Intel’s next generation Sandy Bridge seems aimed at the hearts and minds of mobile and mainstream PC users
If you’re a performance enthusiast, then it’s unlikely you’ll be chomping at the bit to build a new system based on Intel’s Sandy Bridge CPUs when those processors ship in early 2011. That’s because the CPU Intel is calling their second generation Core architecture is aimed at the vast majority of mainstream users – and mostly laptop users a that.
First, let’s go over a quick rundown on the CPU itself. Sandy Bridge is built on Intel’s 32nm process. Intel’s “tick-tock” product creation process dictates that a brand new architecture be built on proven manufacturing capabilities to minimize risks to the design.
The new CPUs will offer on-die graphics, a first for Intel mainstream CPUs. The graphics core shares a high speed ring bus and what Intel calls the “LLC”, or “last level cache” (typically the L3 cache on current gen CPUs.) The CPU will also include the latest AVX instruction set, which will complement and eventually replace the older SSE media and floating point instructions.
The actual CPU cores are enhanced versions of the current Westmere cores, and Intel will be shipping versions with two and four cores built in. They’ll support Hyper-Threading (simultaneous multithreading), like the current Core i5/i7 generation. The onboard memory controller is limited to two channels of DDR3, and external, discrete graphics support is limited to one PCIe x16 or two PCIe x8 slots.
Some new features have been added to the integer cores, including improvements in branch prediction, dual load/store ports and a dedicated micro-op cache for decoded instructions. While each of these improvements is incremental, the overall effect is a an increase in performance with a decrease in power usage.
The floating point unit is a substantial enhancement. The AVX engine is a fully 256-bit wide floating point engine that doubles overall floating point performance over the older, 128-bit SSE engine.
The integrated graphics core is said to significantly improve performance over older Intel CPUs. Intel demoed Starcraft 2 running at good frame rates, but divulged no details on graphics settings or resolution. However, the new graphics core is still a DirectX 10 GPU, and won’t support DirectX 11. As with the CPU cores, the graphics cores are scalable, so different Sandy Bridge models may offer differing levels of graphics performance.
While Sandy Bridge graphics won’t support DX11, Intel estimates that graphics performance will be over 20x (“up to 25x”) faster than the GMA 4500 from several years ago – safely hitting their “10x improvement in graphics performance by 2010” mantra. The GPU, in conjunction with the AVX floating point engine, should offer big gains in video transcoding performance as well.
Since the GPU is now built onto the CPU die, it will support Turbo Boost – the ability to run at higher than default clock rates to ramp up for brief periods at higher than default clock speeds, as long as certain voltage and heat thresholds aren’t exceeded. Turbo Boost will work independently and dynamically, letting the CPU or GPU scale up performance independently as needed.
One of the more interesting design decision for Sandy Bridge is building the GPU behind the last level (typically L3) cache. This allows the GPU to share the high level cache with the CPU cores. The graphics driver determines which graphical elements are cached. Hitting the cache instead of having to go to main memory results in big bandwidth and power savings for graphics.
All of these disparate elements need to share or move data at high speeds. Facilitating this is a high speed ring bus, capable of peak throughput of 96 GB/second per connection. That “per connection” phrase is significant, as Sandy Bridge is highly modular. The amount of cache is tied to the number of cores, and the overall maximum throughput is also dependent on the core and cache.
For example, a four core Sandy Bridge CPU can move up to 384 GB/second over the ring bus, while a dual core CPU may max out at 192GB/second. That’s not really a limitation, since the amount of computer power and data needed decreases as the number of cores decrease.
Managing all these disparate elements is the System Agent. Think of the System Agent as an on-die north bridge, bringing together the memory controller, PCI Express controller, PCU (power control unit), DMI and display controller. The PCU is an enhanced version of the PCU found on Nehalem, and handles overall power management for the entire CPU (including graphics.)
In many ways, Intel’s new CPU will be akin to the Arrandale – targeted squarely at laptops and mainstream destkops. Unlike Arrandale, there will be four core versions available, though product information on specific CPU releases isn’t available yet. Even the socket format is similar: socket 1155. Yes, that means new motherboards if you want to go with Sandy Bridge.
While performance enthusiasts in love with their six core Westmere systems may not be too excited, Sandy Bridge should offer substantial improvements in graphics and CPU performance for mainstream laptops. The current mobile quad core CPU lineup has been looking increasingly anemic, but when the quad core Sandy Bridge CPUs hit the scene, that will change.