Subtitles section Play video Print subtitles Hi, my name is Alex Hudak, I'm an Offering Manager at IBM, and today I'm going to talk to you about: What is a GPU? So, I get some pretty basic questions on GPUs, and that's what I'm going to go over today. First question is, what is a GPU? What is the difference between a GPU and CPU, so I'm going to represent those here. And then, lastly, why use a GPU? And, is it even important to use a GPU on cloud? So, let's start first with, what is a GPU? GPU stands for "Graphic Processing Unit". But, oftentimes people are more familiar with CPUs. So, CPUs are actually made up of just a few cores. You can think of these cores as the power, the ability of a CPU to do certain calculations or computations. On the other hand, though, GPUs are made up of hundreds of cores. But, what difference does it make? So, the thing with the CPU is that when it does a computation it does so in a serial form. So, it does one computation at a time. But with the GPU, it does it in parallel. So, the importance of these two differences is that with a GPU, you're able to do computations all at once, and very intense computations at that. So, oftentimes when you have app codes, a lot of it's going to be going to the CPU. But then every now and then, you're going to have an application that's going to require quite a bit of compute-intensive support that the CPU just can't do, so it's going to be offloaded to the GPU. So, you can think of a GPU as that extra muscle or that extra brain power that the CPU just can't do on its own. So, there are two main providers of GPUs in industry: NVIDIA and AMD. Both providers manufacture GPUs that are optimized for certain use cases. So, let's jump into that, because a big question I get is: Why do I even need a GPU? In what industries and in what use cases? So, the first we'll talk about is VDI. VDI stands for "Virtual Desktop Infrastructure". So, GPUs are created to support high-intensive graphic applications. Think about if you are a construction worker, right? And you're out in the field, and you need to access a very high-graphic-intensive 3D CAD program. So, rather than having the server right next to you or right in the field with you, you can have a server that's a country away in a cloud data center and be able to view that 3D graphic as if that server was right with you. And that's going to be supported by the GPU because the GPU supports graphic-intensive applications. Another example of this would be movie animation or rendering. So, in fact, GPUs actually first got their name mainly with the gaming industry. Oftentimes they were referred to as "gaming processing units" because of this ability to provide end users with low-latency graphics. But gaming is no longer the focus in industry anymore. It's a big piece of it, but now financial services, life sciences, and even healthcare are starting to get into it with artificial intelligence. So, artificial intelligence has two big pieces to it - there's machine learning and there's deep learning. So now there are also GPUs that are optimized and created specifically for those applications. There are some that are created for inferencing for machine learning purposes, and there are some that are created to help data scientists create and train neural networks. In other words, they're trying to create these algorithms that can think like a human brain. That's something that a CPU can simply not do on its own, and it requires GPU capabilities. And then, lastly, let's talk about HPC. HPC is a buzz word that's been going around, it stands for "High Performance Computing". While a GPU is not absolutely necessary for HPC, it's an important part of it. So, high performance computing is the company's ability to spread out their compute-intensive workloads amongst multiple compute nodes (or, in the case of cloud, servers). Oftentimes, though, these applications are very compute-intensive - it could include rendering, it could include AI - and that's where a GPU comes in. You can add a GPU to these servers that are spread out amongst an HPC application and utilize those in that manner. So, this is a nice little segue into why should use GPUs on cloud. If HPC is such a big piece of that, what else is important about cloud? So, the first part of that is you get high performance - you need the cloud for that. The GPUs are great . . . but not on their own. So, back in the day (and even still today), there are companies that use a lot of on-prem infrastructure, and they utilize that infrastructure for any of their compute-intensive applications. However, especially in the case of GPUs, the technology is ever-changing. In fact, there's typically a new GPU coming out almost every single year. So, it's actually very expensive and nearly impractical for companies to keep up with the latest technology at this point. Cloud providers actually have the ability to continually update their technology and provide GPUs to these companies to utilize them when they need them. So, on a more granular basis though, cloud technology can often be broken down from an infrastructure perspective between bare metal and virtual servers. So, let's talk about the differences. There are advantages of using a GPU on both types of infrastructure. If you utilize a GPU on a bare metal infrastructure, the companies oftentimes have access to the entire server itself, and they can customize the configuration. So, this is great for companies that are going to be really utilizing that server and that GPU-intense application on a pretty consistent basis. But, for companies that need a GPU maybe just on a burst-workload scenario, the virtual server option might be even better. And the nice thing about virtual is that there are often different pricing models as well, including hourly. And the cool thing about cloud is that you only pay for what you use. So, if a company is using on-prem technology or infrastructure but they're not utilizing it at the time, that technology is depreciating, and it's essentially a waste of money for that company. When they offload to the cloud, they only pay for what they're using. And, so, it just makes a lot more sense from a cost perspective; and then, because the GPU is so great at performance, it just makes sense for a performance perspective as well. So, companies are able to focus way more on output than they are on keeping up with the latest technology.