Intel's newest generation of desktop and mobile CPUs, code-named Ivy Bridge, debut today. In our tests, the chips showed moderate overall performance gains and substantial graphics improvement, all while sipping significantly less power.

Built in the company's 22-nanometer, tri-gate manufacturing process, the new CPU contains 1.4 billion transistors in a scant 160mm2 die area. The CPU includes the redesigned Intel HD 4000 graphics processing unit, which delivers the best integrated graphics performance we've ever seen. In fact, in our tests, HD 4000 graphics blew away the performance of an entry-level discrete graphics card. 

Today, you'll find Ivy Bridge chips primarily in desktops and all-in-ones. The first Ivy Bridge chips on the market will be the higher-wattage models best suited to those systems and to powerful desktop replacement laptops (you'll see a few Ivy Bridge models in that segment, too). It will take another six weeks or so for the dual-core and lower-wattage models to hit the market, so the majority of thin-and-light and everyday affordable laptops won't update to the new chips until then. 

So how will you be able to tell whether you have an Ivy Bridge CPU in your new computer? Look for the "3rd Generation Core Processor" brand name. You can also tell from the model number: Systems with a 3000 number after "Core i5" or "Core i7" use Ivy Bridge, while systems with a 2000 model number are Sandy Bridge. So, a Core i5-2600K, for instance, is a current-generation Sandy Bridge processor, while a Core i5-3550 is a new Ivy Bridge model.

Will you pay more for Ivy Bridge? Not really. Enthusiasts buying parts to build their own systems might pay a little more at first when supplies are scarce, but you shouldn't expect a significant difference in price between complete systems made with Ivy Bridge chips and the similarly configured systems of today using Sandy Bridge.

A moderate performance boost

Intel estimates that the Core i7-3770K should run most processor-intensive chores about 5 to 10 percent faster than the similarly clocked Core i7-2700K Sandy Bridge CPU and our tests generally bore that out.

The PC World Labs ran WorldBench 7 on three platforms. Our baseline system is built around an Intel Core i7-2600K (not 2700K), which runs at 3.4GHz and has a maximum Turbo Boost clock of 3.8GHz. The baseline system also includes an Nvidia GeForce GTX 560 Ti discrete graphics card.

You'll see three systems compared below. The Core i7-3770K is the new Ivy Bridge processor; it was tested both with the Intel HD 4000 integrated graphics and with a GeForce GTX 560 Ti graphics card. The Core i7 2600K is the current-generation Sandy Bridge processor, equipped with a GeForce GTX 560 Ti graphics card. This is our "baseline" WorldBench 7 system used to determine what a WorldBench score of 100 should be.

When both systems were equipped with the discrete graphics card, the Ivy Bridge CPU scored 10 percent above the baseline -which is greater than a mere 3 percent clock-rate disparity would indicate. Also, the Ivy Bridge system running with only the Intel HD 4000 GPU still scored about 4 percent higher than the Core i7-2600K. Bear in mind that WorldBench 7 does not incorporate 3D gaming-graphics tests.

Overall, the Ivy Bridge CPU looks to perform slightly better than an equivalent Sandy Bridge CPU, while consuming less overall power. That’s a win on the desktop side, but the real gains will likely be on laptop systems, where the performance gain combined with a lower power draw may result in laptops that offer longer battery life while improving performance.

Next Page: The Architecture

Intel uses what it dubs the “tick-tock” model of product design. The “tick” refers to rolling out a new manufacturing process--in this case, 22nm. The “tock” occurs when Intel ships a new microarchitecture using an existing manufacturing process. The current-generation Core i5/i7 Sandy Bridge CPUs represented a “tock.”

So the new Ivy Bridge processor is just a “tick,” right? The same thing on a new manufacturing process? That’s mostly true: The x86 CPU part of Ivy Bridge contains only minor tweaks to the current Sandy Bridge architecture. However, the GPU inside Ivy Bridge is a substantial redesign.

Basic speeds and feeds

On the surface, Ivy Bridge seems startlingly similar to Sandy Bridge. Comparing the two high-end models of the desktop CPU product line reveals almost no surface differences.

