By Bill Lee

As Supercomputing Conference 2018 (SC18), held this year in Dallas, TX, quickly approaches, the High Performance Computing (HPC) industry gears up to discuss topics ranging from HPC in the Cloud to the Convergence of HPC and Big Data. Each of the topics highlighted during the show will provide insight into the current HPC technology landscape, as well as the future of HPC and its potential impact on science, education and more.

During SC17, we chatted with industry experts on what they see as the most important aspects of the HPC industry today and what they mean for the future of HPC. With perspectives stemming from a wide variety of companies and organizations, they could all agree on three main factors that have and will continue to have a significant impact on their respective industries: Remote Direct Memory Access (RDMA) technology, standards bodies and data/compute-heavy applications such as Artificial Intelligence (AI). As we look to the future, let’s review these important topics and what they mean to the industry.

 

On the Impact of AI on Specific Industries

“I think the better question is what industries aren’t going to benefit [from AI]? If you look at the mountain of data that is available to customers today and the capability of actually crunching through that and coming up with meaningful results in a very quick manner – machine intelligence, artificial intelligence, all that is going to play a very big role.”

-Greg Gibby, AMD

“There’s a lot of people who are going around saying that AI is the new electricity, and that pretty much every industry will be impacted by AI in the same way that electricity changed every industry. I truly believe that.”

-Sumit Gupta, IBM

 

On the Benefits of RDMA Technology on AI

“Most of the analytics if you look at an AI workflow, the heavy lifting is on the back end, where your storage systems are trying to feed the GPU to do the learning and the training of the systems and the models themselves…this is the domain of high throughput, low latency infrastructure – the interconnect.”

-David Hyatt, WekaIO

“Today you’ve got a mountain of data that you need to quickly analyze and come up with decision making. RDMA helps out in two kinds of major ways. One, it provides scalability, allowing you to have multiple nodes and access those nodes. And then it improves on the performance by offloading the things it would normally be taking care of in the NIC, it can go access the memory directly, allowing the processors to have better performance.”

“The scalability that RDMA enables to allow multiple node systems to access that mountain of data. And then the performance advantage that you get by being able to…access that quickly across multiple nodes without a performance impact is very, very compelling.”

-Greg Gibby, AMD

“RDMA is extremely important for us, and to put it even more extreme, we cannot actually operate our architecture without RDMA. The reason for that is that we use NVMe drives, which are the new generation of SSDs…and using RDMA allows us to access these drives in a parallel fashion from many different sources concurrently.”

-Zivan Ori, E8 Solutions

 

On the Importance of Standards Bodies for the Advancement of AI

“Innovation and adoption are enabled by collaboration and open standards. And I see the IBTA and open standards as being critical to supporting that and to making AI successful.”

-David Hyatt, WekaIO

“Open source and industry standards are very important. Not only for AI but actually across the board in any kind of scientific discipline. AI stands to benefit from this because you have communities within communities that are working on very specific elements of AI.”

-Scot Schultz, Mellanox Technologies

 

Attending SC18?

The following IBTA members will be in attendance during SC18. Be sure to stop by their booths to discuss RDMA, InfiniBand and how these technologies are changing the face of HPC as we know it.

  • Cray (#2413)
  • Fujitsu (#1226)
  • HPE (#2428, #2828)
  • Huawei (#3047)
  • IBM (#3433)
  • Intel (#3223)
  • Mellanox (#3207)
  • Microsoft (#3442, #4515)
  • Molex (#3941)
  • NetApp (#605)
  • Oracle (#2806)
  • Samtec (#521)