In the PC market, Intel has been in top form for quite some time. The Sandy Bridge CPU and more recent Ivy Bridge graphics opened up a nigh unassailable lead for Intel over AMD across all major market segments. At the same time the overall market for computing devices has fundamentally shifted in favor of highly integrated System-on-Chip (SoC) solutions, particularly those found in mobile devices such as tablets and smartphones. The world of mobile devices is dominated by an entirely different set of players including Apple, Mediatek, Qualcomm, and Samsung. Almost uniformly, these mobile devices are powered by ARM-based SoCs that integrate the CPU, graphics, wireless and other functionality.
Intel’s mobile strategy to date has been slow and lackluster at best, embarrassing at worst. The first generation of mobile products was introduced in 2008. The 45nm Silverthorne processor was paired with a discrete 130nm chipset and a third support chip; the total dissipation for the solution was 3W. In 2010, Intel released Moorestown, a two chip solution that was still far too power hungry for the market. Last year, Intel started shipping truly viable products for smartphones and tablets. The 2012 Medfield platform is an improvement, it is a single chip 32nm solution, the performance is tolerable and it hits the right power budgets. However, Medfield relies on the archaic Atom core (Saltwell) and a front-side bus (FSB) SoC fabric; choices which were made largely on the basis of availability and time to market, rather than excellent microarchitecture. Despite the less than stellar provenance, Medfield seems to be reasonably competitive with SoCs based on the ARM A9 and Qualcomm’s Krait. While Medfield did not garner many design wins, Clovertrail (a higher performance dual core relative) has done better in tablets.
Going forward, Intel’s fate in mobile depends on not just offering acceptable performance, but beating the competition. The upcoming Haswell architecture should solidly anchor Intel in notebooks and higher performance tablets. The three biggest steps for Intel in low power mobile devices are shifting mobile SoCs to the 22nm FinFET process, integrating LTE for smartphones, and delivering a new CPU microarchitecture, codenamed Silvermont.
Silvermont is Intel’s first low power core designed for 22nm. It is accompanied by a modern SoC fabric and will form the basis of products in smartphones (Merrifield), tablets (Baytrail and Valleyview), embedded networking (Rangeley), and microservers (Avoton). Silvermont adds support for a number of existing instruction set extensions, offering compatibility with Westmere, along with a few more modern tidbits, such as RDRAND.
Silvermont retains some similiarities to Saltwell, as both are dual-issue designs, but nearly every aspect has been fundamentally changed to enhance architectural efficiency. The result is a microarchitecture that is substantially different from any of Intel’s prior designs. The basic pipeline for Silvermont is 2 cycles shorter than Saltwell, and the branch mispredict penalty is 3 cycles lower, as shown in Figure 1.
Like its predecessor, Silvermont focuses on decoding and issuing x86 instructions into the back-end, rather than converting them into µops. However, Silvermont eschews multi-threading and instead hides memory latency and extracts single threaded performance through out-of-order execution. Intel’s architects use several clever techniques to reduce the number and size of power hungry structures in the critical path, particularly in the memory system. Silvermont is a 64-bit, out-of-order microprocessor that decodes and issues 2 instructions and dispatches 5 operations per cycle. Silvermont consumes under 1W/core and will be Intel’s CPU core of choice for low power SoCs in the 1-4W range, such as phones and tablets, as well as microservers, with shipments expected in the latter half of 2013.
Discuss (409 comments)