Major changes to Intel's CPU line-up are afoot with the introduction of Core i7 (codenamed Nehalem), which improves on the design of the Core 2. We won't beat around the bush: it looks like it's the CPU line-up to go for if you want to own the fastest PC rig on the block.

Core i7 CPUs won't fit in to existing motherboards, as they work in a different CPU socket to Core 2 Duo CPUs. They also require a new chipset and motherboard configuration (initially, boards based on the X58 chipset), so there isn't really an upgrade path for this CPU: if you want to build a system based on the new Core, you'll have to commence a PC build from the ground up.

It's a little strange to say it, but the brand new Core i7 actually borrows technology from the Pentium 4 line, as well as AMD's Athlon 64 CPUs, so its features aren't all completely new. However, as they're centred around a super-fast micro-architecture, they combine to produce a killer product, especially for PC enthusiasts, content creators and overclockers.

The micro-architecture is codenamed Nehalem and is built using Intel's smallest manufacturing process — 45 nanometres. However, it's physically bigger than the Core 2 Duo micro-architecture, as it packs a lot more features, which means it won't fit into a motherboard with an LGA775 CPU socket. Instead, it requires a motherboard with an LGA1366 socket.

Some of the features that immediately stand out are the integrated memory controller, which requires DDR3 memory, and the implementation of QuickPath Interconnect, which replaces the front-side bus link to the chipset and memory. AMD has had an integrated memory controller in its chips since the days of the Athlon 64, as well as a point-to-point bus — HyperTransport — between the CPU and the chipset, so these aren't brand new innovations. But they are welcome.

As the memory is not controlled by the chipset, and is actually controlled within the CPU, there is a separate connection for the memory traffic and for the I/O (input/output) traffic via the chipset. This can help reduce latency when the CPU requests data in applications that are heavily memory-dependent, as memory data doesn't have to compete with I/O data.

In Intel's design, the front-side bus has been replaced by a central clock, which runs at 133MHz and controls the speed of the CPU, memory and PCI bus components. Clock multipliers determine the speed of each component (for example, a multiplier of 24 results in the 3.2GHz speed of the Core i7-965 Extreme Edition model), but the multiplier for the CPU can be changed if you buy an Extreme Edition CPU. In fact, not only can you control the speed of the CPU, memory and PCI bus, but you can actually control the speed of each of the CPU's individual four cores. Conversely, you can leave the multipliers alone and just change the speed of the central clock.

What's interesting about Core i7 is that HyperThreading also makes a comeback. This was first introduced in the Pentium 4, and it allows each core to work on two tasks at the same time. This means that the Core i7-965 Extreme Edition can process up to eight threads at a time. What you are looking at is a virtual eight-core CPU. This is beneficial for users who run applications that are multithreaded, and can therefore take advantage of all the cores on offer.

In our testing, we used the Core i7-965 Extreme Edition (which is codenamed Bloomfield), running at its default speed of 3.2GHz, and it delivered a blazing performance. We ran the CPU in an Intel DX58SO Extreme Series motherboard, with 3GB of QIMONDA, 1067MHz DDR3 memory, an Intel 80GB High Performance solid-state drive and a Gigabyte GeForce GTX 260-based graphics card. Our WorldBench 6 test suite completed swiftly, and returned a score of 144, which is the fastest overall we've seen in this benchmark to date. It's approximately a 10 per cent improvement over the nearest Core 2 comparison — Desktop Board D5400XS (Skulltrail).

When breaking down the scores in WorldBench 6, the individual application times were as follows: 280sec in Adobe Photoshop, 320sec in 3dsMax, 202 sec in Firefox, 318sec in Microsoft Office and 159sec in Window Media Encoder.

These scores are important to note, as they are the key indicators in the platform's application performance when we went on to overclock it, and also when it ran on 2GB of RAM, instead of 3GB. We changed the CPU multiplier to 30 (from its default of 24), and bumped up the core voltage, to take the CPU up to 4GHz. At this speed it took 248sec to complete the Adobe Photoshop test, 302sec in the 3dsMax, 178sec in Firefox, 303sec for Microsoft Office, and 139sec for Windows Media Encoder.

As you can see, the extra 25 per cent straight-line speed increased the system's performance in all of these applications by up to 15 per cent in some cases. Air cooling was used to overclock the CPU (we used a Thermalright Ultra-120 eXtreme RT cooler supplied by Intel) and our test rig was stable, except for a couple of random restarts in the middle of one of WorldBench 6 test runs.

In the Blender 3D rendering and iTunes MP3 encoding tests, the CPU, while running at its default 3.2GHz, recorded times of 24sec and 46sec respectively. The iTunes test took longer than the Blender score as iTunes only uses a maximum of two CPU cores. When overclocked to 4GHz, the results weren't as fast as we'd hoped — the tests recorded times of 23sec and 45sec, respectively.

One of the other key differences between this platform and the Core 2 platform is that it uses a triple channel memory configuration, instead of dual-channel. That means you'll get optimal performance out of it if you use three memory modules, instead of two or four. You can populate all four slots if you have memory-hungry applications, but the latency will be affected. Likewise, with only two memory modules (for a total of 2GB), the system's performance was impacted noticeably; it recorded a time of 317 in Adobe Photoshop, for example, which is 37sec slower than what it recorded with 3GB. It will be interesting to see if motherboard manufacturers ever decide to implement nine memory slots for this platform.

As for the CPU's power consumption, it's actually very economical, especially for a high-end component. We measured its consumption while it was running at full capacity (running eight threads) at 3.2GHz, and its consumption peaked at only 221 Watts. It consumed 203W when running four threads, and it consumed 104W when idle. While switched off but still plugged in to an outlet, our test configuration consumed 15W. The CPU also contains dynamic speed adjustment so that the operating system can step down the CPU by up to 50 per cent, depending on what is being processed at the time. Cores that aren't being used can be switched off, and the cores in use can be set to run more quickly. This works primarily with applications that use one and two cores.

All up, there is no doubt that the Core i7 is efficient and plenty fast. You'll need a new motherboard to run it, as well as new memory, but if you've got the cash for a new setup, it'll be worth it.