Review: Intel's New Quad-Core CPUs
In our lab tests, Intel's 45nm Penryn CPU didn't blow away the previous generation, but the tech behind it should keep Intel ahead.CyberPower's Power Infinity Pro is the first system we've tested with Intel's 45nm, 3-GHz QX9650 Core 2 Extreme chip.
Photograph by Marc Simon
Let's Get Small
For now, the real news is that the 45nm manufacturing process Intel uses for Penryn should allow the company to keep churning out superfast desktop chips for the forseeable future.
If Intel were to have its own TV show, it would probably be entitled Honey, I Shrunk the Chips. The Penryn family of CPUs, set to launch on November 12, are built on a manufacturing process that shrinks the features of the chip down to a mere 45 nanometers (or about 1/18000 the width of a human hair). That's down from the 65nm process the company has used for its current Core line and the 90nm process it used on some Pentium 4 chips. The company has already demonstrated a 32nm process that it intends to begin using to produce chips in two years.
By shrinking the size of the transistors in its chips, Intel can produce more CPUs from the same amount of silicon, or build more-complex chips in the same amount of space. For example: A Celeron 300 made in 1995 using a 250nm process measured 131 square millimeters, yet contained a mere 7.5 million transistors; a current 65nm Core 2 Duo is a scant 11 square millimeters larger but contains 291 million transistors; and the new 45nm, quad-core Core 2 Extreme QX9650 that we tested for this article measures 214 square millimeters but contains a whopping 820 million transistors.
Hands-On
The first desktop Penryn chip, Intel's QX9650 Core 2 Extreme, runs at 3 GHz.
Photograph by Marc Simon
Like the existing quad-core crop of Core 2 Extremes, the new product is actually two dual-core CPUs paired on a single silicon package with a shared bus interface, running at 1333 MHz in this case. Each of the two dual-core CPUs carries a shared 6MB of secondary (L2) cache, up from the 4MB of each core of the previous QX6850 chip, for a total of 12MB. This larger secondary cache is partly responsible for the new chip's high transistor count.
Performance
Thanks to some optimizations and enhancements, such as faster divide operations and the larger L2 cache, Intel's own benchmark results from this spring's Developer Forum showed modest performance gains for Penryn over the last 65nm generation of chips running at the same clock speed, as well as moderate power savings.
To see just how much you stand to gain with a Penryn CPU running current hardware and software, we put together a test system built with Asus's Maximus Formula X38-based motherboard, 2GB of DDR2-800 memory, a pair of Seagate ST3320620AS 320GB hard drives in a striped array, and an EVGA GeForce 8800GTS graphics card. We tested both the QX9650 and the older QX6850, using PC World's application-based WorldBench 6 Beta 2.
The QX9650 bested its older sibling by a mere point, 127 to 126. In the majority of our test apps, the new chip was 2 to 5 percent faster, but slower times in Nero and especially WinZip dropped the overall number. Both WorldBench scores would place right in the middle of our top five gaming PC chart, though the GPU and hard-drive setup we used weren't cutting-edge, just close to it. As mentioned earlier, none of the applications in the WorldBench 6 Beta 2 suite are optimized to take advantage of the new SSE4 (Streaming SIMD [Single Instruction, Multiple Data] Extensions 4) instruction set, and only a few WorldBench apps can take advantage of more than two cores.
Intel's own benchmark results released on October 28, as well as the demos we witnessed at the fall Developer Forum, showed much larger performance gains with SSE4-optimized applications, such as an HD-optimized DivX encoder.
Infrastructure
The Asus Maximus Formula is one of several Penryn-ready X38-based boards that use DDR2 RAM despite chip-set support for DDR3.
Photograph by Marc Simon
For example, Gigabyte's Tomas Lee confirms that the company's P35, G33, and P31 motherboards will run the new CPUs after a simple BIOS update. And nVidia tells us that its nForce 600i series of motherboards, as well as the recently launched GeForce 7150 and 7100 Series chip sets, will also work with Penryn CPUs.
Intel is pushing DDR3 as a preferred memory companion for its 45nm CPUs, and has incorporated support for it (as well as DDR2) into its more recent chip sets. But given the high cost and small performance benefits of DDR3, many motherboard manufacturers are still designing their X38 motherboards around DDR2. Even the high-end Asus Maximus board we used for testing--equipped with a built-in water-block assembly so you can water-cool its chip set--uses DDR2. Some motherboards, such as the Asus P5KC, support both kinds of memory.
And some early Penryn PCs, such as the Power Infinity Pro from CyberPower that we tested for this story, will stick with motherboards using Intel's P35 chip set.
The Competition
AMD has lagged behind Intel in shrinking die sizes, putting itself at a distinct disadvantage in the economics of producing CPUs. Despite its tardiness, the company's Athlon CPUs have sold well because they outperformed Intel's products for a good three years starting in 2003. Athlon's dominance in speed tests ended abruptly in the summer of 2006 when Intel introduced its Core 2 line, though AMD's CPUs still compete nicely in terms of power consumption.
The financial advantages of shrinking die sizes are huge. Says IDC's Shane Rau: "It's classic Intel. Shrinking the die gives them more leverage over pricing, allowing them to outmaneuver the competition in the marketplace anytime they choose. In a price war it's a huge advantage." In other words, AMD has its hands full.
AMD isn't just sitting around idly while Intel kicks its, err...circuits. According to AMD's Simon Solotko, December should see shipments of the company's quad-core Phenom desktop processors based on its new Stars core, which includes the Hyper Transport 3 bus and support for DDR2-800 memory, plus Opteron-like features such as a shared L3 cache. The latest announcement from the company is a 65nm triple-core Phenom chip that should ship early next year. Generally the number three doesn't sit well in an industry based on powers of two, but with application support for more than two cores still a rarity, AMD believes that its triple-core chips can perform as well as quad-core processors do in many cases. And a triple-core option gives AMD a way to use quad-core chips with one defective or deactivated core to hit lower price points.
Unfortunately, since AMD hasn't yet shipped out any Phenoms--quad-core, triple-core, or otherwise--for testing, we can't say how they will match up with Intel's latest. Though they lack SSE4, their design and projected specs indicate that they should perform well. AMD has also started exploring methods of linking the GPU and CPU as an alternative way to improve multimedia performance.
What It Means
For the moment, most users would gain little advantage in upgrading to a Penryn CPU--the chip may not be compatible with your motherboard, the market has few SSE4-optimized applications that would allow it to shine performance-wise, and it will certainly be extremely expensive. And rumors have hinted at new chip sets just down the pike that may offer superior support for the new processor line. So unless you simply must live on the bleeding edge, wait a few months to see how the market shapes up.
Penryn Speed: Minor Gains in Mainstream Apps (chart)
Intel's new high-end CPU, the 3-GHz Core 2 Extreme QX9650, showed only modest gains over its predecessor in our tests with mainstream applications.
 Click the icon to see our chart.
