The End of Moore's Law

By Maxxwell A.R. Chatsko - May 30, 2012 | Tickers: AMD, INTC, NVDA | 0 Comments

• Email
• Print

Most people are familiar with Moore’s Law, which states that computer chips double their capacity about every 18 months. But how many people know that it will one day cease to exist? That’s because real laws, such as those that govern physics, will overtake Moore’s Law sometime in the 2020’s. In his new book Physics of the Future theoretical physicist Michio Kaku describes exactly why this will happen. Today chips are created by shining UV light on a silicon wafer and then bathing it in acid to create circuits containing millions of transistors. Kaku further explains:

“But this process cannot go on forever. At some point, it will be physically impossible to etch transistors the size of atoms. You can even calculate roughly when Moore’s law will finally collapse: when you finally hit transistors the size of individual atoms. Around 2020 or so Moore’s law will cease to exist. (Page 39)”

The industry is taking a proactive approach to the problem now by investing heavily in high performance computers (HPCs), also known as supercomputers. The technology will gradually trickle down to everyday users as innovation in graphic processing units (GPUs) drive breakthroughs in parallel computing technology. With engineering and financial applications demanding more computing power every year, investors don’t have to wait for supercomputers to break into homes to jump into the impressive growth in GPUs.

Much like the computers of the 1950’s, the Blue Gene supercomputer created by IBM fills an entire room. Most supercomputers currently focus on science and engineering research, but will become available for individual use in the next decade or two.

The focus on parallel computing puts emphasis on innovative designs instead of etch-and-sketch wafer manufacturing, which will put hardware companies such as NVIDIA Corporation (NASDAQ: NVDA),  Advanced Micro Devices, Inc. (NYSE: AMD), and Intel Corporation (NASDAQ: INTC) in the limelight. Investigating the GPU divisions of these companies is a good start for developing a plan to catch the future trend of supercomputing.

Mobile is great, but GPUs are king

Parallel computing hinges on solving one big problem: how do you efficiently break up everyday computational tasks, solve them, and piece them back together in order? Companies have been chipping away at it for years, but NVIDIA seems to have provided a major breakthrough with its new Kepler GPU (opens PDF). Sure, investors point to the impressive growth in mobile device processors and the company’s Tegra 2 and Tegra 3 CPUs. The market is certainly an important piece of the puzzle, but it is not integral in driving the company forward. Consider that the company’s consumer products revenue ($132.6 million) made up only 14.3% of total 1Q13, while GPU revenue ($579.7 million) totaled 62.7%. That’s not even close.

Since their launch in 2007, Tesla GPUs have found their way into 35 of the top 500 supercomputers and 3 of the top 5. Source: NVIDIA

Demand for Kepler outpaced supply last quarter. That’s because the new Kepler GPU isn’t just the next model of incrementally better hardware; it is a game changer that will set the standard for future GPUs. Kepler changes how GPUs are accessed by the CPU with its Dynamic Parallelism feature, which is a monumental step toward making parallel computing – and supercomputers –  cheaper and more accessible.

In fact, Oakridge National Laboratory’s Jaguar supercomputer will be upgrading 78% of its nodes to Kepler GPUs. Once the work is completed in early 2013 it will have a peak performance of more than 20 petaflops, a 870% increase over the current system and more than double the current fastest computer in the world. Just for kicks, that is 20,000,000,000,000,000 floating point calculations per second.

Depending on the source, NVIDIA is either pacing the GPU industry or never far behind. But there is a consensus on one thing: the competition is fierce. AMD has kept constant pressure on NVIDIA in GPUs and CPUs for mobile devices. In fact, AMD reported CPU revenue of $1.203 billion in 1Q12, which handily beat NVIDIA’s revenue of $132.6 million for consumer products. Despite walloping its competitor in CPUs, the company has had no such luck in GPUs.

*Includes $212.6 million from Quadro and Tesla GPU sales listed under professional solutions.

Is AMD overlooking a much bigger future trend in GPUs and parallel computing in favor of the smaller, current trend in mobile device processors?

Intel does not make GPUs, but often includes integrated graphics solutions hardware as GPUs. This must be included in any discussion about parallel computing since NVIDIA (63.4%) and AMD (36.3%) control the discrete GPU market. When integrated graphics are included Intel takes an impressive lead with 59.1%, followed by AMD (24.8%) and NVIDIA (15.7%). Regardless, integrated graphics are not an ideal solution for supercomputing and are traditionally years behind GPUs. Realizing this trend, Intel purchased supercomputer hardware manufacturer Cray, Inc.’s interconnected hardware development program for $140 million. The move was made to compete with NVIDIA’s Professional Solutions business unit and to cash in on the impressive growth of large-scale HPC systems, which grew 24% and generated record worldwide revenue of $4.4 billion in 2011.

Foolish bottom line

Realizing the interconnected trend of GPUs, parallel computing, and supercomputing will reward opportunistic investors well before the rest of the investing community jumps on board. Successful implementation of parallel computing will transform virtually every industry, from bioinformatics to finance, by increasing computation speeds by orders of magnitude. The windfall will be compounded by the explosion of scientific advances that will spur further innovations and create entirely new industries (proteomics, nuclear fusion, universal language translation, real-time image processing). Companies such as NVIDIA are developing innovative new hardware systems such as Kepler GPUs that will change the course of the industry. The age of supercomputing is still in its infancy, but it still generated $4.4 billion in revenue last year. How much revenue will the industry command in 5, 10, 15 years?

Did you enjoy this article? Follow me on Twitter to keep up with my future posts on value opportunities, energy, and sustainable chemicals @BlacknGoldFool.

Make better investments with science.

• Email
• Print
• Feedback