CUDA development and GPUs scored big in the world of supercomputing recently. An HPC application using a GPU computing cluster and software written in NVIDIA’s CUDA C was awarded a highly coveted ACM Gordon Bell prize. The Gordon Bell awards are often referred to as the Nobel Prizes of the supercomputing industry and are awarded each year to recognize outstanding achievements in HPC.
The award went to Dr. Tsuyoshi Hamada of Nagasaki University, Japan, and his colleagues for achieving the best price/performance in an HPC application. Their submitted results were produced using a 256 GPU cluster running leading-edge research into simulations in astrophysics and fluid turbulence. For less than $230,000 in hardware costs, this GPU computing cluster achieved unprecedented efficiency, resulting in an impressive 124 MFlops per dollar.
The news about Dr. Hamada’s low-cost supercomputer comes at an interesting time for HPC in Japan. According to a recent PC World article, a Japanese government panel has recommended freezing funding for what could have been the largest supercomputer in the world, due to concerns about costs, which were projected to be more than $1 billion. Dr. Hamada’s work shows that with GPUs, you can get equal if not better performance than large, expensive and power-hungry CPU servers at a fraction of the cost, power and overall carbon footprint. And we’re seeing more and more examples of this every day from all over the world.
To see GPUs and the CUDA C development language be a part of a Gordon Bell award is an extremely significant milestone and shows that GPUs are truly having a powerful and transformative effect in the world of computing. We’d like to formally congratulate Dr. Hamada and his fellow researchers on their incredible achievement.