Ask About Computers

Keeping up with Technology (so you don't have to)

Tid-bits

logo

AMD LogoAMD 2006 Analyst Day

Core 2 ProcessorCore 2 Processor Family

Core MicroarchitectureCore Microarchitecture

front and backThe Full Monte-cito: Itanium 2

AMD / ATI MergerAMD's Acquisition of ATI

Special Interest

J.R.R. Tolkien and The Lord of the Rings

Energy Efficient Performance

January 21, 2007

Intel CEO and PresidentIntel CEO and President Keynote

There's an enormous emphasis on power efficiency in computing these days. At the same time, almost everyone still wants performance.

The current emphasis on power efficiency represents, per Intel, "the biggest single shift in computer microarchitecture we've seen in well over a decade" (p. 2).

Performance today is largely being driven by the shrinking size of transistors, which enables chipmakers to double the number of cores on a CPU.

Many applications are able to make good use of the additional cores. "Performance matters again after four or five years" (p. 1).

Apple

Apple Logo

2006 marked the first time that an executive from Apple appeared onstage at an Intel Developer Forum.

However, this should come as no surprise, since Paul Otellini, Intel's president and CEO, had already appeared at Apple's Worldwide Developers Conference.

The Intel CEO announced that Apple was moving to Intel hardware. The Apple executive that the transition to Intel had been completed.

Since the transition began in January ('06), Apple was able to double the market share of its notebooks in the space of about six months: "market share of notebooks had risen from 6 percent in US retail in January to 12 percent this summer" (p. 3).

Apple moved their notebooks to Core Duo and later Core 2 Duo. In their professional workstations, however, they put Xeon server processors based on the Core microarchitecture.

Though Apple completed the transition to Intel hardware, this is also "just the beginning" (p. 4).

Core 2 Processor Family

Significant bumps in clock speed used to result in comparatively little performance increase. It was "an era where 5 percent or 10 percent increments in performance seem normal" (p. 4).

Performance did not scale linearly with clock speed because of other bottlenecks in the system, such as memory and the hard disk subsystem. These bottlenecks still exist.

However, some applications are heavily bound by the CPU. It is these applications that tend to show the most improvement in doubling the number of cores on a CPU.

Core 2 Duo and Core 2 Extreme LogosThe Core 2 family now includes Core 2 Quad, not pictured

Possibly because of its performance advantages, Core 2 Duo has been the fastest ramping product in Intel's history. "It's the fastest 60-day ramp for both the desktop and notebook processors in Intel's history" (p. 4).

Core 2 Extreme is a brand. Under the hood of the quad-core version, there's basically a couple of Core 2 processors.

By putting together dual-core processors in this way, Intel was able to double the number of cores within a short period of time and took "a huge step forward for the second time inside of a four-month window" (p. 5).

Process Technologies

Clean Room Worker with Product

Chipmakers recently ran into a wall where they could no longer bump up the speed of processors at the same rate of the past.

A next barrier shall occur when chipmakers can no longer shrink the size of transistors at the same rate of the past.

Chipmakers, however, should know in advance when they are about to hit this barrier of shrinking transistor size, since chipmakers start working on new process technologies about five years in advance. Microprocessor manufacturers "are now working on processes for use five years or more from now" (p. 7).

Current state-of-the-art process technology is 65nm (nanometers).

Up next is 45nm.

SRAM (static RAM) has already been produced on 45nm; test wafers are currently being produced; and processors based on 45nm should appear in 2007.

Then comes 32nm.

The following is a good illustration of just how small the current 65nm process technology is: "up to 10 million transistors can now fit on the tip of your ballpoint pen" (p. 8).

The reason why Intel has about $9 billion invested in 45nm fabs is because they have three fabs. Each fab costs about $3 billion. So 3 x $3 billion equals $9 billion.

Two of the fabs are still under construction. The one currently producing test wafers is the "first and only 45-nanometer production facility" (p. 8).

Microarchitectures

Diagram of Microarchitectures

It has taken Intel time to adjust after having pushed clock speeds as high as they would go under the NetBurst microarchitecture. However, the chipmaker says that it now has its rhythm down. "We have a model now for sustained technology leadership" (p. 9).

Intel has for some time been shrinking the size of its transistors about once every two years. From 130nm to 90nm, from 90nm to 65nm, from 65nm to 45nm, to 32nm, and so on.

However, now Intel intends to introduce a new microarchitecture every two years as well. Before, Intel changed microarchitectures about once every five or six years.