Two key differences are immediately noticeable, and suggest what Ivy Bridge is really about. At the reference clock frequencies, Ivy Bridge is rated at 77W, versus 95W for Sandy Bridge. And the die size, at 160mm2, is 25 percent smaller. Ivy Bridge is both more power efficient and cheaper to produce--in terms of numbers of die per wafer--than Sandy Bridge. The Core i7-3770K is the high end of the Ivy Bridge lineup at launch, while the 3.4GHz Core i7-2600K was the high end at the initial launch of Sandy Bridge. It’s likely, then, that we’ll see higher-clocked Ivy Bridge CPUs in the future.

Another interesting difference is the clock-frequency disparity between the two graphics cores. Ivy Bridge’s GPU runs 200MHz slower than the Sandy Bridge GPU. However, the Ivy Bridge HD 4000 GPU offers additional performance since it has more execution units (16 versus 12) and texture units (two in Ivy Bridge versus a single texture unit in Sandy Bridge). 

Ivy Bridge processor enhancements

In addition to the die shrink to a 22nm process, Intel made a few tweaks to the Sandy Bridge architecture for Ivy Bridge. Many of these adjustments - including efforts to improve instruction efficiency by increasing the number of instructions per clock - aren’t obvious from a look at the spec sheets.

One interesting aspect of Ivy Bridge is just how large a portion of the chip the new graphics engine consumes. Toss in the memory controller and display I/O, and Ivy Bridge is almost a System-on-Chip, though it lacks on-chip I/O for networking, USB, and storage.

One new feature is the incorporation of a sophisticated, on-chip, digital random number generator. This addition improves overall encryption security by making key generation less predictable; the DRNG is exposed via a new CPU instruction, though, and may be used by any application needing better random number generation. Supervisory mode execute protection is another new feature designed to protect against certain types of malware attacks on the processor.

Perhaps the greatest focus of the Ivy Bridge design is improved power efficiency. Enhancements include:

  • DDR memory I/O Power Gating: When the processor enters deep sleep--which it can do even during brief idle periods - the power needed for memory I/O is minimised.

  • Configurable TDP (thermal design power): One particular CPU product can support multiple TDP points, which allows OEMs to build CPUs into cases with different thermal envelopes while maximising performance.

  • Power Aware Interrupt Routing: This mouthful simply means that application tasks or threads can be routed to a particular CPU core based on power-efficiency needs, rather than simply to maximise raw performance.

Some additional features, such as more-granular frequency adjustments for memory and higher multiplier ratios, exist mainly for overclockers who want better control of the overclocking or underclocking capabilities.

The Z77 platform

Along with the new CPU comes a new chipset, the Intel Z77. Intel finally adds native USB 3.0 into its core logic with the Z77, supporting up to 14 USB ports in total, with up to 4 of them being USB 3.0 (SuperSpeed) capable.

Intel regards the motherboard chipset--which is really just an I/O Controller Hub - as only one part of the platform. The integrated memory controller and PCI Express controller built into the Ivy Bridge CPU itself is the other half of the platform. Like Sandy Bridge, Ivy Bridge supplies 16 PCI Express lanes usable for graphics. The full complement of 16 lanes is allocated to a single installed graphics card; adding a second GPU means that each graphics card gets just eight lanes, but given the overall bandwidth available with PCIe 3.0, most users won’t notice any real limit to performance in that configuration.

The I/O Controller Hub adds another eight PCI Express lanes, but those are better used for expansion cards such as audio and networking, rather than graphics, since going through the Z77 I/O controller, up through Intel’s DMI interface, and into the CPU would likely add more latency than you'd want for your graphics card.

The extra PCIe lanes might also be used for Thunderbolt I/O, helping to bring the interface to PC users, though not every motherboard may offer a Thunderbolt connector. (The Gigabyte Z77-UD3 motherboard we used for testing, for example, didn’t include Thunderbolt.) Regardless, Thunderbolt will still require a separate controller chip, as it does today. Display support has broadened, with full support for up to three simultaneous, independent displays.

Storage support includes up to two 6-gbps SATA and four 3-gbps SATA connections. Some boards may include Intel’s Rapid Storage technology feature, which allows users to add a small solid-state drive to act as a fast, persistent cache for standard hard drives. Rapid Storage made its debut with Intel’s earlier Z68 chipset.

The Z77 occupies the high end of the Ivy Bridge platform spectrum. Intel will also ship the H77 and Z75 chipsets, which will likely be designed for lower-cost platforms. Both remove some features, such as Thunderbolt, or offer support for only a single PCI Express graphics card.