The New Intel Core™ i7 and Simulations

by John Reynolds



i7 badgeIntel has been riding on a tremendous wave of success over the last two years, largely due to the market reception of the Core 2 Duo processors. Introduced in June of 2006, the original Core 2s were a tock, a new micro architecture, and were followed in late 2007 by a tick, an update to an existing CPU design. Yet hardware enthusiasts have been well aware of a new tock code-named Nehalem that was rumored to be the most significant architectural overhaul from Intel in years. Unveiled on November 2, Intel’s Core i7 CPU launch saw a new processor design with an integrated memory controller that put to rest the aged front-side bus, a new socket format, reintroduces simultaneous multithreading, and a host of other micro architectural tweaks that we’ll cover here shortly. The initial Core i7 rollout consists of three products hitting the market this November, ranging in price from $999 down to slightly under $300 (in quantities of 1,000 and barring a high demand that sees e-tailers raising prices above MSRP).

Core i7 brings some significant improvements to the Core 2 design. And considering its transistor density and relative chip size, this comes as no surprise. Core i7 is comprised roughly of 731 million transistors fabricated using the same 45nm, high-k process as the newest (Penryn) Core 2 CPUs. In comparison, AMD’s Phenom quad cores consist of 463 million transistors. A larger chip than Core 2, Core i7 also sees the introduction of the LGA1366 socket, which is similar in design to the older LGA775 format, just physically larger and with more pins. All three Core i7 CPUs to be launched this month are 130W TDP parts.

True Quad-core Design

As mentioned above, the Core i7 is Intel’s first true quad-core design. Each core contains a 32 KB L1 instruction cache, a 32 KB L1 data cache, and a 256 KB L2 cache (both L1 and L2 caches are 8-way set associative). Separate from the four cores is the much larger, smarter 8 MB L3 cache (16-way associative) that allows the individual cores to share data. Worth noting is that while larger than AMD’s Phenom (2 MB L3 cache) caches, Intel’s new architecture allows for lower latencies than the competition’s design. Intel has also finally caught up with AMD by using an integrated memory controller, moving it out of the north bridge and on-die with the processor. The new controller supports only DDR3, three channels per socket. And this integrated controller sports a higher clock or frequency than one traditionally found in the north bridge, though it does not match full CPU speeds. Along with the new controller, Intel has also developed a point-to-point system connect similar to AMD’s HyperTransport known as QuickPath Interconnect (QPI). QPI can move data between the CPU and I/O chip at up to 25.6GB/s for the higher-end parts, and at 19.2 GB/s for more mainstream CPUs.

The Return of SMT

Simultaneous multithreading, better known as its marketing name of Hyper-Threading, makes a return with Core i7. Introduced with the latter P4 CPUs, SMT was not supported by the Core 2 architecture, but each core processor in Core i7 can execute two independent threads, though some hardware resource sharing is required to implement this design. This increased parallelism will take advantage of multithreaded software to increase performance of such code, though performance certainly won’t be doubled due to resource sharing involved. A quad-core design with SMT means support for eight total threads, a level of parallelism that will be a challenge for the vast majority of game designers to fully exploit. Core i7 also supports the latest SSE enhancements Intel has made, bringing the instruction set up to version 4.2, for faster media transcoding and 3D graphics performance.

Advanced Power Management

Like AMD’s Phenom design, Core i7 can vary each core’s clock speed independently based on processing demand. In fact, if one core goes into a deep idle state, power to that core can be completely shut off. This is handled by a dedicated on-chip controller for advanced power management that also handles Core i7’s new Turbo mode. Turbo mode works transparent to the OS and is designed to increase individual core frequencies if needed. The power management controller treats Turbo mode as yet another power state, raising baseline clock speeds by up to several tiers (3.2 GHz to 3.33 GHz or 3.46 GHz for the 965 CPU) for single-threaded applications, and by one tier across the cores for multithreaded apps.

The chart below shows some of the specification differences among the upcoming Core i7 CPUs:

Core i7 920
Core i7 940
Core i7 965
Clock speed
2.66 GHz
2.93 GHz
3.2 GHz
QPI rate
4.8 GT/s
6.4 GT/s

Intel’s review kit contained the 965 and 920 processors, which we tested against the Penryn-based QX9770. AMD has been attempting to compete against the Core 2s by massively slashing prices, leaving their Phenom lineup priced at more mainstream market segments so we decided to test the Core i7 against only the Core 2.

Intel X58SO “SmackOver” Extreme Motherboard

Also included in Intel’s Core i7 review kit was the X58SO (SmackOver) Extreme motherboard. As noted above, the memory controller has now been moved from its traditional location in the north bridge to the CPU, with Core i7 processors talking directly to system memory via QPI. The new I/O hub in the X58 chipset supports 36 PCIe 2.0 lanes, though board manufacturers have some flexility in how these lanes can be used on individual board designs. Two x16 slots are available on X58SO, but a different board may have three or four graphics slots (a single full x16 slot with two x8 slots, or four x8 slots). With a front-side bus no longer present, the X58 uses a default clock speed of 133MHz that acts as a sort of reference clock that drives all CPU-related speeds. The X58SO BIOS has a wealth of new options for the overclocker under Performance Overrides section, with new settings unique to the Core i7 processor that can be confusing at first, particularly due to the Turbo mode described above. An example of this is the CPU multiplier now being named the Maximum Non-Turbo Ratio and a 4-Core Ratio Limit setting that determines Turbo mode’s multipliers (both of these settings are only present with a Core i7 Extreme part installed). Hardware enthusiasts and overclockers will need a few moments to wrap their heads around the new verbage. Last on the X58 chipset is multi-GPU support. Like earlier Intel offerings, AMD’s CrossFire is supported and NVIDIA’s SLI will work fine with the new chipset so long as the motherboard’s manufacturer has purchased SLI certification from the company.

Intel® X58 Express Chipset Block Diagram

Intel® X58 Express Chipset Block Diagram

The Intel DX58SO (X58 chipset) Extreme motherboard

The Intel DX58SO (X58 chipset) Extreme motherboard

Go To Page 2

Powered by WordPress. Designed by WooThemes