IBM and Nvidia Team Up on Supercomputing and Servers
Here’s a second bit of interesting news on the supercomputing front. Computing giant IBM and chipmaker Nvidia are today announcing a significant partnership that will have them teaming up on the design and building of new supercomputing systems and servers.
If you look at the machines on the Top 500 list of the world’s most powerful supercomputers, which was released today, you’ll see IBM and Nvidia popping up quite a bit. Only Hewlett-Packard built more systems on the list than IBM, and not that many more. And Nvidia chips are used to accelerate the computing in 38 of the machines on the list, while at the same time helping to keep power consumption down.
Here’s what’s going on. Nvidia makes a type of chip called a graphics processing unit. Fundamentally, a GPU is really good at doing a certain kind of computation known as a floating point operation. This kind of computing work is useful in rendering the graphics of computer games — which is what Nvidia first built its business on. And floating point operations are mathematically similar to the kind of computation needed to visualize and simulate complex problems for engineers and scientists, and also in creating the visual effects in movies.
Generally speaking, GPUs are better at this kind of computation than your traditional CPU chip, like an Intel Xeon or an AMD Opteron. The difference is that a GPU chip is designed to handle lots of small computational tasks that are carried out all at once, while keeping a lid on power consumption. In computer science, this is called parallel computing, and CPU chips aren’t as good at it as GPU chips. CPUs are better at doing one job at a time, getting it done really fast, and then moving on to the next thing.
So, the important thing to understand is that GPU chips — most of them come from Nvidia, though some come from AMD and Intel, too — have been showing up in the world’s most powerful supercomputers with increasing frequency. They’re the backbone of Titan, which was once the world supercomputing champ and is still a very respectable No. 2 on the new Top 500 list.
Now, IBM — which would naturally like to take some supercomputing business away from HP — is going to work with Nvidia on supercomputers, and also on its Power line of enterprise servers. The two companies said they will share technology that will make it easier for Nvidia’s Tesla GPUs to talk to IBM’s Power8 processors.
Obviously, there’s lot of status in supercomputing. But working with Nvidia will also give IBM something new to stir up its Power server business. While Big Blue takes in more revenue from server sales than anyone else — about $3.3 billion in the second quarter, according to IDC — the fact is that the overall Unix server business is slowing down considerably. IDC reckons that Unix server revenue will decline by more than $1 billion by 2017.
These expensive and specialized Unix machines are increasingly being supplanted by less-expensive Linux machines using industry-standard Intel and AMD processors. Adding GPUs as an option won’t reverse that decline, but it won’t hurt IBM’s efforts to manage it. For Nvidia, it’s a big endorsement of its GPU technology in the enterprise.