Intel: From The Bottom Up

One of the most immediately identifiable reasons that dual-core will take some time to catch on at the mainstream level is that it actually sacrifices clock speed in a bid to improve efficiency. If you're familiar with the term speed-binning, it's a little easier to see why the frequency knob has to be relaxed somewhat. A single Pentium 4 running at 3.8 GHz consists of 125 million transistors. When Intel unveiled the 600-series featuring 2MB of cache, transistor count jumped to 170 million. Whether because of added complexity or marketing insight, the 2MB part topped out at 3.6 GHz—down a notch from single-core Prescott's peak. The Pentium D, Intel's new dual-core desktop series, hosts a whopping 230 million transistors manufactured on the same 90nm process, yielding one huge 206 square millimeter die. It is hardly a surprise, then, that the fastest model is further detuned to 3.2 GHz. Both Intel and AMD have to use slower transistors in order to keep the power levels manageable. As the number of transistors increases, that issue becomes more of a problem, and the increased complexity leads to slower operating speed.

Meet the New Extreme Intel's latest desktop flagship is the dual-core Pentium Extreme Edition. Note that, as with the Pentium D, the "4" (as in Pentium 4) nomenclature has been dropped for dual-core designs.

"When you push transistors to the limit of their process technology, they draw power at a non-linear rate," explains Jeff Austin, multi-core product marketing manager for Intel. "Power is basically related to voltage squared times frequency, so as the frequency goes up it draws more power. Also, to get the frequency up, you have to feed it more voltage. So you have a non-linear growth in power consumption, hence you see dissipation. But with dual-core, we're able to effectively maintain the frequency at a slightly lesser range such that the voltage doesn't need to be cranked up. We're not running the transistors at the upper end of their capability, and that gives us the ability to maintain lower power. For example, a 3.73 Pentium 4 Extreme Edition is around 120W. We have the ability to bring in two execution cores running at 3.2 GHz and fit them within a max power envelope of 130W."

Note that since 130W is not an increase over single-core Extreme chips, there is no need for novel heatsink solutions. The coolers already on the market work just fine.

What the Pentium D lacks in frequency it makes up for in efficiency. The previous discussion of Hyper-Threading technology illustrates Intel's first foray into working smarter rather than harder, and dual-core takes that approach even further. In short, the Pentium D is an evolutionary step beyond Hyper-Threading because it leverages two separate sets of execution resources—one per thread, just like you'd encounter in a dual-processor server or workstation system. Revisit the processor usage history panel in Windows XP and you'll again see a pair of running timelines. This time, both represent physical cores, which are much better equipped to power through multi-threaded apps than a single core equipped with Hyper-Threading.

P4 x 2 = Pentium D A great advance that still gets criticized as being "two Prescott cores glued together," the Pentium D has twice the power of P4 but without an increase in bus bandwidth.

The two Pentium D cores are architecturally identical to the first Prescott core with 1MB of L2 cache. EM64T, Intel's 64-bit instruction set, is supported, as is the virus-fighting Execute Disable bit and advanced power management functionality. At least from a value-added feature perspective, the Pentium D is comparable to any Pentium 4. That's good from an education standpoint, since you already understand how the design is capable of performing. But it also means that Prescott's oft-lamented thermal qualities and ultra-long execution pipeline will remain with us on the desktop through at least the end of this year.

Fortunately, Intel's Pentium D employs the same fundamental infrastructure components as Pentium 4 processors. Designed to fit into the same LGA775 socket interface, it's nearly impossible to distinguish the difference from a single-core Pentium 4 and Pentium D by visual inspection. The Pentium D is also optimized for Intel's existing 800 MHz front-side bus setting. If you really want to get technical, the Pentium D actually suffers a minor setback in that the 800 MHz bus is responsible for handling memory traffic and communications between the two processing cores. As a result, effective throughput drops drastically in memory bandwidth tests. Intel recognizes this as a weakness and is already planning to add a second front side bus to its Blackford (Xeon DP) chipset architecture due out in 2006.

Of course, at the end of the day, your customers want real-world performance indicators and not hypothetical suppositions. Intel's dual-core Pentium D may not represent any major architectural advances, but it does effectively demonstrate the benefits of combining two independent cores on a single die. Evolutionary more than anything, the stage is now set for future dual-core projects, of which Intel claims to have about 15 in the works, including a highly anticipated mobile offering (Yonah) set to employ the 65nm manufacturing node.

The X Factor

When I first read the Pentium D's specification sheet, the absence of Hyper-Threading surprised me. Although dual-core delivers the same end result with more efficiency, Intel put in the research and development to bring Hyper-Threading to fruition. The technology does help improve overall utilization, and although that's supposed to be dual-core's forté, Hyper-Threading is a logical (pardon the pun) value-add. Simply discarding it seemed awfully wasteful.

