Intel's first dual-core CPU benchmarked

A new era for desktop CPUs has arrived with the introduction of the dual-core processor. AMD announced as early as June 2004 its plans to manufacture dual-core processors, but the first dual-core desktop processors you'll actually see in shipping systems will have Intel's name on them -- specifically, Intel's new dual-core Pentium Processor Extreme Edition 840 CPU, along with the new 955X chipset.

What is dual core?
Simply put, dual-core technology places two independent execution units onto the same processor die -- think of it as two processors in one. This idea differs from Intel's HyperThreading (HT) technology, which uses a single (physical) execution unit but allows the processor to run two separate (logical) execution threads. Some of the dual-core processors will also include HyperThreading, so there will be some dual-core CPUs that support four independent threads (two of the threads are running on physical execution units, and two are running on logical units).

Making its debut
The first dual-core processors we'll see from Intel will be in high-end desktops, featuring a Pentium Processor Extreme Edition 840 CPU. That processor's execution units include 1MB of L2 cache and support HyperThreading -- for a total of 2MB of L2 cache and support for four execution threads. The CPU will operate on an 800MHz frontside bus (FSB) and will include Intel's new 64-bit Extended Memory 64 Technology (EM64T), making it fully prepared to take advantage of Microsoft's new 64-bit version of the Windows operating system when it becomes available.

You'll notice that we haven't mentioned clock speed yet. While the Extreme Edition 840 chip is clocked at a pedestrian 3.2GHz, its launch is particularly notable because it's the first time since 1998 that Intel has released an overclockable CPU. AMD's Athlon FX-55 -- because of its own overclocking potential and its excellent standard performance -- has become the enthusiast's chip of choice. By giving its new flagship CPU room to overachieve, Intel is making a strong play to reclaim its spot as the maker of the fastest processors in the land. Intel doesn't endorse overclocking, nor does it use the term anywhere in its literature. However, it's likely that vendors will take advantage of this new ability to be overclocked. And you can bet that Intel won't mind vendors doing so if it allows Intel-based systems to re-emerge at the top of comparative benchmark charts.

The Intel 955X Express is the new chipset for the dual-core Extreme Edition. It will support both 800MHz and 1,066MHz frontside bus speeds, 667MHz dual-channel DDR2 memory, and up to 24 PCI Express lanes. While the 955X will support dual x16 PCI Express slots, it does not currently support Nvidia's SLI (Scalable Link Interface) dual-graphics-card configuration. The reason is that one of the 955X Express PCIe slots clocks down to 4X, not the full dual-8X bandwidth that SLI requires.

The other dual-core CPU we'll see soon from Intel will be the 3.2GHz Pentium D. The Pentium D is almost identical to the dual-core Extreme Edition -- even down to the same number of transistors. The only significant difference is that the Pentium D will not support HyperThreading, so this CPU supports only two threads. The Pentium D will be paired with the upcoming 945 Express chipset.

Making it work
In order to use dual-core technology to its best advantage, the operating system and the applications need to support thread-level parallelism, which basically means running multiple execution threads simultaneously. Microsoft's Windows XP and more than 200 applications, such as Adobe's Photoshop CS and Roxio VideoWave 7.0, are multithreaded. Most of today's multithreaded applications are of the content-creation ilk, which tend to perform many operations in parallel. As dual-core technology becomes more prevalent, you can expect to see more multithreaded programs become commonplace -- for example, 3D-intensive games can take advantage of dual-core technology by using more robust physics and AI engines for more realistic effects and gameplay. But since Windows XP itself is multithreaded, you don't necessarily have to be running multithreaded applications to see a performance gain. Windows is a multitasking environment, and as such, there are usually applications running in both the foreground (such as the browser you are using to read this) and the background (such as real-time virus scanning). A dual-core processor should execute the multiple threads of these applications more efficiently.

Intel claims it saw from 50 to 124 percent performance gains for a dual-core-based system over one using a 3.73GHz Pentium 4 Extreme Edition on its own multimedia application tests. For more-mainstream programs, Intel estimates about a 40 percent performance improvement.

We have developed a number of benchmarks to test the potential performance gains provided by a dual-core CPU when compared to a single-core CPU running on the same chipset. Using a test PC submitted by Intel, we compared a 3.2GHz Pentium Processor Extreme Edition 840 to a 3.73GHz Pentium 4 Extreme Edition chip. Our preliminary test results back up Intel's claims, at least in part.

Of our seven tests, five were real-world multithreaded applications. In two of those tests, Adobe Photoshop and Apple iTunes, the dual-core 3.2GHz Extreme Edition 840 CPU beat the single-core 3.73GHz Extreme Edition chip -- most notably on the iTunes test, where the Extreme Edition 840 encoded a collection of MP3 files in 24 seconds, or 18 percent faster than its competitor. The Extreme Edition 840 also beat the 3.7GHz Extreme Edition on the 3D animation portion of Maxon's Cinebench 2003 benchmark suite by a full 56 percent. Also of note are the results of our multitasking test. While running a virus scan in the background with McAfee's VirusScan 9.0, we simultaneously encoded a video using DivX Networks' Dr.DivX software. The Extreme Edition 840 chip beat the higher-clocked competition by 17 percent. On this particular test, at least, the dual-core processor lived up to its promise of improved multitasking.

Before we can say that the dual-core chip completely exceeds expectations, we should point out that on the Sorenson Squeeze 4.0 video-encoding test and the VirusScan 9.0 and Dr.DivX standalone tests, the 3.73GHz Extreme Edition chip came out on top. But despite the fact that the dual-core chip didn't win, it's notable that the Extreme Edition 840 chip performed only 13 percent slower at worst (on the Sorenson test), where the 3.7GHz chip is 17 percent faster in raw clock speed. We'll need to spend more time with finished dual-core systems before we can make a complete evaluation, but our early impression is that the new technology shows great promise for improving productivity.

Waiting in the wings
Intel predicts that by the end of 2006, dual-core processors will have sufficiently infiltrated mainstream systems to the extent that roughly 70 percent of all desktops and notebooks shipped will use dual-core CPUs. Intel also expects to move over to a 65nm manufacturing process next year. Dual-core won't arrive on the mobile front from Intel until 2006, when the company releases its 65nm, dual-core mobile processor -- Yonah -- with 2MB of L2 cache that will be shared by the two cores. Yonah will be part of Intel's next update to the Centrino platform, currently codenamed Napa.

Now that Intel has launched the first true dual-core salvo, only time will tell how rival will AMD respond.


All tests were conducted on a 'white box' system supplied by Intel, with the following specification: Intel D955XBK motherboard; Intel 955X Express chipset; 1,024MB Micron DDR2 memory running at 667MHz; Sapphire Radeon X850XT graphics card with 256MB; 160GB Seagate ST3160827AS SATA hard drive; 550W Enermax EG651AX power supply; Microsoft Windows XP Professional SP2.

The two processors tested are:
-- 3.2GHz dual-core Intel Pentium Processor Extreme Edition 840 (800MHz frontside bus, 2MB L2 cache, 90nm, with HyperThreading)
-- 3.73GHz single-core Intel Pentium 4 Processor Extreme Edition (1,066MHz frontside bus, 2MB L2 cache, 90nm, with HyperThreading)