There's more to Intel's next-generation processor family than the 45nm process node
Intel plans to unleash its
Penryn family processors next quarter, shortly after
AMD releases Barcelona.
Penryn is the umbrella for all 45nm Core 2 micro architecture products, including
quad-core Xeon Harpertown, quad-core Core 2
Yorkfield and dual-core Xeon, Core 2
Wolfdale processors.
On the surface,
Penryn looks like die shrink of
last year’s Conroe micro architecture, but Intel sought additional
tweaks to the micro architecture to achieve greater performance at the same clock speeds as
Conroe processors.
Intel improves existing Wide Dynamic Execution, Advanced Smart Cache, Advanced Digital Media Boost and Intelligent Power Capability, technologies that previously made its debut with
Conroe and
Merom.
Penryn enhances Wide Dynamic Execution technology with a fast radix-16 divider and improved Virtualization technology. With a fast radix-16 divider, the processor can process 4-bits per cycle instead of the 2-bits per cycle of
Conroe – doubling the divide instruction capabilities. Intel VT technology receives enhancements that reduce virtual machine transition latencies by 25-to-75%.
Intel Advanced Smart Cache technology receives additional enhancements, besides the increased L2 cache.
Penryn-based quad and dual-core processors will have up to 12MB and 6MB L2 cache, respectively. Intel reduces cache latency in addition to the larger sizes.
Penryn features a 24-way associative cache, an upgrade from
Conroe’s 16-way associative cache.
New to the Advanced Digital Media Boost technology is the inclusion of a new Intel SSE4 instruction set. SSE4 introduces 47 new instructions to improve performance of video accelerators, graphics building blocks and streaming load. Intel claims a 2x performance gain in video acceleration tasks. There are 14 new instructions for video accelerator performance enhancement. Intel improves compiler auto-vectorization performance with 32 new instructions.
Intel expects SSE4 optimizations to deliver
performance improvements in video authoring, imagine, graphics, video search, off-chip accelerators, gaming and physics applications. Also new to Advanced Digital Media Boost is the Super Shuffle Engine. Intel’s Super Shuffle Engine allows for shuffling unpacking, packing, align concatenated sources, wide shifts, insertion and extraction, and setup for horizontal arithmetic functions. Intel claims a “2x faster SSE shuffle instruction execution,” according to briefing documents.
Mobile Penryn processors receive enhanced power saving technologies. New to the mobile
Penryn is a deep power down state. In the deep power down state, the processor lowers the core voltage, more so than in the C4 state, and turns off the L1 and L2 caches. Intel claims significant power savings in idle modes for extended battery life with the new power state.
For servers and workstations, Intel has designed
Harpertown with additional headroom for the front-side bus. Intel plans to debut
Penryn-based Xeon DP, MP and UP processors with 1333 MHz front-side bus, but the architecture has headroom for up to 1600 MHz front-side bus.
Penryn-based quad and dual-core Xeons will have three thermal envelopes. Quad-core
Harpertown Xeons will have 50, 80 and 120-watt TDP ratings while dual-core
Wolfdale Xeons have 40, 65 and 80-watt ratings.
Desktop Core 2 and Core 2 Extreme processors have less thermal bins. Quad-core
Yorkfield models have 95 and 130-watt TDPs while dual-core
Wolfdale models have a single 65-watt TDP. Intel hasn’t set thermal ratings for its mobile
Penryn processors yet.
Penryn follows Intel’s plans of
alternating between new fabrication processes and a completely new core within two years. Intel previously released its 65nm fabrication process with the last of its Netburst Pentium D and Xeon DP processors, with
Conroe, a new architecture, following months after. Once again,
Penryn is a die shrink of
Conroe set to debut Intel’s 45nm fabrication process.
Following
Penryn is a
new Nehalem architecture, based on 45nm with Intel’s new QuickPath technology, formerly known as common-system interface.
Nehalem is set for a 2008 introduction,
beginning with the Xeon family for servers and workstations. Intel plans to shrink
Nehalem to 32nm with the
Westmere core. Following
Westmere is
Sandy Bridge, a new micro architecture based on 32nm.
Expect Intel to debut
Penryn later this year with the Xeon family.