Oracle Cloud has made NVIDIA’s latest generation of A100 Tensor Core GPUs available on its cloud infrastructure, which enables Rocky DEM users to further accelerate solving complex and larger problems involving particle dynamics.
A recent performance benchmark done by the ESSS team on Oracle’s Second-Generation Cloud, using newly available Oracle Cloud Infrastructure (OCI) GPU 4.8 machines powered by NVIDIA A100 GPUs, has shown more than 40 percent performance gains while running on a single GPU, compared with the previous generation V100.
GPU-powered applications, such as Rocky on Oracle Cloud, offer companies more time to start thinking about engineering and designing their next product rather than worrying about simulation runtimes or computing resource availability.Karan Batta, Vice President of Oracle Cloud Infrastructure
In addition to the more than 40-percent performance gain (a figure statistically significant in just a single-generation advance), the setup is powered by 40 GB of GPU RAM, versus 16 GB per board (more than twice the previous generation, which is critical for large cases). But perhaps the biggest gain from the OCI GPU 4.8 is when progressing into multi-GPU simulations.
This higher communication bandwidth between all GPUs, which results in lower latency, has proven to be key in achieving better scalability as more GPUs are added. The benefit to Rocky DEM users is the critical performance gains that result from combined processors available in each of the NVIDIA GPUs — as well as from stacking the memory from multiple GPUs to run cases that do not fit into a single GPU.Clovis Maliska Jr., President of ESSS
The NVIDIA A100 brings unprecedented AI and HPC performance to cloud computing, giving developers easy access to world-class compute capacity to accelerate their time to insight. Using A100 on Oracle Cloud, the ESSS’s Rocky DEM team will be able to unlock next-level discoveries.Paresh Kharya, Senior Director of Product Management and Marketing for Accelerated Computing at NVIDIA
Both plots below show the simulation performance comparison between NVIDIA A100 Tensor Core GPU and NVIDIA V100 Tensor Core GPU on the Oracle Cloud Infrastructure relative to eight CPU cores of an Intel Xeon Gold 6230 @ 2.1GHz. The case is a rotating drum with polyhedral particles, with both particle and GPU numbers varying to show scalability. Best performance is achieved by 4 x A100 configuration, with performance equivalent to more than 750 CPU cores.