As it turns out, Hyper-Threading does make its return in the Pentium Extreme Edition. It may come as a disappointment that Hyper-Threading is the only feature distinguishing Pentium D processors from the Extreme Edition family, save for an unlocked clock multiplier so that overclockers can finally take another stab at viewing Intel as an enthusiast's product. There's no 1,066 MHz front side bus or larger L2 memory cache this time around. However, high-end customers will be fascinated to see a single chip addressing four threads through two physical cores and two logical processors (yes, you can see four CPU utilization panes in Task Manager), even if most of the applications I've tested seem unable to coax extra speed from the additional threads.

Intel's current flagship Extreme Edition is the model 840, running at 3.2 GHz. Intel representatives won't comment on its exact pricing quite yet, but we've been told to expect something in the neighborhood of existing Extreme Edition processors, or roughly $1,000.

Priced to Move

It's difficult to fall in love with the dual-core desktop processors available today. Fortunately, Intel seems to have an inclination that the technology is a difficult sell at the mainstream level given today's predominantly single-threaded software landscape and has priced the new Pentium D accordingly. Granted, Intel didn't help its cause by preaching the gospel of clock speed over the past several years—a notion that it must now debunk while stepping back from the 4 GHz barrier.

The top-end Pentium D 840, operating at 3.2 GHz, lists at $528, followed by the 3.0 GHz model 830 for $314 and a 2.8 GHz Pentium D 820 for $240. Those are fairly competitive prices for such incredibly complex processors. In comparison, even AMD's slowest upcoming desktop offering costs more then the Pentium D 840. When you look at the price of a single-core 3.0 GHz Pentium 4 and compare it to the Pentium D 830, for example, you're actually getting that second core at a discount. Not bad for an entry-level video editing machine or lightweight media server.

Platform Considerations

Customers looking to drop existing LGA775 processors in favor of a dual-core chip may be disappointed to learn that only a handful of Intel's current platforms recognize Pentium D processors. An upgrade requires a brand new motherboard centering on the 955X, 945P, or 945G MCH (memory controller hub) with ICH7 I/O controller. Although the latest chipset generation introduces a handful of attractive features, it may not be compelling enough to motivate many upgrades based on the merits of dual-core alone. This makes Intel dual-core an unlikely candidate in the upgrading segment.

An easier target audience is with those buying new systems since they're already shelling out for a new motherboard. The 955X and 945P/G Express chipsets support both single- and dual-core processors spanning from mainstream to high-end. Supporting motherboards will undoubtedly bear a premium at first but should fall off to today's 925/915 levels shortly thereafter, enabling many different configurations and price points. The 955X sits atop Intel's chipset hierarchy with the ECC memory support needed to get the platform into workstations. Performance desktops are recommended to the 945P chipset, identical in almost every way minus the ECC option. Both accommodate dual-channel memory modules of different sizes through Intel's Flex Memory technology. The 945G platform is intended for mainstream consumption and includes integrated graphics.

Accelerating I/O remains one of Intel's principal foci with these new platforms, spearheaded by PCI Express. DDR2 memory support also extends to 667MHz, which, in dual-channel mode, theoretically yields 10.7 GBps of bandwidth. Unfortunately, the 800 MHz system bus essentially cripples that throughput, in particular when you're using high-latency modules. A transition to 1,066 MHz would certainly help even things out, and while the chipset supports that setting, today's dual-core processors do not.

Complementing the new slew of memory controller hubs is Intel's ICH7. It's very much similar to the preceding ICH6 with the exception of a couple extra PCI Express x1 links, SATA/300 (3 Gbps) support with RAID 0, 1, 10, and 5, and a lower operating voltage. There's one other feature of particular interest to VARs with distanced clients called Active Management Technology, which enables remote diagnostics, remote BIOS recovery, OS lock-up alert, and serial over LAN. The whole concept seems tailored for enterprise customers, but with the proper components in place, Active Management might provide service providers the tools to quickly troubleshoot and diagnose software issues with remarkable turnaround. How's that for offering a competitive edge? And, naturally, the ability to carry out remote service can be another upsell facet you plug into your maintenance contracts.

Why did Intel opt to start its dual-core lineup near the bottom of the price scale? Theories abound, but Intel's Austin offers this:

"From a timing perspective, a lot of what went into the Xeon rationale was our end customer platform cycle times. In the server space, cycle times tend to be quite a bit slower than cycle times on the desktop. Desktop is very much a yearly cadence. The server space is much more drawn out than that. The more natural refresh time for servers, both for OEM customers and our end customers, lines up more toward the early part of next year."

...more