Apple Drops PowerPC for Pentium
Kevin Krewell - Editor in Chief {06/20/2005}
On Monday, June 6, 2005, Steve Jobs took the stage in the keynote at the Apple Computer Worldwide Developer Conference 2005 (WWDC2005) and announced that Apple would transition from the PowerPC processor and move to Intel processors, starting in 2006. The first Intel-based Mac computers (which Jobs called "Mac-ntels") should be shipping by next year¹s WWDC, and the transition would be complete by 2007.
Apple realizes that not every PowerPC application will be recompiled for the Intel processors. For these programs, Apple announced "Rosetta" software that will translate PPC code to Intel code. The dynamic binary translation software will run transparent to the Mac user, as it is tightly integrated into the Mac OS. However, Rosetta has a number of limitations, the biggest being that it will not support AltiVec instructions. Any emulated application will be emulated in a G3 mode. Rosetta technology is based on technology from a startup called Transitive Technology.
To support the transition with third-party developers, Apple will lease an Intel 3.6GHz P4-based development platform for $999 for registered Apple developers. The systems must be returned to Apple in 2006.
This transition will also mean the end of Apple support for OS 9 and earlier operating systems. The document on the Apple developer site on the software transition makes it clear that Rosetta will not support the older Apple operating systems. The "classic Mac" support will not be available on Intel-based Macs. Along with the end of OS 9, it will also mean the end of 68000 emulation as well.
The risks for Apple are great, but two years from now, the rewards may be worth it. As a measure of that, a number of PC users MPR talked to would consider buying an Intel-based Apple computer, assuming they could get the best of both worlds‹Windows and Mac OS all on one, high-performing, attractive computer. For some traditional Apple customers, the problem is that Apple is not thinking quite as different as it once was.
Elixent Improves D-Fabrix
Tom R. Halfhill - Senior Editor {06/20/2005}
Elixent took the stage at Spring Processor Forum 2005 to prove that listening to customers isn't a lost art. Using feedback from early adopters of its massively parallel configurable-processor core, Elixent has introduced D-Fabrix v2.0, which significantly boosts performance without increasing the overall gate count.
D-Fabrix embodies a concept that Elixent calls reconfigurable algorithm processing (RAP). It's an outgrowth of the Chess architecture described in an academic paper presented in 1999 at the ACM/SIGDA International Symposium on FPGAs. One of the paper's authors, Alan Marshall, is the chief technology officer of Elixent, a five-year-old U.K.-based spinoff from Hewlett-Packard Labs in Bristol, England. Marshall also presented D-Fabrix v2.0 at SPF.
The general concept of RAP is to build a chessboard-like array of four-bit ALUs, multiplexers, registers, local memories, and interconnecting switchboxes that work together as a massively parallel on-chip fabric. Using Elixent's proprietary tools, SoC developers can configure the fabric at design time to optimize it for specific algorithms. The D-Fabrix architecture lends itself to applications with inherent data parallelism. (See MPR 2/9/04, "Extreme CPUs Defy Conventions," and MPR 7/21/03, "Elixent Expands SoCs.")
On the basis of user feedback and internal analysis, Elixent found that many of the four-bit ALUs in the fabric were being used as simple multiplexers instead of as compute resources in the critical datapath. Elixent also found that programmers were writing more bit-manipulation code than expected and that the architecture could be improved for those operations. As a result, Elixent has introduced D-Fabrix v2.0, which addresses those issues.
Editorial: Best of Spring Processor Forum '05
Kevin Krewell - Editor in Chief {06/27/2005}
We recently completed our first Spring Processor Forum (successor
to the Embedded Processor Forum) at the Doubletree Hotel in San
Jose, California. The new location was quite a hit. We packed 31
presentations into two full days, perhaps a bit too full. We tried an experiment by running two sessions in parallel to form two
tracks. Unfortunately, a number of people wanted to attend both
sessions and couldn't. We'll fix that at the Fall Forum.
As I mentioned in my previous editorial, the keynote by Wally
Rhines, CEO of Mentor Graphics, followed up on my earlier editorial
on Moore's law, and it was quite a hit with many audience members.
On the second day of the Forum, we had a special presentation on
the Cell processor by Jim Kahle of IBM. Kahle made a special effort
by flying up from the PlayStation 3 launch event at the E3 trade
in Los Angeles to show how the Cell processor can be used and announced
IBM's plan to open the architecture by releasing an open-source
tool chain and simulator. The Cell processor continues to excite
the engineering community with its unique approach to computational
performance, and Kahle's presentation received the most positive
reviews of the Forum. Based on audience feedback, Kahle's Cell presentation
was "best of show."
The ongoing theme that touches almost all processor designs today
is multicore processing. Part of the appeal of the Cell processor
is the approach the team from IBM, Sony, and Toshiba has taken to
aggressive multiprocessing.
We also got to see the first public demonstration of the XLR processor
from Raza Microelectronics and the revelation of the first superscalar,
out-of-order ARM core from Marvell. And as I mentioned in my previous
editorial, multicore design will be the theme for our Fall Processor
Forum, and we're planning to move beyond chip and core introductions
to explore all aspects of multicore design, including software.
