New 64-bit PWRficient Multicore CPUs
Emerging from more than two years of stealth development, P.A. Semi today unveiled the PWRficientTM processor family—a 64-bit multicore, scalable processor line based on the Power ArchitectureTM from IBM—that delivers high performance at very low power consumption, offering up to a tenfold advantage in performance per watt over the industry. The first PWRficient chip, the PA6T-1682M, which dissipates between just 5-13 watts, depending upon the application, is a dual-core implementation running at 2GHz with two DDR2 memory controllers, 2MB of L2 cache, and a flexible I/O subsystem that supports eight PCI Express controllers, two 10 Gigabit Ethernet XAUI controllers, and four Gigabit Ethernet SGMII controllers sharing 24 serdes lanes. It will sample in the third calendar quarter of 2006, with single-core and quad-core versions due in early and late 2007, respectively, and an eight-core version planned for 2008.
Emerging from more than two years of stealth development, P.A. Semi today
unveiled the PWRficientTM processor family—a 64-bit multicore, scalable
processor line based on the Power ArchitectureTM from IBM—that delivers high
performance at very low power consumption, offering up to a tenfold advantage in
performance per watt over the industry. P.A. Semi is headed by Dan Dobberpuhl,
the acclaimed lead designer of the DEC Alpha series of microprocessors, the
ultra-power-efficient StrongARM microprocessors, and the first commercial
multicore processors including the SiByte 1250. The 150-strong processor, ASIC,
software and systems engineering team also includes key designers of other
defining processor architectures, such as Opteron, Itanium, and UltraSPARC.
"The next wave of microprocessor innovation is contingent on solving the problem
of dramatically increased power consumption," said cofounder, president, and CEO
Dan Dobberpuhl. "We had to start from scratch, rethinking every step, to achieve
our breakthrough performance-per-watt design. The result is a paradigm-shifting
processor that has been enthusiastically received by our customers, who look
forward to building a new breed of cool, efficient, yet high-performance,
systems around the PWRficient processor."
PWRficient HIGH PERFORMANCE AT LOW POWER
The PWRficient processors address the multibillion-dollar high-performance
embedded and computing markets to redefine power, cost, and throughput
efficiency in high-performance processing. The unique system-on-chip
architecture and design, underpinned by 50 patents filed and pending, delivers
high performance (up to 2.5GHz per-core) at phenomenally low power consumption.
In terms of performance per watt, the defining metric for all next- generation
processors, PWRficient is up to 10 times superior to the competition. For
example, the first PWRficient processor, a dual-core chip running at 2GHz,
dissipates just 5-13 watts typical, depending upon the application.
Beyond performance per watt, the PWRficient processor delivers key breakthroughs
in cost and throughput efficiency. PWRficient processors are the first
processors in their class to integrate what is typically a three- to
five-chip-set platform into a single chip, called a "platform processor." Not
only does the integration of the cores, memory, south bridge, and high-speed I/O
onto one chip dramatically reduce the cost of silicon and power consumption, but
it also delivers high throughput at low latency.
Through its unique modular architecture, which allows the number of cores,
memory controllers, cache, serdes lanes, and protocols to easily scale, P.A.
Semi will deliver a family of PWRficient processors targeting a variety of
applications, including high-performance computing, embedded datacom and
telecom, storage, and other embedded consumer applications. Additionally, this
modularity advantage enables P.A. Semi to tape out new PWRficient processors in
one quarter, versus the years common in the industry
"P.A. Semi’s PWRficient processor addresses the fundamental challenge facing all
next-generation processors by delivering higher performance and reduced power,"
says In-Stat analyst and Microprocessor Report editor in chief Kevin Krewell.
"In being phenomenally low-power while still being able to run at high clock
speeds, the PWRficient processor is ahead of today’s processors and will be a
significant challenger to the much vaunted devices on its competitors’
THE PWRficient PROCESSOR ROLLOUT
The first PWRficient chip, the PA6T-1682M, which dissipates between just 5-13
watts, depending upon the application, is a dual-core implementation running at
2GHz with two DDR2 memory controllers, 2MB of L2 cache, and a flexible I/O
subsystem that supports eight PCI Express controllers, two 10 Gigabit Ethernet
XAUI controllers, and four Gigabit Ethernet SGMII controllers sharing 24 serdes
lanes. It will sample in the third calendar quarter of 2006, with single-core
and quad-core versions due in early and late 2007, respectively, and an
eight-core version planned for 2008.
PWRficient ARCHITECTURAL ELEMENTS
The PWRficient family of platform processors is derived from a common set of
fundamental architectural elements. A coherent, ordered crossbar called
CONEXIUMTM interconnects multiple Power cores, L2 caches, memory controllers,
and the ENVOITM I/O subsystem. ENVOI combines a set of configurable serdes lanes
with a set of protocol controllers for such I/O standards as PCI Express,
Gigabit Ethernet, and 10 Gigabit Ethernet.
These controllers share a bridge to CONEXIUM, as well as a set of centralized
DMA channels, offload engines, and a coherent I/O cache. The architecture
supports a variety of offload engines, including support for TCP/IP, iSCSI,
cryptography (IPSec and SSL), and RAID. This layered, scalable architecture
results in versatile single-chip solutions that can be quickly developed by
combining the appropriate number of Power cores, memory controllers, and L2
caches with a suitable number of serdes lanes and protocol controllers.
P.A. Semi also employs a unique scalable-socket plan, which provides several
options for performance upgrades or cost reductions with little or no design
effort. P. A. Semi defines a "socket" (package, pinout, and power envelope) by
the number of memory controllers (up to four), the number of serdes I/O lanes
(up to 32), and the supported system peripherals. Each socket supports several
performance levels by varying the number of cores (up to eight on a chip) and
the size of the L2 cache (up to 8MB). Within a socket definition, processors are
tailored to different applications by adjusting the number and type of the
high-speed I/O protocols (for example PCI Express, 10 Gigabit Ethernet, 1
Gigabit Ethernet, SATA/SAS, RapidIO, and Fibre Channel). Initial socket
definitions include the "E" socket (entry), "M" socket (midrange), and "P"
socket (performance). Customers can design to a specific socket, instead of a
specific processor, to enable easy migration to compatible processors.
P.A. SEMI STRATEGIC PARTNERS AND ECOSYSTEM
P.A. Semi is partnered with some of the most notable names in technology, having
licensed the Power Architecture from IBM [see other news release, P.A. Semi
Signs Power Architecture License], and is supported by an ecosystem of partners,
including Macraigor Systems, Micron, MontaVista Software, QNX Software Systems,
SMART Modular Technologies, Inc., Terra Soft Solutions, and Wind River.
P.A. Semi is backed by two of the most respected venture-capital firms, Bessemer
Venture Partners and Venrock Associates.
FALL PROCESSOR FORUM PRESENTATION
Jim Keller, vice president of engineering, Architecture Group, P.A. Semi, will
present the PWRficient processor architecture at Fall Processor Forum in San
Jose, California, on Tuesday, October 25 at 9:50 a.m. in a session titled "A
Power-Efficient, Scalable Processor Family." The new PWRficient processor will
also be demonstrated at the show on an EVE (Emulation and Verification
Engineering) ZeBu-XL emulation platform.
More information on P.A. Semi, its founders, the PWRficient processor family,
and its partners appears on the new P.A. Semi web site, also launched today, at