Maximum PC transistors RSS Feed

First Programmable Quantum Computer Created

Ryan Whitwam — Wed, 18 Nov 2009 19:27:22 -0600

Moore’s Law states that approximately every two years, the number of transistors that can be placed on an integrated circuit doubles. This has held true for the last 50 years. But there will come a point one day when physics puts a stop to that. Eventually the boundaries of atomic scale will limit transistor density. However, a new breakthrough in the field of quantum computing may provide hope for future advances. Until now, a quantum computing device had to be designed for one, and only one, operation. But scientists from the National Institute of Standards and Technology (NIST) have constructed the first programmable quantum processor.

Quantum processing units are fundamentally different in a number of ways. First, where a regular bit can be only 1 or 0, a quantum bit (or qubit) only assumes a value of 1 or 0 when it is observed. Additionally, Quantum computers aren’t bound by Boolean operators like ‘and, ‘or’ and ‘not’. Finally, two qubits can be “entangled”, meaning they will always have the same value when observed, even if separated.

The NIST computer consists of two quantum gates, one single qubit gate and an entangled two qubit gate. The gates utilized two beryllium ions stimulated with UV lasers to represent operations. The test programs run came back with 79% correct results. Certainly not perfect, but a huge step forward. You won’t be dropping one of these into a socket on your motherboard anytime soon, but maybe someday.

Hard Case: Too Many Transistors!

Loyd Case — Thu, 10 Sep 2009 12:00:00 -0500

Francois Piednoel is worried.

For those of you who have never met Francois, he’s a member of the performance marketing team at Intel. It’s always entertaining to carry on a conversation with Francois. He was the guy at Intel who first steered me to the idea of building small systems around an X58 micro ATX motherboard and undervolting the CPU while maintaining the reference clock speed. This is sort of the inverse of overclocking, and results in pretty high performance systems that run cooler and quieter than the norm.

What worries Piednoel, though, is this: what are desktop users ever going to do with six cores?

By the end of this year, Intel will be rolling out CPUs built on their next generation 32nm manufacturing process. This will enable them to ship a CPU code-named Gulftown. Gulftown is 32nm CPU with six cores, and will drop into socket 1366 motherboards built on the current X58 chipset.

Six cores sound great, right? After all, more is always better. There are a small subset of users who will benefit from six cores. Folks who do a lot of video editing or professional photo editing will benefit. Some other workstation class applications which are multicore aware will also benefit. Performance enthusiasts and gamers who buy Gulftown and drop it into their X58 systems will discover that two or more cores are actually idle most of the time.

Right now, I’m typing this on a four core Nehalem (Core i7) system. According to a nifty tool called TMonitor, I’ve got one core relatively active, because it’s playing back music. (TMonitor can be downloaded from the CPUID web site). Another core occasionally blips as I type this. The other two cores (and four logical cores, since Hyper-threading is enabled) is quiescent. If I fire up a game, a couple of cores are hammered, and the others are mostly idle. Photoshop will sometimes hammer two or three cores, but only for a few seconds.

So what will most of us actually do with a six core CPU, other than brag about it?

One possible future is heterogeneous cores. These are CPUs that may have two to four general purpose cores, and several specialized cores. For example, you could imaging an entry level or even midrange graphics processor built onto a CPU core. In fact, AMD is pursuing this with their Fusion strategy. Intel’s 32nm Arandale mobile CPU will have the graphic integrated onto the chip package, and it’s not much of as stretch to see that the next step is to bring that graphics functionality onto the CPU die.

At what point, though, is a core not a core? Take Larrabee, for example. Larrabee is Intel’s foray into the world of graphics processors. One key component of Larrabee will be a hefty set of vector instructions in the form of a vector-processing unit to assist with graphics processing. If those vector extensions are built onto a general purpose CPU in the future, would you consider that to be a separate core?

Still, adding more capability in the form of new instructions can make effective use of a large transistor budget. What I’m driving at here is that the CPU companies know two things:

• Their transistor budget is going up.
• Adding more general purpose cores is pointless for most desktop users.

So they’re trying to be smart about what to do with all those new transistors, but they're moving into uncharted territory. It will be interesting to watch developments.

Of course, the other approach is to shrink the microprocessor. That’s what Intel is doing with Atom. In fact, Intel is both shriking and adding new capability as the manufacturing process allows for more circuit density. The problem is that the profit margin on these tiny, low power CPUs is thinner than on the more capable CPUs – but they’re still capable CPUs. So we see artificial limits, like reduced marketing dollars for companies that try to build Atom-based systems that look more like traditional laptops.

In the end, Moore’s Law still has a few tricks left. The problem is that our ability to take advantage of more transistors is increasing at a slower rate than Moore's Law adds new transistors. Someone, somewhere, will have to think of new ways to use all those transistors.

Moore's Law to Hit a Brick Wall at 18nm

Paul Lilly — Tue, 16 Jun 2009 16:00:10 -0500

Intel co-founder Gordon Moore once predicted that the number of transistors on an integrated circuit would double every 18 to 24 months, a prediction which has been famously dubbed Moore's Law. But according to market research firm iSuppli, the move to 18nm will signal the end of Moore's Law.

"The usable limit for semiconductor process technology will be reached when chip process geometries shrink to be smaller than 20nm, to 18nm nodes," said Len Jelinek, director and chief analyst, semiconductor manufacturing, for iSuppli. "At those nodes, the industry will start getting to the point where semiconductor manufacturing tools are too expensive to depreciate with volume production, i.e., their costs will be so high, that the value of their lifetime productivity can never justify it."

So when exactly will it happen? According to iSupply, in the year 2014. In 2007, Gordon Moore said his prediction could be upheld for at least another decade. Five years from now, one of them is going to be wrong.