The keynote speech given by Dr. Jack Dongarra at HPC Asia 2009 examines the history of high performance computing (HPC) from its beginnings in the 1950s, through the present, and into the very near future. The talk also takes an in-depth look at the TOP500 supercomputing list which was begun by Dr. Dongarra and several of his colleagues in 1993.
This extremely educational and enlightening talk also takes a look at current trends in HPC such as “many-core” chips and GPUs as well as examines future obstacles in the ongoing development of HPC.
Part1 | Part2 | Part3 | Part4 | Part5 | Part6 | Part7 [PDF Download] [Video]
Part 4ˇGLife at the Top of the List; The Top 10
What we’re looking at here on this slide is the total core count inside of the machines on the TOP500 list. If we take a look at this slide, we see that the majority of machines have a core count of between 2,000~4,000 cores. Think of that is the “sweet spot” for HPC today—between 2,000~4,000 cores. We also see some machines that have a great many number of cores, for example, there’s one machine that has more than 128,000 cores. There are also a few machines which have a very low number of cores but, of course, there are no machines with just one core. That just means that everything is “parallel.”
Next, if we take a look at countries that have machines on the TOP500 list, here is the breakdown: The U.S. is quite dominant with 58% of the machines on the TOP500 list. The UK is #2, France and Germany are tied for #3, Japan is #4, and China is #5 on the list. Also, India has 2% of the machines on the list. As you can see, The U.S. is quite dominant. This is a breakdown across various segments regarding how these machines are being used. The striking thing here is that industry has a large and growing fraction of HPC use. What that means is that industry realizes that HPC offers a strategic advantage to the development of their products.
This is the distribution of the 500 machines today in terms of their computational power. The thing that’s very noticeable here is that there are very few machines at the top the list with the highest performance and then it has this very long tail to it. These are the machines at the top of the list. There are two machines greater than a PFlop. And, again, the lowest machine is at 12.6 TFlop/s. If we look more closely here--two machines have greater than a PFlop, 19 systems are greater than 100 TFlop/s, 51 machines have greater than 50 TFlop/s, and 119 systems are greater than 25 TFlop/s. Then, again, there’s this long tail. There are two machines on the list here in Taiwan. There’s a machine at the National Center for High-Performance Computing that’s #181 on the list and then there’s another machine very close to the end of the list which is used for semiconductor manufacturing. That long tail brings about a lot of change to the list. You’ll notice that the TOP500 list has a lot of turnover. Looking at the list, 267 machines dropped off the last time it came out. That long tail got “whacked off” as new machines came on the list and replaced the old ones. So what happens here is that the whole list’s performance increases every time, of course, and because of that long tail, a lot of machines drop off the list.
This is the top 10 machines on the list -- the 10 fastest computers in the world today. One striking thing we see here is that nine out of 10 of them are in the U.S. and the other one is in China. The first column represents the ranking of the machine. The second column shows the institution where the machine is located. The third column shows the name of the computer. The fourth column shows the country the machine is located. The fifth column shows how many cores are in the machine. The sixth column represents the machine’s rate of execution. The first thing that we notice in this list is that there are a large number of machines at the Department of Energy’s laboratories. The Department of Energy’s laboratories really do dominate the list here. In the U.S., the Department of Energy is using high performance computers to help solve some of the most challenging problems that they are facing.
Also on this list of the top 10 machines, there’s a scattering of companies represented --IBM, Cray, SGI, Sun, and Dawning. Dawning, of course, is a Chinese company. Shanghai Supercomputer Center is #10 on the list. Another thing that’s interesting is that the supercomputer at the Shanghai Supercomputer Center, the one that’s ranked #10, is running Windows as its operating system. So Windows actually does just fine running this large cluster machine. All of the other machines are running some version of Linux or UNIX.