The current microarchitecture is Core, Nehalem is scheduled for 2008, Gesher 2010.

The Data Center

Intel CEO and president holds up a wafer of teraflop processorsA wafer featuring research prototype chips with 80 floating point cores

Intel hopes to address the needs of the data center with a teraflop chip. That's a chip capable of executing a trillion floating point operations a second.

The chip is only a research prototype. It's still "the world's first teraflop on a chip die" (p. 9).

Energy efficiency is so important that data centers are being built in locations where energy is the cheapest: "it's one of the reasons why you've seen some of the three largest data center operators in the world, Yahoo, Google, and Microsoft, all announce that their next-generation data centers are going to be built in the Pacific Northwest, because that's the cheapest electricity available in the United States" (p. 9).

Xeon 5100 Family

Xeon 5100 seriesXeon 5100 series

Processors based on the Core microarchitecture power devices ranging from the smallest thin and light computers to dual-processor (DP) servers and workstations.

The brand name of the mobile and desktop implementation of the Core microarchitecture is Core 2 Duo. However, the name of the server version of the Core microarchitecture is still Xeon. Xeon 5100 is the dual-core series, Xeon 5300 the quad-core series.

The dual-core series--the Xeon 5100--has had, per Intel, "the fastest ramp of any server chip in the history of our corporation" (p. 10).

Core Processor Challenge

Intel is offering $1 million to whoever can design the most stylish PC.

The primary objective behind the Core 2 Challenge is to stimulate the desktop form factor.

Content is changing. However, "the desktop is not evolving nearly as rapidly" (p. 14).

Laptop Growth

diagram of laptop growth

The notebook space continues to explode: "notebook growth is accelerating" (p. 14).

It's not that the market share of notebooks is spectacular. Rather, it's the market share of notebooks compared to just a few years ago.

In the United States and Europe, more consumer notebooks are sold than consumer desktops. "This is up from less than 20 percent of the total market in the pre-Centrino days" (p. 15).

Next Generation Centrino

Next Generation Centrino SlideSource: Keynote Presentation, p. 66

The new platform for Centrino debuts early 2007, codename Santa Rosa.

Robson Technology

Santa Rosa is to contain Robson Technology, which is NAND on the motherboard. The NAND performs many of the functions of the hard drive, only faster and with less energy.

It should be Centrino notebooks that "have NAND on the motherboard for the first time" (p. 15).

802.11n

I guess you don't need B if you have G. Most notebooks ship with wireless 802.11ag silicon, rather than 802.11abg. 802.11g is backwards compatible with 802.11b networks, so apparently there's no need for 802.11b if you have 802.11g.

Be that as it may, the next generation of WiFi shall be 802.11n.

illustration of enhanced range of 802.11n802.11n offers up to five times the performance and up to twice the range compared to the earlier 802.11g standard

WiMAX

Diagram of how WiMAX works

A technology to watch is WiMAX, which is wireless broadband that does not require an access point nearby the way 802.11 WiFi wireless requires an access point only a few hundred feet away.

One way that WiMAX differs from current cellular network wireless solutions is that it's supposed to be faster.

There are two version of WiMAX: fixed and mobile. 802.16e is the mobile version. By 2008, the U.S. should have "almost nationwide coverage" (p. 16).

Intel hopes to drive the adoption of WiMAX the same way it drove the adoption of 802.11 WiFi wireless, viz. with the Centrino platform.

It "was integration into the platform of Wi-Fi that drove the ubiquity of Wi-Fi. We believe the same thing is going to happen as we integrate WiMAX into the platform" (p. 16).

Ultra Mobile PC (UMPC) Silicon

While there has been a Mobility Group at Intel for some time, the Ultra Mobile Group is new.

Intel is a supporter of the Ultra Mobile PC (UMPC). However, Intel primarily makes processors, not computers, so they shall make a new processor for this new PC.

Compared to their current ultra-low-voltage (ULV) CPU, UMPC silicon will consume "one-half the power and takes one-fourth the surface space" (p. 16).

Sooner Than You Think

Third party prototypes already exist that have built-in WiFi and WiMAX chips.

NAND hard drives will be upon us before you know it, too.

In addition to being faster and more energy efficient, NAND hard drives make a device more durable because there are fewer parts to break.

The Classmate PC is expected in the first quarter of 2007 and will have "a NAND hard drive for durability" (p. 21).

Classmate PCClassmate PC (Source: Keynote Presentation, p. 76)

Send Comment