Intel debuted Hyper-Threading on the Xeon line before moving it down into the Pentium 4 series. As should be expected, initial observations about HT's efficacy were mixed following the technology's launch, but the industry unanimously agrees that performance gradually increased over time. By early 2005, Intel was releasing Xeon-based benchmarking numbers showing a 27% gain in WebBench and a 33% gain with SPECjbb2000 when running with HT enabled. Most consumer apps and benchmarks that made use of HT consistently showed improvements well into the double-digit range. Soon enough, HT-enabled CPUs approached the cost of their older, less efficient counterparts, and today only a bare handful on non-HT Pentium 4 SKUs (the 505, 505J, and 506) are still in production.

The Pentium D (now updated with virtualization in the 900 series) is essentially two 600 cores sitting side by side. As you may remember, the original Pentium 4 Prescott core took some heat over its thermal properties. Well, the 500J series integrated some new power controls and implemented XD-Bit functionality. The 600 series kept the 500J Prescott and added 64-bit (EMT64) support as well as Enhanced Intel Speed Step (EIST). This was the prelude necessary before a truly reliable, versatile dual-core desktop chip could arrive.

The Pentium Processor Extreme Edition and contemporary Xeon chips all include HT as well as dual-core. This means that each chip, while containing two physical processors, actually yields four logical processors. (The primary difference between these two families is that the Xeon platform can run multiple processors while the Extreme Edition is a single-chip solution.)

Before you bust out the bubbly over quad-thread processors, though, realize that whether all four of those logical processors are being used ultimately depends on the application(s) running. Multi-threaded computing relies on applications designed to utilize thread-level parallelism (TLP), which essentially means that they can break workloads into multiple threads. However, software designers have the ability to pick the maximum number of threads that can be supported. A title coded for dual-thread support is going to show little or no benefit moving from a Pentium D to an Extreme Edition, all other specs being equal. If you were to grab a copy of PCMark05, for instance, and use it to benchmark a 2.80GHz dual-core Xeon against a 2.80GHz standard Xeon, both in dual-processor configurations, your results would be quite similar. The reason is that PCMark05 only recognizes up to four threads, and a dual-core, dual-processor config with Hyper-Threading enabled represents eight threads.

Back in the real World

Dual-threaded apps now pervade the market, especially when it comes to audio or video encoding. But how many apps on the market support four threads? The answer is tough to find. Even Intel's list is still evolving. Most multi-threaded titles are developed within the academic and supercomputing worlds. Those that exist in the desktop domain seem to congregate around multimedia handling, which makes sense since this and gaming are the highest-bandwidth applications in the consumer world. Because demonstrating value to customers is critical in the multi-thread sales process, here's a quick rundown of the quad-thread titles that were ready when Intel launched the Pentium Processor Extreme Edition in 2005:

• Adobe (www.adobe.com): Photoshop CS, Premiere Pro, and After Effects all were ready for the dual-core arrival, ensuring that the world's favorite desktop photo and video production apps were ready to thread like mad. Perhaps more interesting for everyday consumers, the scaled-down Elements versions of Photoshop and Premiere are also quad-thread-ready.
• Discreet/Audodesk (www.autodesk.com): 3D Studio Max 7.0 (now 3ds Max 8) is one of the industry's most powerful 3D content creation packages, capable of generating animations ranging from planetary movement in a school tutorial to the biggest blockbuster action games.
• MAGIX (www.magix.net): Far less known as a software vendor than it should be, Magix's MP3 2004 Diamond (now MP3 Maker 10 deluxe) is a hot audio encoder and transcoder packed with a slew of consumer-friendly mixing and filtering tools.
• Maxon (www.maxon.net): Half application and half platform, CINEMA 4D R8.1 (now R9.5) is a leading modeling, animation, and rendering package that makes extensive use of effect and feature modules so that users only buy what they need. CINEBENCH 2003 (now 9.5) is a free benchmarking tool based on CINEMA 4D that, thanks to its ability to task up to 16 multiprocessors on one system, is a terrific showcase for multi-threaded systems.
• Microsoft (www.microsoft.com): Strange but true in this lineup, Windows Movie Maker 2.1, Microsoft's newbie-oriented but nonetheless robust video editor, was one of the first quad-thread desktop apps on the market--and probably the only free one.
• NewTek (www.newtek.com): Much like 3ds Max, LightWave 3D is a rendering and animation program enjoyed by everyone from hobbyists (thanks to a relatively low price) to the largest motion film houses.
• TMPGEnc Net (www.tmpgenc.net) TMPGEnc 2.521 Plus (now 2.524) is a widely used AVI-to-MPEG1/2 encoder tool.

