 So here at the IBM booth here at the super computing conference, and who are you? My name is Etra Teotron. I work for the IBM Excel compiler team. So the compiler team is a team that developed like compiler technology, which is used for the core machine and people as summit and Sierra will use our compiler to create good To take that application and put it to open. So right here. Is this a super computer? What is this? This is an example of the installation of the summit system that it's currently being delivered to To call and so that's Of how the machine actually looks like each cabinet contains several Racks with power night and NVIDIA GPU So it's NVIDIA GPU kind of solution It's not just the NVIDIA GPU is a mix solution between You know the IBM chips the power mine chips and then the best of breed of GPU computation Which is the NVIDIA bolt GPUs and together they create a system that is very Balanced and adapt very well to different high-performance computing So how many super computers in the top 500 use power? So I don't know the number So there are some For sure has a history of being in the super computer for a long time. Well IBM has been one of the company that has been Committed to high-performance computing for a very very long time back to you know the Power system and now with the transition to hybrid system IBM is again trying to deliver the best of breed between Both CPU and GPU So the video over there is going to show an installation of of the system for And Example of our compiler which is being used to take some of the coral scientific Computing application and ported to open MP 4.5 So we are a working collaboration with IPM research and our customers LLNL and Oak Ridge National Lab to take this application and Put them to the machine in preparation for the delivery of the system Showing them What this is about is a Demonstration of several application. This is a coral micro mini application developed by LLNL and Collaboration with Laurence Levermore. We have taken this application Ported it to 4.5 and started to measure the performance difference between running the application just on the power system That's a opening the application on running on on power and then moving it to the GPUs and we have done a study to figure out how to Develop the compiler the application itself to get better performance on the GPU Here already we have a ten times speed up between running it just on the two power processors or on an NVIDIA Pascal GPU and furthermore It's very easy to use opening all the Developer have to do is to identify the looks that they have parallelism and add a little bit of Syntax this is opening P direct test to tell to the compiler the system look or flow this computation to the GPU If you do that We integrate very well with the NVIDIA tool. That's an image from the NVIDIA visual profiler these are the The application the opening the application compile with our Excel compilers Integrates with the NVIDIA tools you can use the tool to try to figure out where the bottlenecks are in your application For example, you can see Carbons are all floated to the GPU Sometimes what the car is not actually doing anything over here, so We have enough of the open be directed to allow you to dispatch the application By adding this no way Yes So it's very easy that means that the austere will not stop at this point To a flow of the kernel and the programmer is able to enforce dependencies when you do that Your kernels are perfectly pipeline on on your car, so they are all running back-to-back without any time in between And the performance difference is from the red line to the yellow greenish line It's about 24% better than anything to offload the application So is that what you work on you work on this kind of stuff? I am a developer I'm the lead developer for open and P In the IBM Excel compiler for both C-C++ and Forty And what else is around here? Can we look at that? So over here, let's grab it Over here is a This is the NVIDIA V100 Volta GPU So that's one of six four to six GPUs that are in the IBM System so in the I believe that the summit system left six of these Connected to two power nine CPUs if we move over here you can see the entire Yeah So that's the IBM power system AC 922 and and you can see that Six GPUs each And there is a MV link to bus between OST processor and the GPUs and that's unique So in the car you cannot see it here, but there is a bus that Call MV link to the connect the GPUs To the OST processor the power nine and NVIDIA provides that only to In this system for the IBM solution. So is this one of the most cutting-edge Super computing systems? Yeah, so this is actually the node that you saw over those racks So each racks are several of these nodes And that's basically that's the the system that the coral will use The communication across network for the machine. All right. So in this system, you don't just have one node You're gonna have several Cabinets and you're going to You know need to communicate across those cabinets. So what we need you need an infinite band connection That's the switch that provide the connection So there's a lot of activity in the super computing space and IBM is at the forefront of this effort at the forefront out of the Three coral contracts two of the sites were awarded to IBM. So, you know both the Oak Ridge National Lab and the Lawrence Livermore National Lab chose IBM as their provider together with NVIDIA for you know, the next generation of super computers and