In this special guest feature, Alex Larzelere from the U.S. Council on Competitiveness writes that the new Sunway supercomputer is a bigger story than China overtaking the U.S. in the TOP500.
Having worked in our national labs for 25 years as part of the U.S. Department of Energy’s high performance computing initiatives, I can only imagine our scientists’ dejection on learning that the Chinese are kicking the “FLOP” out of us.
I want to congratulate China’s Sunway TaihuLight for winning the latest round in the high performance computing (HPC) race and becoming the world’s fastest computer. Now that it’s happened, how will the U.S. supercomputer community respond to stay competitive?
Achieving the No. 1 ranking is significant for China’s economic and energy security, not to mention national security. With 125 petaFLOP/s (peak), China’s supercomputer is firmly on the path toward applying incredible modeling and simulation capabilities enabling them to spur innovations in the fields of clean energy, manufacturing, and yes, nuclear weapons and other military applications. The strong probability of China gaining advantages in these areas should be setting off loud alarms, but it is hard to see what the U.S. is going to do differently to respond.
Wake Up: China Created a HPC Ecosystem
The Chinese accomplishment is far more significant than one computer coming in first. Other countries have done it in the past with what’s called “stunt” machines ― fast and powerful, but unable to apply it to anything. The stunning news here is that China created a High Performance Computing “ecosystem” in an amazingly short period of time.
In 1993, when the first TOP500 list was published, China was not even on it. Ten years ago, in 2006, China had 28 systems on the list and the U.S. dominated with 298 systems. Five years ago the count was China 61 and the U.S. 255. As of June 2016, the tables are turned. China has 168 systems on the list and the U.S. dropped to #2 overall with 165 systems.
Another important part of the Sunway TaihuLight announcement is that for the first time, the Chinese – built a 100 percent indigenous supercomputer. In an ironic twist, our scientists are trying to learn the recipe of the secret-sauce to the Chinese processor. China’s ecosystem also created the hardware and the middleware required to operate the computer.
This is a big deal. Let’s shake ourselves out of sleep mode. It took the U.S. over 50 years to develop an ecosystem that once dominated the world of computing. For national and economic security reasons, the United States made the necessary investment to build a HPC ecosystem. Now in about 10 years it has been matched by China. They clearly believe they cannot “be a superpower without supercomputing.” The question now is, does the United States still believe that too, and if so, what are we going to do about it?
A good place to start is the recently released White House National Strategic Computing Initiative (NSCI) Executive Order. The order provides a great overview on the need for HPC supremacy and prioritizes the development of this technology calling it an “all of government” approach – a national priority. The order also talks about giving more accessibility of these systems to U.S. industry to improve competitiveness. My top-5 list on ways the U.S. can build the fastest computer adds a few more points including an embarrassing role reversal: learn from the Chinese.
The problem is that while the U.S. was studying how hard the next generation of computing is going to be (aka exascale), we stopped deploying large systems and let the U.S lead slip away.
Maintaining leadership in high performance computing is a national and economic security imperative. As history has shown, HPC translates into innovations in aircraft, automobiles, oil and gas, biomedical, nuclear energy and military systems. For the last 10 years, the U.S. Council on Competitiveness has run the High Performance Computing Advisory Council. The mission of the group, whose members come from industry, academia and national laboratories, is to promote the idea that “to out-compute is to out-compete.”
Let’s get our heads back in the game. In this Olympic year, do we really want to be No. 2 in the supercomputing race? A lot more is at stake in this global contest than the color of a medal.
About the author
Alex Larzelere is senior fellow at the U.S. Council on Competitiveness and is a member of its High Performance Computing Advisory Committee. Previously, using our national lab supercomputers, he developed advanced modeling and simulation capabilities for nuclear energy at the U.S. Department of Energy. He has worked with Oak Ridge, Lawrence Livermore, Los Alamos, Sandia, Argonne, and Idaho national laboratories to provide tools used to create improved insight into advanced nuclear energy technologies.
Does reliability improvement follow the speed and Flops improvement. What about errors related to natural radiative environment?
Gaillardia – the answer to your first question is just the opposite. As computers faster, more parts are used and thus reliability goes down. Computer engineers have found ways to live with this. As for national radiation – yes that is an issue. A computer in Los Alamos (elevation 7,000 ft) will have more errors than one in Livermore (about 500 ft). Once again, ways have been found to live with these issues.
No.
Let them eat Intel. The USA did little but maintain nukes and run some neat science, but it was up to the Fortune 500 to invest in HPC or not. And they did not. The people did. If by people you mean the federal government. Then yea, the people took a lot of borrowed Chinese money and made many HPC’s. We’ve stopped and the Fortune 500 never stepped up.
Now you recommend ‘we the people’ go back to the Chinese for MORE money to figure out how to bet the Chinese? Who have their own 3-way contest for EXE? No.
We lost and you did it.
You get your own money from now on. And just to make sure I will campaign and vote for politicians that see raising taxes and stopping the rape of the poor as necessary. Yea, this ‘we the people’ had to eat his own Intel.