Surprisingly, PC gaming has been quite slow in making the jump to multi-threading. The shift didn't really start until 2004 and didn't take off in earnest until just last year when Epic's Tim Sweeney grew vocal about the challenges (2X to 3X the development time vs. single-thread) and rewards of making the jump to multi-thread right about when Epic's multi-thread-enabled Unreal Engine 3 arrived. Apparently, the extra time and expense of adapting to multi-thread is still taking a toll as gaming companies by and large have yet to embrace the update en masse. No one doubts that the move will happen, though, most likely as new engines such as Unreal 3 gradually arrive. Major games take about two years to develop, and dual-core chips are still in their infancy.

"Today, we have voice over IP in Unreal Tournament 2004," notes Epic Games vice president Mark Rein. "Down the road when everyone has multi-core, we will be able to enhance in-game user-to-user communication by adding real-time high-quality video. Imagine a window in the game where you conduct real-time video communication with your teammate while the game is taking place. That's going to be very cool, but it will take a lot of CPU power to accomplish that while still rendering a high-definition game environment. That is something that is not practical today in a single-core environment but is completely plausible with multi-core. You could even expect to see multiple video windows in the future."

The good news is that applications don't have to be multi-threaded in order for a dual-core system to yield big performance improvements today. So long as the CPU, BIOS, and operating system are all on the same dual-core page, the processing benefits of multiple cores can be applied to multi-tasking single-thread applications. Asking a single CPU core to handle HDTV encoding, transcoding an MPEG-2 file into WMV or DivX, and running an antivirus scan in the background concurrently can cripple even the fastest processors. Dual-core, and especially dual-core with Hyper-Threading, makes such heavy lifting feasible.

Click to see Processor Sequence Current Feature Comparison

Core Duo's Quantum Leap

PC buffs may look back on the release of Intel's Core Duo as the most significant event of 2006. The Core Duo is in large part a Pentium M with a faster FSB (667 MHz over the M's 533 MHz) and two processor cores courtesy of the 65 nm process shrink. With T-series parts showing 31W power consumption and L-series SKUs sipping a slight 15W, the Core Duo delivers a quantum leap in performance balanced against power savings.

Core Duo is today's chip of choice for diminutive living room entertainment systems, so be careful to think of this processor as a desktop solution in addition to its Centrino Duo alter-ego. There are plenty of environments where high-performance/low-power solutions make sense, and don't expect the corporate world to be left waiting for long. Imagine the benefits of a near-silent, power-saving, hardback book-sized PC in cubicles that still has the horsepower to run 3D presentation slides, multiple VoIP feeds, and stream a corporate executive briefing video simultaneously--and affordably.

Direct comparisons between Core Duo and conventional desktop chips are rare and involve many variables. However, we've done some Core Duo vs. Pentium M testing in-house in order to judge the difference made by that extra core. With 2.13 GHz chips working under Windows Media Encoder, we encoded a 1080p WMV HD file in 5:07 on the Core Duo while the same operation took 10:30 on the Pentium M.

In a preliminary look at the Core Duo vs. Pentium M on an AOpen desktop platform, AnandTech recorded some remarkable if predictable numbers. In SYSMark 2004's Overall Internet Content Creation, the 2.0 GHz Core Duo T2500 scored 272 while the 2.0 GHz Pentium M 760 hit only 191. Encoding Star Wars Episode VI down to 4.5GB with DVD Shrink, the Core Duo blew through the task in 8.1 minutes while the Pentium M took 13.4 minutes.

With its twin dies, the Core Duo benchmarks very similarly to standard dual-core desktop processors. We expect soon enough to see reviews emerge showing mid-level Core Duo SKUs acing out low-end desktop dual core parts, particularly in applications where the Core Duo's superior cache and memory addressing outweigh its slower FSB speed.

TWO COREs, One Direction