|
Embedded
Processor Watch
MicroDesign
Resources --- October 17, 2000 #120
Editor:
Cary D. Snyder
Contributors
to this issue: Steve Leibson, Cary Snyder, and Kevin Krewell
In This
Issue:
- SH-Zam!
SH Map Leads to SH-7
- Embedded
Processor World War
- Sun
Shines on UltraSPARC III
SH-Zam!
SH Map Leads to SH-7
Hitachi,
ST Reveal Additional SH-5 Details
By Steve
Leibson {10/16/00-02}
In the
comics, Billy Batson transformed himself into Captain Marvel
by uttering the word Shazam! Hitachi and STMicroelectronics
appear to have used that particular spell to transform the
32-bit SuperH processor into a 64-bit multimedia machine called
the SH-5. The corporate duo first disclosed plans for the
SH-5 at last year's Microprocessor Forum. The scalar, single-issue
SH-5 executes all 208 instructions of its 32-bit SuperH predecessors.
These existing instructions are actually 16 bits long and
are now labeled the SHcompact instruction set. The SH-5 also
executes 209 additional instructions, called the SHmedia set,
which are 32-bits long and include SIMD and floating-point
instructions. The LSB of the SH5's program counter serves
as an instruction-set mode selector.
At this
year's Microprocessor Forum, ST and Hitachi provided additional
technical details on the SH-5 core design and a roadmap that
extends the SuperH series an additional two generations, to
the SH-6 and SH-7. The extra technical detail provides a deeper
understanding of the SH-5 pipeline and virtual cache architecture.
The seventh-generation SuperH processor takes the processor
family well into the gigahertz realm. The processor core is
taping out now. SH-5 partners Hitachi and STMicroelectronics
are not yet saying when we can expect the SH-6 and SH-7. (The
full version of this article is available online to Microprocessor
Report subscribers at http://www.mdronline.com/mpr/h/2000/1016/144202.html
)
Embedded
Processor World War
Action
Heating Up on Multiple Fronts
By Cary
D. Snyder {10/16/00-01}
The opening
salvo of Altera's Excalibur embedded-processor battle is not
swordplay but a bold declaration of war. The expected end-of-year
barrage will be a very capable $2,000+ (100-piece quantity)
EPXA10 FPGA embedded-processor hybrid device shipping by the
end of the year. The first Excalibur FPGA embedded-processor
device is an ARM9-based embedded-processor stripe crafted
to a million-gate Altera APEX 20KE1000E FPGA. It will be closely
followed by a similar MIPS32 4Kc-core-based million-gate FPGA.
Smaller, less costly devices of both processor camps will
ship in 1H01.
Altera
is likely to gain a significant lead over Xilinx in offering
serious hard and soft FPGA embedded-processor hybrids. Xilinx's
own PowerPC embedded-processor architecture has yet to debut
in an actual product announcement that contains details on
what IBM PowerPC core will be used or any other product details
(see MRP 9/18/00-01, "FPGA Processor Cores Get Serious").
Altera's own PowerPC product announcement and details of a
licensing agreement with Motorola are slow in coming, a hint
that licensing issues between Motorola and IBM are complicating
related agreements.
In the
meantime, PowerPC may lose some interesting opportunities.
Altera's Excalibur family creates interesting battlegrounds
for easy performance evaluation of MIPS and ARM processor
architectures with similar features in pin-compatible FPGA
embedded-processor hybrid chips and other soft-core applications.
Another important related area will be the emergence of the
favorite on-chip bus: will it be the open ARM AMBA bus, IBM's
CoreConnect, or something else?
All the
action in the serious FPGA embedded-processor arena that is
hosting soft cores, on-chip buses, and hybrid devices is quickly
heating up battles in the embedded-processor market. Today's
battles for whatever becomes the favored SOC design methodology
or process will push the greater innovations of tomorrow.
The embedded-processor industry is showing signs that everyone
in these hard-fought battles can be a winner by making a profit,
like a golf tournament in which everyone finishes in the money.
(The full version of this article is available online to Microprocessor
Report subscribers at http://www.mdronline.com/mpr/h/2000/1016/144201.html
)
Sun
Shines on UltraSPARC III
By Kevin
Krewell {10/16/00-04}
Sun Microsystems
has finally released the first of a series of UltraSPARC III-based
products. It is now three years since Sun first revealed details
of the UltraSPARC III at Microprocessor Forum '97 and about
a year later than UltraSPARC III-based systems were expected
to ship. In those three years, Sun raised the top target frequency
at launch from 600MHz to 900MHz (which will be available in
4Q00). Sun also changed the main memory interface bus, widening
it from 128 bits to 256 bits of data plus 32 bits of ECC.
In the
first of three stages of UltraSPARC III-based product releases,
Sun announced the Sun Blade 1000 workstations and the Sun
Fire 280R small server. These announcements included only
single- and dual-processor (SMP) configurations; large server
configurations will be announced in 1H01. Sun announced shipments
of 600MHz and 750MHz Sun Blade and Sun Fire products, with
900MHz products to follow in 4Q00 when Sun's fab partner,
Texas Instruments, converts the 0.18-micron process (used
to fabricate the UltraSPARC III processor) from aluminum interconnects
to copper. In the aluminum process, the UltraSPARC III die
measures 15.5mm by 15.9mm (246mm2), and in the copper process,
the die size shrinks to 14.3mm by 14.7mm (210mm2). The 900MHz-based
Sun Blade 1000 workstation has posted very competitive SPEC
numbers of 438 SPECint2000 (base) and 427 SPECfp2000 (base),
a major leap forward from the anemic performance of the aging
UltraSPARC II family. Sun also revealed an outline of its
processor roadmap, with UltraSPARC III processors scaling
up to 1.5GHz; followed by the UltraSPARC IV processor, starting
at 1.8GHz in late 2002; and leading to UltraSPARC V in late
2003, running at 2.1GHz. (The full version of this article
is available online to Microprocessor
Report subscribers at http://www.mdronline.com/mpr/h/2000/1016/144204.html
)
|