Apple Redux
I have to eat some crow after my last editorial, as it was revealed
at Apple's developer conference that the company will indeed
move to Intel processors. Complete in-depth coverage appears in
this issue. (See MPR 6/27/05-01,
"Apple Drops PowerPC for Pentium.") The move is a calculated risk
for Apple and a setback for the PowerPC. The winner in this is clearly
Intel, but the short-term impact on IBM and Freescale is minor.
Intel gains an additional customer it didn't have before and locks
up the personal computer business. Intel also gains a relationship
with a company that can challenge Microsoft's stranglehold on the
desktop as well as a partner with good connections to media companies
and for developing digital rights management for PCs.
For IBM, Apple's move is an embarrassment, and that company loses
a partner that helped advance the PowerPC architecture (not to mention,
Apple was the original impetus for creating the PowerPC). But IBM
still has the opportunity to sell perhaps one hundred times the
potential Apple volume with the game consoles from Microsoft, Nintendo,
and Sony. Freescale had already focused its PowerPC processors on
the embedded market a few years ago.
The mistake I made was in thinking about the Apple/PowerPC relationship
as an engineering problem when it was actually a purely business
decision. There were no technical barriers to developing the PowerPC
processor for Apple's needs, just economic barriers. In the end,
the business was not valuable enough for IBM to expend its limited
resources to maintain it.
Lessons From Formula 1 Racing
I had the unique opportunity to attend the U.S. Grand Prix in
Indianapolis on June 19, 2005, as a guest of AMD. The race turned
out to be not much of a race, but it did provide a good lesson in
supply-chain and vendor selection.
Only two companies supplied tires for the all the Formula 1 (F1)
race cars: Bridgestone and Michelin. Most of the teams (seven out
of ten) in the race were running on Michelin tires and had done
well with the tires up to this race. In the first practice run on
Friday, the Toyota team experienced two tire failures, one of which
led to a spectacular crash into a wall for driver Rolf Schumacher.
(He was shaken up, but not injured.) The reason for the crash was
found to be failure of the Michelin tires in a particularly high
g-force turn.
This accident was a serious failure for Michelin's race program
and a problem not just for Toyota but for every team running on
Michelin tires. For each racetrack, the race teams and tire companies
configure the cars and tires for the desired results. Each team
spends millions of dollars on computer simulations and wind-tunnel
tests. Even with this extensive development program, the cars and
tires are so finely tuned that it's nearly impossible to make a
significant change a couple of days before the race. While Bridgestone
engineers analyzed the failure, the Michelin-equipped teams tried
to convince the Federation Internationale de l'Automobile
or FIA (the race-sanctioning body) to find a way to let them run.
Even after extensive negotiations and a request for a chicane (road
barrier) to slow the cars in one critical turn where the Michelin
tires failed, in the end they could not find a solution that would
be safe and that met FIA rules. The end result was that after the
first parade lap of the race, all the cars with Michelin tires retired
to the pits and quit the race. The remaining six cars, from the
three teams that were running the Bridgestone tires, continued to
race, leading to a first- and second-place finish for the Ferrari
team.
The teams running on Michelin tires lost a valuable opportunity
to score points toward the season championship, and they wasted
all the time and money preparing for the race. The fans lost most
of all: many had traveled from all over the world to see their favorite
drivers and teams compete, and most were sadly disappointed. Michelin
may have also lost the confidence of the drivers and teams. Michelin
was considered the faster tire, but now the teams will have to consider
how much they can trust a critical component on their car. This
is a serious challenge for Michelin: to find the reason the tires
were not properly designed and constructed for the fast Indy track
and to regain the confidence of the teams and drivers. It is very
difficult, if not contractually impossible, for the teams to switch
tire suppliers midseason, much less just before a race. The F1 racing
sport has become so complex that relying on multiple suppliers is
not an option.
In just three days, Michelin went from being the dominant supplier
to a position in which it must go into damage control to save its
reputation and the F1 business. Yesterday's hero can be tomorrow's
goat. Picking the correct supplier for Ferrari meant standing on
the winner's podium while the other teams were packing up and catching
the next plane out of Indianapolis. So sometimes the fastest doesn't
win the race.
The other instructional lesson is that computer simulations are
not yet good enough to completely model the real world. Even simulations
run on supercomputers are still not the real thing.
The Passing of a Legend
It is with considerable sadness that we heard that one of the
true founders of the semiconductor industry, Jack S. Kilby, passed
away on June 20, 2005 at age 81. In 1958, while at Texas Instruments,
Kilby created the first integrated circuit design. Kilby received
the Nobel Prize in Physics in 2000 for his work on the integrated
circuit and had more than 60 patents over his career. Kilby's work,
and the work of Robert Noyce at Fairchild, laid the foundation for
the modern electronics business.
To find out more about Microprocessor Report, please visit: www.mdronline.com
