 Foundation Summit here in San Jose and with me I've got Brock Palin who is Director of Advanced Research Computing Technology Services at the University of Michigan. And today you announced a joint project with IBM. Can you tell us more about that project? Yeah, so this project is putting together a system we call Conflux. Conflux is going to be a platform for our Center for Data-Driving Computational Physics which is inside our Michigan Institute for Computational Discovery Engineering which is one of the faculty-led institutes inside the University's strategy around advanced research computing. So this system is going to be a way of coupling machine learning with traditional computational physics. That's ambition. If I heard that right, that's data-centric high-performance computing. That's quite a mouthful so can you unpack that a bit? Yeah, so we're really looking at a lot of your traditional computational physics problems such as weather modeling, turbulence and blood flow, materials, combustion, computational fluid dynamics and a whole host of others. But in these cases these are very traditional computational physics problems. We understand a lot of the equations but it's very hard even with the high-performance machines of today to be able to really realize all the physics we want to understand. So what we're trying here is we're trying to couple machine learning in which we'll be taking in data sets either from sensors and observational studies as well as other larger-scale simulations to be able to help guide and improve our existing computational physics. That's a lofty ambition. So it's bringing together the compute side and the design side and the software side and the data side all together in one piece. Is that correct? So the focus here really is around the science and the machine itself was designed around a lot of data movement because we are going to be dealing with a large quantity of data here. So this is why with the power platform from IBM we are also utilizing NVLink which is a very high-performance way of moving data from the main system memory to the NVIDIA Pascal GPU that will reside on the compute nodes and this is really the fastest way to be able to move data from a host system into an accelerator today and that is really one of the main bottlenecks we have faced in the past. We've also utilized IBM Spectrum Scale Storage which allows us to be able to access the same set of data both from our traditional physics codes because we have an existing tested base of applications as well as from a lot of the platforms that we're building our machine learning in such as Spark and Hadoop and other types of things that come from the big data world. So that sharing of that data between all these different projects that you've got on that sharing the same data between all of these different areas at very high speed I presume is that right? Yeah yeah so we're also utilizing Melanox 100 gig in FinaBand which is also accessible over the CAPI interface which is another unique feature of the power platform but really the advantage of the Spectrum Scale being able to access the same data in both ways is that we don't have to move the data around we don't have to end up with two like a shadow copy of that data in another system and when you're really working with data of the size we're talking about here you really don't want to move it you want to move it as few times as possible and really only access it when you're directly computing on the data. So that's a lot of lot of research to do in one small thing what's your vision in three years time when we come and visit you in Michigan what are you hoping that will come out of this what breakthroughs are you hoping for? Right so the University of Michigan is unique as a public research institution we have one of the largest research budgets but we also have one of the most diverse research footprints the university last year had over 101 top 10 graduate programs which is unique in the entire country and really for my job at ArcTS advanced research computing technology services is to try to make a platform of resources that serves both data privacy computation visualization analytics consulting and support to be able to support a very diverse group set of researchers today we support several thousand researchers in the future I expect to be touching even more and also create part of the teaching mission of the institution also we're also getting into a lot more with our hospital around things like patient care and direct patient care we got a lot of focuses around data as well as simulation and modeling and this goes well beyond the conflicts effort conflicts is something we hope to grow and build off of such that we are able to provide a better set of understanding and resources so Michigan can be the top and maintain its position as a top research institution and create some of the best science we ever have and I think our best days are still ahead of us I hope so too and congratulations on the work that you're doing with this and congratulations on this program it sounds exciting I have many relatives have been to University of Michigan so looking forward to more of them going there so thanks very much