In a recent press release, AMAX, “a leading manufacturer of high-performance servers for AI and Deep Learning development, announced the launch of the DL-E48A, a reconfigurable single/dual root high-density GPU platform designed for AI training and inference. AMAX will be showcasing the highly anticipated solution at CVPR 2018 (Booth #304), to be held at Salt Palace Convention Center in Salt Lake City, Utah, June 19-21, 2018.”
“The DL-E48A is a high-performance, high-density 8x GPU server platform with first-in-industry remote-configurable single and dual PCIe root complex architectures, enabling ease of hardware optimization for AI and DL training, inference, HPC compute, rendering and virtualization. The platform supports the latest generation of NVIDIA Tesla Volta V100 32GB HBM2 GPU, Mellanox EDR Infiniband and Intel Omnipath fabric to process largest datasets for the most demanding AI and DL applications. Utilizing Intel’s Scalable Processor Architecture, the system provides a 56 percent increase in memory bandwidth and a 54 percent bandwidth increase for communication between CPUs (UPI) compared to platforms featuring the previous Intel Processor generation. NVME storage options are available with the DL-E48A, removing storage bottlenecks for ultra-fast data performance.”
“The DL-E48A’s software-configurable PCIe architecture allows IT organizations to remotely change the PCIe hardware topology between single and dual root complex architectures in order to optimize machine and cluster performance for specific workloads. Single root complex architectures are ideal for reducing GPU to GPU memory copy latency for improved performance of GPU-intensive DL workloads.”
Read more at PR Newswire.
Photo credit: AMAX