Hi, welcome to today's video. Today, we're going to be talking about HPC architecture. The purpose of this video is to ensure a basic understanding of computing hardware and infrastructure on a supercomputer. This is going to include a brief discussion of a typical supercomputer. We'll be talking about processors, memory, network, and storage, with a particular focus on processors and memory. The architecture of a supercomputer falls from highest level to lowest level, the highest level being the system as a whole, nodes, and then all the way down to the core's memory and storage. A network interconnect connects the components of the system as a whole and allows it to operate together. Supercomputers can encompass entire rooms, much like just our regular computers did in the mid-twentieth century. How large your system is going to be is largely dependent on how much space you have and how much money you have. Components of a system are mounted in racks, and racks can be nice cabinets with rails or some other system that can be used to contain the infrastructure. You can purchase standard racks or you can customize them, depends on what your needs are. We here at the University Colorado operate a supercomputer called RMACC Summit. RMACC Summit stands for the Rocky Mountain Advanced Computing Consortium. Our RMACC Summit system has 10 racks. Those are nine compute racks and one storage rack. And racks are black metal cabinets that enclose infrastructure. You can see in this picture here that we have five black metal cabinets that you can see as part of RMACC Summit. Behind each of the racks is a cooling unit. The purpose of the cooling unit is to move hot air through heat exchangers, which allows the system to keep an optimal temperature. Infrastructure is made up of compute, networking, or storage. Storage is made up of disks. Network is made up of high-speed switches and cable. And compute, which is the most interest here, is made up primarily of compute nodes, cores, and memory. In the picture that you see here, we're looking at eight chassis that contains four nodes each. Which means that we have 32 nodes total that you're looking at on the system. So what are nodes? Let's take a look at what a compute node is. There are two main components that we're going to be talking about on a compute node. One are the socketed CPUs, and the other is memory. So let's start with the socketed CPUs. The socketed CPUs sit within sockets on a node under large silver heat sinks. At least in this picture, we see that we have two large silver heat sinks, two sockets, and each one of them has the CPU. Within the CPUs are cores, which are your main processing power and memory. We can't see the cores or the memory in this picture because they're underneath the heat sinks. But this particular node has 12 cores per socketed CPU, which means that this particular node has 24 cores total. And when you hear about how powerful a system is, it's usually because they're talking about how many cores the system is, that's where you're going to derive your power from. Next, let's talk about memory. There are two types of memory that are on the nodes in this particular picture. The first, which is potentially more obvious, are the RAM memory, the Random Access Memory. And this you can see here. These are these eight green thin cards that are surrounded by the black rectangles. So we have one, two, three, four, five six, seven, eight. What RAM memory does is it shares a memory on one node. So I'm on all 24 cores on the node. Each one of these memory cards is 16 gigabytes. Also, there's memory within the socketed CPUs, as I mentioned earlier. And again, that's underneath the heat sink, so we can't see it. The memory there is cache and shared between the cores on one socket. Meaning, it's shared within 12 cores. Finally, we'll talk about the interconnect. So supercomputers work together as one big unit to solve larger problems. And that's where you get all your large processing power. You can imagine that if you can use the entire system, it would be way bigger than any of your laptops. But of course, we have to limit how much of a proportion of the system one person can use. But the way that all these components work together is through an interconnect. And that allows the entire system to have access to all the memory and all the computing power on all of the nodes. The nodes can talk to each other through two types of interconnects, and these are two main interconnects that are out there on the market now. OmniPath and InfiniBand. OmniPath is what we're using on the RMACC Summit supercomputer, but our previous generation of supercomputer used InfiniBand. So now that we know about HPC infrastructure, how would you use all of this infrastructure to solve a larger problem? We'll talk about that next.
