X-ScaleSolutions is pleased to announce the release of its HPL-DPU v2023.05, the DPU Optimized version of the High-Performance Linpack Code v2.3. 

This version of the HPL is enhanced to offload the communication to the NVIDIA BlueField DPU. The HPL-DPU can be run in two modes: Host mode and DPU mode with the MVAPICH2-DPU MPI library. In the DPU mode, communication offloading to DPU is enabled. In the Host mode, no such offloading occurs. This version of HPL is optimized in both Host mode and DPU mode. 

Using this version, we evaluated the performance of HPL on the HPC-AI cluster on 16 and 31 Xeon nodes with the MVAPICH2-DPU MPI library. As illustrated in Figure 1, the DPU-offloaded version of this HPL library accelerates HPL benchmark by up to 18% while being used in conjunction with the latest MVAPICH2-DPU 2023.05 version.

Figure 1: Capability of the MVAPICH2-DPU library to accelerate HPL benchmark performance.

For a trial license for the HPL-DPU code and how to modify your HPC and AI applications to take advantage of the MVAPICH2-DPU library, please contact X-ScaleSolutions at contactus@x-scalesolutions.com.