This year has been an exciting year in the supercomputing industry as there have been many new developments. Two weeks after the world’s largest conference in the industry, SC19, all the announcements have made it to the internet and opinions have been shared, which makes it the perfect time to highlight the developments of the year.
In my opinion, the top two developments in supercomputing have been two new players in the processor market finally taking a significant share in the industry. The past few years have been dominated by Intel and Nvidia. However, this year two new players began to take some of the glory. AMD and ARM have both designed new processors and seem to be taking a market share from Intel with computing systems on the TOP500 list, a list of the top 500 most powerful computers in the world.
AMD brought products to the market competing with both Intel and Nvidia. The AMD EPYC processors offer a higher memory bandwidth than Intel processors and push the envelope of performance with an average of 30 percent higher performance on supercomputing workloads than Intel. The AMD Radeon Instinct beat Nvidia to the punch, releasing the world’s first PCIe4.0 bases graphics processing units. The increased communication bandwidth coupled with the optimized software stack and virtualization technologies drive higher efficiency and datacenter optimization than current technologies by Nvidia. I would say the biggest obstacle to AMD is the fear of adopting new technologies in research-based workloads.
ARM made its entrance into supercomputing in 2011 as part of the European Mont-Blanc project. ARM had already dominated the mobile market; chances are you have an ARM processor in your pocket right now powering your smart phone. However, making the leap from mobile to the datacenter was a challenging proposition. This year they made a big splash with introduction of a product in partnership with NVIDIA, and various hardware vendors to build GPU-accelerated Arm-based servers. These servers hit the strongest area of the HPC market as a majority of the TOP500 supercomputers are powered primarily by Nvidia GPUs. Coupling the energy efficiency of GPUs and Arm CPUs results in some of the most energy efficient computing platforms available today.
The other big change in supercomputing are the new breakthroughs in machine learning and artificial intelligence. These breakthroughs are redefining scientific methods and enabling exciting opportunities for new architectures. The floodgates are open for new technologies that accelerate the machine learning methods, which are less computationally intensive and more memory intensive. Driven by these breakthroughs, ARM and AMD both bring higher memory performance than we have ever seen before into the market.
The final new technology breakthrough in supercomputing in 2019 was the proof of quantum supremacy, which we have talked about before. It does not mean a lot right now and will probably steal the show in 2030 as the adoption of new technologies in supercomputing seems to move slow.
Why is any of this important to you? Today’s supercomputers are used for weather forecasting, new product development, drug research and scientific modeling. Supercomputers have replaced the laboratory in many industries, reducing the total cost of product development thus reducing the cost of the product. As supercomputing power increases, the cost of consumer goods should decrease and the level of public safety should increase.