Inside 'Sandy Bridge': Intel's All-New CPU

The term "modular" seems to be a common one when it comes to upcoming processor designs. AMD will use a dual-integer-core module to power its "Bulldozer" processor line, and now Intel has jumped on the same bandwagon: The silicon giant's 2011 "Sandy Bridge" architecture will utilize modular and highly flexible technology.

This is partially due to the inherent challenges multicore processor designs face. The customer base is crying for a greater number of cores and never-before-seen performance levels, while keeping a very close eye on the electricity bill. That means both Intel and AMD need to keep cranking up the performance, while maintaining power and heat requirements at reasonable levels.

At last week's Intel Developer Forum, the chipmaker took the wrapping paper off its upcoming Sandy Bridge processor release, confirming some rumors while negating others. It's an ambitious design, with the processor featuring more than a billion transistors, but it remains consistent in other respects, such continuing to use the Core i7/i5/i3 brand name.

Although previous Intel multicore CPU generations have used traditional multi-CPU architectures, Sandy Bridge is a true single-chip design, and will feature a modular package with all the components on one piece of silicon. This ensures that the chip remains as small and cool-running as possible, while adding new features like on-die integrated graphics.

Sandy Bridge has been built from the ground up, and is very flexible in terms of producing a variety of processors for different markets. It easily supports the removal or addition of not only the integrated graphics component, but also of CPU cores and caches.

This ensures there will also be a wide range of different Sandy Bridge models and features, with some models shipping unlocked, while Hyper-Threading support will vary based on model. The new 32-nanometer-process CPUs will require a new socket design, LGA1155, which should come as no surprise to the Intel faithful.

Initial clock speeds will range from 3.1 to 3.4GHz and L3 cache levels will start at 3MB for dual-core and go up to 8MB for high-end quad-core processors. The onboard memory controller will support dual-channel DDR3, and Intel has upgraded its Turbo Boost auto-overclocking feature to be even more powerful.

One Ring to Rule Them All

A key feature to the Sandy Bridge microprocessor is the ring interconnect, which not only lowers the number of wires on the die, but also enhances its flexibility. The ring interconnect is the communications hub for the entire CPU, with each of the components "talking" to the ring and having that information forwarded to the cache and system agent.

This ring interconnect is also segregated into four 32-byte subdivisions, Data, Request, Acknowledge, and Snoop rings, with logic behind this system -- a 32-byte ring perfectly divides a 64-byte cache line into two different packets. The arbitration system is localized and distributed at the ring level to scale better to frequency increases and provide an easy transition to additional cores.

It also scales well to clock speed, and runs at the same core frequency as the processor. This is a significant design step, as previous CPU architectures had components that would not scale, and therefore held back higher-clocked processors from achieving performance in line with the core speed increase. The ring interconnect is actually based on high-end server technology and its speed potential is incredible -- at 3.0GHz it can provide up to 96GB/sec of bandwidth.

The system agent comprises the basic functionality of a Northbridge chip, and includes features like the memory controller and PCI Express interface. It is also highly integrated with the ring interconnect to provide the CPU/GPU with low latency/high bandwidth access to system memory. It also handles the power management for connected PCI devices and system memory.

Intel now refers to the L3 cache as Last Level Cache, or LLC, and since it now runs at the full clock speed and makes full use of the ring interconnect, cache latency is reduced and performance is four times that of the previous generation.

Graphics on a Di(m)e

Sandy Bridge is the first Intel processor to feature integrated graphics directly in the CPU die. Previous offerings were multichip efforts that were more of a stopgap solution to AMD's impending Fusion release. But Sandy Bridge is the real deal, with a 32-nanometer graphics core fully integrated into the CPU silicon.

This level of integration allows Intel to bring down the overall power requirements compared to a multichip format. With the graphics core directly on the CPU, power efficiency is improved and the GPU can potentially make use of the same CPU power saving techniques. By fitting into the processor TDP, it also lowers overall power consumption and thermal demands.

There have also been improvements to the Sandy Bridge GPU, with enhanced graphics performance and better overall throughput. The shared LLC not only boosts applications, but it will have a noticeable impact on integrated graphics performance.

Exactly how high that performance ends up being is still a question, although Intel has bandied about the usual hyperbole about how it is 25 times the performance of its top integrated GPU from 2007. The company hedged away from naming specific 3D games and frame rates, falling back again on the tried and true "mainstream performance" idiom.

Realistically, simply due to its on-CPU integration, there are natural concessions. Cooling is certainly a concern, as there would be no way for Intel to put a high-end GPU like a Radeon HD 5970 on a CPU as it would rip the power and thermal envelope to shreds. Even a mainstream GPU would be problematic, as these require 100 to 200 watts of power. Intel also needs to slide the same basic design into low-power mobile systems, and that has an impact.

Sandy Bridge includes a Media Engine processor that handles the basic media tasks like video encoding and HD video playback. In the age of YouTube and portable media devices, video encoding and decoding is an extremely important graphics component, and from early indications, this may be an area where Sandy Bridge can compete with AMD and Nvidia. In fact, it could very well be a major selling point.

While a definite improvement, a Sandy Bridge integrated GPU is not going to replace a dedicated graphics card. Performance will be higher than previous generations, but the main benefits revolve around basic economics. Sandy Bridge will be less expensive for Intel to produce, less expensive for the system vendors, and hopefully, this will translate into lower prices for consumers.

A Sandy Bridge Scratch-and-Win?

Another interesting development on the Intel front was the announcement of a unique CPU upgrade program. It is similar to a gift card approach, whereby owners of certain CPUs can buy a $50 scratch-off card for the Intel Upgrade Service to fully maximize the potential of their systems.

These improvements include enabling Hyper-Threading and activating additional L2 cache levels, taking the Pentium G6951 from 2MB to 4MB of L3 cache with HT. Assuming it happens at the hardware level, this idea actually has some merit, as it could provide an on-demand method of upgrading CPUs, all without getting your hands dirty. This would be extremely useful for improving CPU performance on laptops, and the only danger is if it is hacked.

Currently, this is a pilot project involving the Pentium G6951 processors with select vendors, but if the Intel Upgrade Service turns out to be successful, the wide variety of upcoming Sandy Bridge processors could make this an easy transition to the new line.

Although Intel certainly doesn't need any help, Sandy Bridge is a very impressive microarchitecture, and one that will likely widen the gap with AMD. CPU performance will be higher and that ironclad Intel brand name will ensure business approval at the highest level. The improvements in the GPU are the most promising, and even the ability to run basic 3D applications and games would be a step forward for Intel.

The most amazing part of the upcoming Sandy Bridge release is Intel's belief that they have finally cracked the multicore nut and believe that this microarchitecture will be sustainable well into the future. This is a surprising position, especially as Intel has historically spoken about an upcoming processor while having its successor hidden in its back pocket.

Intel certainly hopes this combination of performance, features and value will translate into increased sales for the new processors. The vast majority of business desktops and notebooks do not require high-end graphics, and bean counters are more interested in the bottom-line purchase price. Since mobile systems make up the majority of business purchases, Sandy Bridge looks exceptionally suited to make a real impact in 2011.