Analysis: Can Merom return Intel to the performance driver's seat?

San Francisco (CA) - Yesterday at the Intel Developers' Forum, the company released the first wave of details about its next-generation Intel Core Microarchitecture (ICM). In so doing, Intel has set high expectations for stunning performance gains. Yet so far, the ratio of words to facts has been heavy, with a smattering of statistics but a lot of verbiage comprising Intel's key value proposition for ICM. Will it really flatten everything in its way, as Intel predicts?

Visitors at IDF got a first taste of what Intel is capable of, once you wake it up. Thus far, its demos have left no doubt that Merom and its derivatives - Conroe on the desktop and Woodcrest on the server side - are monster systems that promise to beat every other processor on the market, in terms of both performance and efficiency.

Yesterday, we reported on a showdown between a 2.4 GHz Opteron-based Sun workstation and a Woodcrest-based HP ProLiant server. The event marked the first time that Intel ever displayed an AMD-based system on stage at IDF, so not unexpectedly, the Opteron was no match in performance and power consumption. The showdown was vaguely reminiscent of Steve Jobs' historic MacWorld demos of Power processors, that put Intel's Pentiums to shame.

There also was a demo benchmark of Call of Duty II, in which an overclocked 4.1 GHz Pentium Extreme Edition computer tried to keep up with Conroe - and trailed its successor with 90 fps to 110 fps. Finally, a Merom CPU was compared to Core Duo within a Napa64 platform: The 2.16 GHz Core Duo achieved 106.6 fps in a Quake 4 benchmark; with a Merom processor, this notebook came in at 134 fps.

And so it was that the theme of Day 1 of IDF was performance, which had not historically been Intel's forté. This is somewhat surprising, as at last year's Fall IDF, Intel gave couched warnings to attendees that future processors my see smaller performance gains, as a result of an increased focus on power efficiency. Perhaps Intel has learned to play a game mastered by American politicians: lowering expectations to heighten yields. While the company "stays the course," to borrow another political phrase, by explaining its progress in the new context of "efficiency per instruction" (EPI), the truth is, everyone involved in microprocessing, from users to engineers, should be mindful of the relationship between power and performance. Power reduction by itself is no miracle. What's truly remarkable is the performance that apparently will be available at these power levels.

Yesterday, we also mentioned that Intel expects Merom to gain 20% more performance compared to Core Duo; Conroe promises performance increases of 40% over Pentium D 950 while decreasing power consumption by the same amount; and Woodcrest is forecasted to achieve an 80% jump in speed coupled with a 35% drop in power over today's dual-core Xeon DP 2.8 GHz. "Merom is the best processor we have ever built," summed up Pat Gelsinger, who runs Intel's digital enterprise group. Gelsinger added that Merom will be the most power-efficient processor on the market, and will give Intel a "very strong position" - strong enough, in fact, that Intel no longer feels the need to develop an integrated memory controller for Merom at this time.

Once you wipe away some of the superlatives and hyperboles, the numbers produced by Intel have certainly impressed analysts and journalists here. But new questions have arisen: Why is Intel just now able to claim such performance and power jumps? What changed, besides the company's attitude, between last fall and today?

According to the company, there are five key components that are responsible for the progress made with Merom:

1. The chip can process four instructions per clock cycle - one more than Core Duo.
2. The shared L2 cache comes with a "smart cache" feature that allows the processor to more efficiently access data stored there.
3. Data pre-fetch algorithms have been improved.
4. Further developments in power gating enables Intel to shut down more components of the chip for longer periods of time, than with today's designs.
5. Perhaps most importantly, Merom is able to combine two high-level instructions and process them as a single unit, and multiple Streaming SIMD Extensions (SSE) instructions that perform uniform operations on multiple data streams simultaneously, can be processed in just one clock cycle.

More speed. Less power consumption. Intel promises a Formula 1 racecar than runs with the efficiency of a Mini Cooper. Is this realistic, or is there a catch?

The simple answer: We don't know.

And here is why. First, when we look at all the announcements thus far, it becomes clear that there is little besides promises to back up the company's fresh wave of claims. Intel demonstrated that its new platform is running on notebooks, desktops, and servers alike. But we have yet to believe the company when it states that ICM architecture will deliver what Intel promises. Second, all product demonstrations compared a future product against processors that are available today. Especially in the case of the Woodcrest / Opteron showdown, we know that AMD is unlikely to stand still, and certainly will continue to improve the performance of its processor line. Intel's aiming at a moving target; in fairness, so now is AMD. But once the targets finally line up with one another, their real differences in terms of performance and power consumption therefore may not be as dramatic as Intel portrayed here at IDF.

Speculation will be the principal ingredient of discussions around Woodcrest, Conroe, and Merom (which will be introduced in exactly this order) until we actually can get our hands on a production-ready processor. Will it be faster and more power efficient than Athlon and Opteron when introduced? Perhaps. But here's a thought: Maybe this is the wrong question for us to be asking anyway. The real question is, what does that mean for you, the user? From the limited perspective we've been given this week, it certainly means that Intel Core Microarchitecture catapults Intel into a much more promising future than many had been envisioning. As a direct result, we are likely to see a higher level of innovation in micro processors and platforms, driven by a more competitive environment, at least in the near term.

We continue to expect 2006 to become the most interesting year in the microprocessor industry thus far. If Intel is able to follow through with its promises made at IDF, we have no doubt that this year will reinvigorate enthusiasm for the microprocessor, as well as bring new opportunities for the industry behind it.