AMD leaps years ahead with new notebook platform

Notebook buyers might find it difficult to get excited about a new notebook PC platform. Notebook buyers hear “new” applied to notebook technology all the time, but when they check out the supposedly new, it’s not only indistinguishable from what came before it, but every brand in a given price range is indistinguishable from all the other brands.

AMD’s notebook platform, comprising the completely redesigned Griffin CPU and a tightly linked Puma chipset, attacks the familiar twin objectives of high performance and low power with a refreshing twist: AMD breaks the performance/power efficiency relationship of inverse proportion that plagues all existing notebook platforms.

The one-chip Griffin CPU combines a completely redesigned microprocessor with dedicated, expanded Level 2 cache for each core, an incomparably fast HyperTransport 3 I/O bus controller, and an on-board, power-scalable DDR2 memory interface. AMD’s design is inspired by SoC (system-on-chip) circuitry used in ultra-low-power embedded devices. Interestingly, the advantage that Griffin gains from pulling everything into one highly integrated SoC is independent control over the power used by each system component. Each of Griffin’s two cores has a broad range of independent settings for clock speed and voltage level, covering the spectrum from deep sleep to full bore. The memory controller has a separately controlled power supply as well, and the HyperTransport 3 bus scales from bus widths of 2 to 16 bits with each narrowing of the bus resulting in a proportional reduction in power.

Like no x86 architecture before it, AMD’s notebook platform’s power state changes take place within the span of one clock cycle with no limits on the combinations of component power states or the magnitude of the change. In today’s PCs, a power state switch can only be performed within a limited range, and a delay of several CPU clocks is required for the new frequency and voltage to stabilize, during which delay the CPU is completely disabled. AMD’s platform can literally swing each core, the memory controller, and the I/O bus, together or independently, from deep sleep to full power and every imaginable state in between in one clock tick. That’s not only unprecedented, it’s impossible on any other x86 architecture.

The Puma half of AMD’s notebook platform is a two-chip solution comprising the RS780M north bridge and SB700 south bridge. Puma switches power states automatically and in lock-step with the Griffin SoC. Puma’s integrated 3-D accelerator is designed by the industry’s leading GPU engineers, not CPU designers just getting their feet wet in graphics. Puma delivers the power savings and cost benefits of unified memory access graphics while clearing the high bar set by Microsoft DirectX 10 compliance and Vista Premium certification. If a portable desktop or gaming notebook design calls for a dedicated GPU, Puma has a HyperTransport 3-to-PCI Express 2.0 bridge with 16 full-speed lanes set aside for just that purpose. Puma can switch, seamlessly and automatically, between an integrated display controller and a discrete GPU when the notebook is switched from battery to AC power, respectively.

AMD’s new notebook platform gives notebook manufacturers practically unlimited freedom to build creative designs that differentiate their models from others. But all notebooks built on AMD’s mobile platform will share two key traits: best-in-class performance and long battery life. Notebook buyers will soon forget the days when they had to choose one or the other.

(The unabridged version of this article can be found in Tom Yager’s Ahead of the Curve blog.)