Deep learning(DL), High-Performance Computing(HPC), Big Data Analysis etc. has achieved remarkable progress in the recent years. To manage the heavy workloads, we need high-speed communication between CPU-GPU and GPU-GPU. At the same time, rapid enhancement of GPGPU programming is required but its so complicated. We know utilizing OpenACC and CUDA Unified Memory will be helpful to reduce programming complexity but it requires more cost between CPU and GPU.

So we choose IBM Power server with the full NVLink connectivity between CPU and GPU which efficiently reduces these costs as possible. It also has high bandwidth path between CPU and GPU than PCIe Gen3 x16.

Our main aim is to provide private cloud as in-house service within our company after integrating IBM Power server to our internal OpenStack environment. Then, comparison is made among IBM Power servers with GPUs and other x86_64 servers with GPUs (like NVIDIA DGX-1 servers)

Search Result

Schedule Matches

Speaker Matches