A Software Development Kit (SDK) that can alter existing code to engender much faster performance from a multi-core, cross-platform environments is huge in the HPC space–especially in finance. Xcelerit’s SDK does this very thing and their recent announcement of a partnership with Intel that utilizes the Xeon Phi really steps things up. We caught up with Hicham Lahlou, Xcelerit CEO & Co-Founder, to find out more.
insideHPC: Xcelerit is being called a hot startup out of Dublin, Ireland. What’s the excitement about? What does your company do?
Hicham Lahlou: Although we are operating now since 2010, and have rapidly grown over the years, we manage to retain the atmosphere of a start-up. Our company was born out of a research center at Dublin’s Trinity College. We were working on complex numerical problems involving hardware accelerators before discovering a novel approach that would open up HPC to non-expert programmers.
Our core product is the Xcelerit SDK, a Software Development Kit that makes it easy for domain specialists (i.e. mathematicians in banks or geophysicists in energy exploration firms) to convert their existing code to take advantage of multi-core, GPU and other hardware accelerators. We do consultancy too though. Often a customer will come to us and ask us to examine their existing code base. Our acceleration experts can spend just a little time with the code and recommend small changes that can speed things up by 2x, 5x, or even 10x over their existing setup. If they want to go further, we can arrange technology pilots and even help them in sizing and sourcing the acceleration hardware before equipping them with the tools that can make it sing.
insideHPC: So far, you have focussed on financial services, how does your technology bring value to your customers?
Hicham Lahlou: In the early days of the company, we focused strongly on Investment Banks and other financial institutions that needed to do complex risk calculations. We have had great success in this area, so much so that seven of the top ten investment banks are now using our software in one form or another. Since the financial crisis, regulators are insisting that banks keep a very tight rein on counterparty risk (i.e. the risk associated with a counterparty’s default before the end of a contract) – this means deeper analysis, with more time steps, more scenarios, etc. In addition to the back-office regulatory-driven analysis, there is also a new emerging need for complex risk calculations to support front-office trading. Clever use of our software allows our customers to minimize their risk exposure while maximizing profits.
insideHPC: What other verticals might your acceleration software help?
Hicham Lahlou: Our technology can make a huge difference wherever a lot of computing power is needed. The next big focus for us is Oil & Gas. This is an area where our target customers are already equipped with large compute clusters and accelerators, but they struggle to extract the best performance from the hardware they have.
We have already had success with Reverse Time Migration (RTM) – one of the hot algorithms used today for Seismic Imaging. We have adapted existing implementations to use our SDK that achieve a performance level on acceleration hardware that is as good as using low-level programming approaches, but involving far less development time. Our implementations can instantly adapt to the addition of new hardware and can work across CPU and GPU platforms. We expect to be able to make our customers lives much easier while delivering the necessary performance.
insideHPC: You recently announced support for the new Intel Xeon Phi platform. Tell us more about that?
Hicham Lahlou: Well we’ve always been pretty good at extracting the best performance from Intel processors. From the beginning, we were able to automatically schedule the workload across the available cores and hardware threads. Lately we have added support for automatic vectorization and this is generating a huge leap in performance – overall, speed-ups of up to 100x are possible on CPUs compared to sequential code. Our most recent announcement with Intel is around providing support for the Xeon Phi Co-processor. From our customers point of view though, this is all completely seamless – a key feature of our technology is that the Xcelerit-enabled code is portable across different hardware. The code just runs on the new platform delivering new performance levels with no need to recode.
insideHPC: This is obviously excellent news for all involved. What does future hold for Xcelerit?
Hicham Lahlou: This expands our hardware coverage nicely. At Xcelerit, we believe that the future is Hybrid – CPUs are good for some tasks, GPUs for others. New platforms will emerge with their own unique advantages and we hope to open up access to them for all of our customers. In terms of where Xcelerit is going, I mentioned that we now have a firm foothold in the computational finance area and that we are establishing a presence in Oil & Gas. There are many more vertical markets we can tackle including academic research, big data analytics and of course catering for the needs of large government laboratories. These customers are already using accelerators and have an almost infinite need for computing power. We can facilitate that – connecting the computing power to the problem in dramatically less time than the tools they are using today.