With the Intel Developer Forum behind it, Intel reported
third-quarter revenue of $9.4bn, beating analysts' expectations.
The company reported operating income of $2.6bn, net income of
$1.9bn and earnings per share of 33 cents.
"As we look ahead, Intel's game-changing 32nm process technology
will usher in another wave of innovation from new, powerful Intel
Xeon server platforms to high-performance Intel Core processors to
low-power Intel Atom processors," says Paul Otellini, Intel
president and CEO. Computer Weekly asked other suppliers to
gaze into their crystal balls.
GPU v CPU
For so long, the CPU has been king for powering everything from
simple bat-and-ball games on early PCs through to crunching through
huge quantities of records in databases. Though the death knell was
sounded in the 1970s for the von Neuman model, it has still largely
persisted, to the extent that manufacturers stitch CPU cores
together to make them look like a single central processing
facility.
Neveretheless, graphics has been found to be very greedy for CPU
cycles and separate GPUs (graphics processing units) have become
commonplace, initially as separate slot-in cards and lately bundled
on most motherboards, although basic integrated graphics are not
coping with the visual requirements of modern operating systems and
hardware. However, in the first quarter of 2010, Westmere, Intel's
collective term for its new 32nm server processors, will integrate
the graphics core previously implemented on a separate chip set
into the same CPU package. A year on, and the graphics will be
implemented on the same die.
The purpose of a GPU is to effect as many pixels (for HD
television) or polygons (for games) as possible. "A GPU is suited
to massively parallel processing tasks. The more things you can
throw at a graphics chip to do at once, the happier it is," says
Benjamin Berraondo, product PR manager at GPU supplier nVidia.
"Over five years ago, Moore's Law started to hit a wall when
ever-faster CPUs were starting to overheat, so first dual-core
processors were developed, then quad-core.
"GPUs were going multicore since their inception, and 240-core
GPUs are currently available using 256-bit cores to provide
extremely high memory throughput. A GPU can be used for more than
just graphics, though, and can be programmed for any parallel
processing tasks like financial modelling, science and
engineering."
For instance, BNP Paribas Corporate and Investment Banking (CIB)
says it has boosted the capabilities of a supercomputer that
simulates the behaviour of financial markets for the bank's Global
Equities and Commodity Derivatives (GECD) group by a factor of 15
while lowering power consumption by a factor of 190.
About one teraflop (one thousand billion calculations per
second), has been transferred to a GPU-based platform, providing a
100-fold increase in the amount of calculation achieved per Watt.
The new platform is based on two nVidia Tesla S1070 GPUs consuming
2kW and will replace more than 500 traditional CPU cores consuming
25kW.
"We are extremely pleased with this performance, which
significantly exceeds our initial expectations. We hope to transfer
more calculations to the GPU architecture in the near future" said
Stéphane Tyc, Head of GECD quantitative research, BNP Paribas
CIB.
Circumventing system crashes
Among IT departments and end-users, little has done more damage
to Microsoft's reputation than the "blue screen of death". In many
instances, the only option is to reformat the disc at low level and
reinstall Windows. Having their hardware incapacitated by software
has long infuriated users. Now it looks like a solution has been
found, not by Microsoft in software, but by Intel in hardware.
In a similar way that allows a user to get access to a dead PC's
hardware through its BIOS, Intel's vPro technology allows IT
departments to reinstall Windows, or to install new patches, over
an IP network without having to send engineers out on the road. It
provides OS-absent manageability and down-the-wire security, even
when the desktop or notebook is switched off, the OS is
unresponsive, or software agents are disabled.
Steve Shakespeare, Intel director of EMEA enterprise solutions,
says, "Imagine vPro as a computer inside a computer. As long as the
machine has power and a network connection, it can be fixed
remotely from a management console within the IT department. You
can re-flash the BIOS, put new firmware into it, and then restart
it.
"vPro has already been introduced in notebooks and will also be
included in the Core i7 and i5 processors which will launch early
next year. They will use the same 45nm technology currently
deployed in the server space."
London Underground operator Tube Lines' desktop and notebook PC
infrastructure was reaching its end of life, with most of the
hardware almost six years old. Together with Intel and its IT
services provider Capgemini, Tube Lines found HP hardware powered
by the Intel Core 2 processor with vPro technology delivered the
best overall performance.
Adrian Davey, head of IT at Tube Lines, says, "By working in
close partnership with Intel and Capgemini we have received a
phenomenal level of service. The dynamic desktop management enabled
by Intel vPro technology is having an extremely positive impact on
both our corporate green philosophy and our bottom line."
SSD v Braidwood
A report by Objective Analysis found that Intel's Braidwood
technology offers comparable performance and power consumption
improvements to solid state drives (SSDs), but at considerably
lower costs.
However Intel's Shakespeare was strangely dismissive about
Braidwood, saying Intel had no new announcements on it, or when it
would be productised. "If you have a vPro-based machine and looking
for accelerated disc support, SSD can help with that. I have a
Centrino-based machine with an SSD hard drive replacement, and I am
delighted with it."
London Borough of Hillingdon has been using a Compellent storage
area network (San) for the past six months but recently updated it
to include two SSD drives. "We were looking for a new storage
solution that could automatically manage data and drive down the
cost per terabyte of stored data over the life of the San. It had
to provide us with affordable system resilience and also contribute
to a greener IT infrastructure," says Roger Bearpark, assistant
head of ICT at Hillingdon.
Hillingdon faced a problem common among many organisations.
Although the borough was storing terabytes of data, only 10% could
be considered mission-critical at any time. However, the other 90%
could not be moved to offline storage as it could be required at
short notice to form the basis of national government reports or to
help solve problems with legacy projects.
The previous San could not automatically distinguish between
primary and secondary data. This meant Hillingdon's IT team had to
either spend time manually moving data or leave secondary data on
expensive, high-speed discs
Bearpark says, "We have no directly attributable saving, as we
introduced the initial SSD as an additional tier to supplement FC
and SATA. What we will do is calculate the potential
environmental/financial benefit of running SSD as a replacement for
some of the FC disc and then look at further deployment in FY
2010/11.
"I am sure that certainly in portable and in some desktop
devices Braidwood will be more attractive than SSD but in the
storage arena it is not in the same league."
Gigabit to the desktop
In early October, the first England football match was streamed
live over the internet instead of traditional television broadcast
following the collapse of pay-TV operator Setanta. Bundling
IP-based quality of service (QoS) functionality within gigabit
switches is becoming increasingly common now that proprietary QoS
implementations from the likes of Microsoft and Cisco have been
superseded by a single converged QoS standard.
"The QoS standard is now ratified by the Internet Engineering
Task Force," said Nigel Moulton, vice-president of business
solutions at D-Link Europe. "QoS is now a function that you can
turn on in the network to effectively prioritise certain traffic
types over others. QoS will typically be implemented in an
application-specific integrated circuit and configured in software
by the user.
"For example, QoS is a requirement with voice traffic if you
want low latency in order to eliminate jitter. If you want
multi-way audio conferencing, QoS may be a requirement to guarantee
voice quality, as well as video multicasting, where there is a
requirement to split the audio stream from the video stream, and
make sure that it all arrives at the destination at the right
time.
"You have to be able to reassemble the audio codec, so QoS in
the network gets over this problem by and large. Multicasting
requires replication of a video stream, and that can be the
responsibility of the network provider or the integrator who has
designed the network for the customer."