As a late sign-up for the recent HPC & AI Meetup in Bristol, I was feeling relaxed coming in without a speaking slot and only the objective of meeting some colleagues and hearing a few lightning talks. Well, all that changed with two minutes from the start when our MC for the night, Verne Global's fabulous Simone Warren, told me that I had just been awarded the honour of giving the closing remarks. Her rationale was sound - it was a technical meeting - perhaps the CTO should give the exit speech. Logical indeed, and while I do in fact enjoy a bit of public speaking, I will admit that I do like to over-prepare myself for any speaking role and I knew I was in for a new type of challenge. My mind had to focus in and quit thinking about that Banksy I was looking forward to seeing around the corner from the venue. Little did I know that he was stealthily working in the background on a Valentine's Day surprise.
Photo Credit: Banksy
What I learned from the overall experience is that when you are the impromptu closer, you end up paying very close attention to the talks, perhaps even more so than I might normally, so I was able to get a lot out of the evening, thanks to the meetup's fantastic speakers. Bonus.
As for our venue, it was fantastic. We were all hosted in Graphcore's Bristol offices which are just up from a beautiful riverside park that buzzes throughout the day with all types of activity. We were greeted with a wide selection of beverages as you would expect, but what I don't believe many expected was that the Verne Global team had arranged for several racks of Icelandic beer. Even in a town like Bristol that is famous for its brews, the bottles of Einstok were well received.
After a warm welcome by Simone, Helen Byrne AI Research Engineering superstar at Graphcore, started things off with a thorough review of the technical benefits of Graphcore’s IPU approach to AI computing. What really impressed me about Helen's talk was that she was able to bring complex topics like mixed-precision training and stochastic rounding down to the level where the casual observer could understand why and how Graphcore's IPU technology can be applied as a better solution. Likely the most important area of focus for Graphcore as they embark on their 2020 focuses is their work on improving the efficiency and the scale of distributed learning systems on large scale machines. Graphcore's IPU layout is clearly built for distributed algorithms from the ground up, and needless to say, Helen really excited the crowd with how their IPU technology can lead us forward into more sustainable options with both conventional and spare training methodologies.
Steve Smith of Dell Technologies woke every one right up with a very engaging talk, complete with props, about the mind-blowing scale of big data. I will admit that I've seen big data talks before and I went into the session wondering if this would be more of the same, but Steve really challenged us to see if we could visualise big data. He chose the big data field of genomics to really make the point. After spooling out across the room metres and metres of thermal paper with tightly printed letters, Steve let us know we were looking at only 1/30,000th or so of his genome. My favourite tweet by @DoctorRobotnik said it all -
Steve's point was clear - for the future of HPC and AI - there is an unbelievably large reservoir of data out there for use and just like the universe itself, big data won't be getting any smaller.
After some amazing pizza from Dough (a must visit if ever in Bristol) and more Icelandic brews, we were back for the last talk from Simon Ponsford, CTO at Bristol native YellowDog. Simon's talk was the perfect progression from the earlier topics. Helen convinced us that faster hardware is needed for the bigger and bigger AI models that are to come and Steve let us know that the only way that we can hope to tackle the data at our disposal is with mega scale - but thankfully not using 3-tonne rolls of paper anymore. As Simon showed, YellowDog offers organisations the ability to scale for HPC & AI workloads through their clever use of multi-cloud management coupled with truly innovative predictive scheduling and workload orchestration. I was especially drawn to their YellowDog API that allows their users to prioritise the HPC cluster features that are important for their workloads. Whether time to insight, minimum cost, or even maximum sustainability is the driver, the YellowDog system harnesses an impressive inventory of cloud resources and then delivers the specific cluster that is required to meet the performance parameters defined by the end users. And if you have any questions about YellowDog's commitment to sustainability, just read CEO Gareth Williams' blog on the next five years for YellowDog.
Needless to say, when it came time to wrap it up, I had no problem connecting the dots between Graphcore's novel approach to better AI computing, Dell's observations on the scale of our challenges and opportunities, and the new class of multi-cloud capabilities that companies like YellowDog are bringing to the HPC & AI community. So that was the easy part. But I also wanted to make a few observations for the rising leaders of computing.
1. Invest in leadership. Don't just develop your skills in technology. Invest in developing your skills in leadership. Chapeau to Novartis CEO Vas Narasimhan for this cogent advice in a recent interview on the a16z podcast - The Science and Business of Innovative Medicines.
2. Grow your network - Meet Up! Don't just come to a meetup to meet with your colleagues. Make the effort to reach outside of your comfort zone. I challenged everyone to introduce themselves to two new people on the way out of the door.
3. Embrace accountability. If we want our visions to live beyond us, we have to work together to lay strong foundations for future growth. These foundations are laid through accountability. More on that topic in a recent blog.
And with that, we closed an A* session. Thank you to everyone who joined us in person and remotely via Zoom. Sign up to the HPC & AI Meetup group to stay updated on the upcoming sessions. We are already looking forward to the next one!