that go hand-in-hand with each other. The first is a continued de-emphasis on chip speed (eg: 2.8 Ghz) as a part of the branding and marketing of its microprocessors. Moving forward, Poulin told me, processors will be numbered (eg: there will be a 7000 series of XEONs) but nothing about the chip's clock speed can be implied from the number. This, of course, is remarkably similar to a strategy long employed by AMD who has forever argued that clock speed is somewhat meaningless when trying to compare offerings. To further emphasize its point, AMD would give model numbers to its chips that were not a reflection of their true clock speeds, but rather an implication as to the clock speeds of the Intel chips whose performance they matched (even though AMD's clock speeds were technically slower). Clockspeeds will continue to be relevant in terms of determining the differences between chips within a single vendor's chip lineup, but they appear to finally be falling by the wayside as a marketing bullet point for differentiation. That said, you and I both know that the clockspeed will continue to be discussed and often misused by marketers.
The second nod to AMD comes in the form of packaging and branding. You can't buy just the processor from AMD. For example, whereas the memory controller is external to Intel's processors, it's integrated into AMD's. Intel argues that the more modular design lead to more flexibility and easier implementation of supporting technologies (supporting to the processor) as they come along. AMD argues that it's OK to lock into a less flexible design for a longer period of time if the integration results in certain long term performance benefits from the design. So, AMD is more of a package deal and Intel is a bit more a la cartesque. But now, even if only in branding, Intel appears to be leaning more in the direction of the package deal. Moving forward for example, Poulin told me that if the processor is a 7000 series processor, then the chipset that goes with it will be branded as 7100. And, taking the packaging one step further, the networking controller will be branded as a 7200. Said Poulin:
Every new platform gets a name unto itself. We're moving away from things like gigahertz and cache size and towards a situation where we're focusing more on elements of platform: Virtualization, management, ability to provision. We are much more focused on features than absolute gigahertz and cache size.
Where have we seen this sort of bundling before? Although it doesn't use numbers, it seems reminiscent of Intel's Centrino brand. Centrino represents a certain combination of processor, chipset, and networking (one that, if you've been reading my blogs for a while, you'd know that I'm not particularly fond of). Even so, for servers, it's clearly a step in the perception direction that AMD took a long time ago. It will be interesting to see where Intel goes with this plan.
In the counterattack department, Poulin didn't mention AMD by name. But he did make it clear that Intel is more than ready for a 32/64 hybrid world. Although he said "85" isn't some magic number, Poulin said that 85 percent of the server chips that Intel is shipping now are 64-bit capable and that by the time the end of 2006 rolls around, 85 percent of the chips it's shipping will be dual core as well. And a significant number of the servers based on these processors will be two-way servers (dual processors). The net-net implication of dual core two way systems is that with each core designed to simultaneously process two execution threads (due to Intel's Hyperthreading technology) and with two processors each with two cores (for a total of four cores), a 2-way dual core system is theoretically capable of simultaneously handling eight threads.
But is it enough to slow down AMD's Opteron which, according to a recent Mercury Research survey ,has surged from 7.4 percent marketshare in 2005Q1 to 11.2 percent share in Q2 (numbers that Intel's Poulin disputes on the basis of the methodology behind them)? That remains to be seen. Even though Poulin says that most existing applications will derive some benefit out of the brute force of a 2x2x2 system (threads x cores x processors), he admits that developers will need to revisit their applications before they can get the most out of all eight threads -- half of which can be attributed to the Hyperthreading technology. While developers contemplate that option, AMD isn't exactly waiting. Its dual core processors are out as well, some of which are married to the 1 GHz version of AMD's Hypertransport bus. For those not familiar with Hyperthreading or Hypertransport, the two are not to be confused with each other. They do two entirely different things. But ultimately, both are designed to improve the same bottom line: system performance.
Currently, Intel has 17 dual or multi-core projects in development. While only three of them are in production right now, Poulin said that Intel's plan "is to ramp dual core aggressively." But, whether or not that dual core push will be enough for Goliath to keep David at bay -- particularly when David is equally committed to a dual core architecture -- remains to be seen. One thing is for sure. In terms of AMD approaches to adopt, there isn't much left for Intel to consider. Unless of course it wants to reconsider what "HT" should stand for.