Big data and machine learning are major factors shaping research and innovation now and will continue to be so in the foreseeable future. Deep learning represents the state-of-the-art in machine learning and data analysis.
The NVIDIA DGX-1 is a supercomputer specifically designed for deep learning. It incorporates two Intel Xeon E5-2698 v3 CPUs, each with 16 cores and taking advantage of the Haswell architecture. It also utilizes 512 Gigabytes (GB) DDR4 memory for the system. The DGX-1 includes 8 Tesla P100 GPUs utilizing the new Pascal architecture. Each GPU has 16 GB of GPU memory which totals to 128 GB of GPU memory. Each GPU has 3584 CUDA cores and the whole system consists of 28672 CUDA cores which provides 170 TFLOPS of performance for floating point operations. Its acquisition provides two major benefits to machine learning research at the University of Utah. First, it will allow state-of-the-art models to be developed and second, it will accelerate experiments allowing researchers to test multiple models and generate results.
Funding for the Nvidia DGX-1 was provided by:
- Research Instrumentation Fund-Core Facilities Competition Award from the VPR
- School of Computing
- Tolga Tasdizen
- SCI Institute
- NVIDIA Cuda Center of Excellence Award