NVIDIA is determined to explore the open-sourcing route, as the firm now open-sources its HPCG benchmark package, targeted at AI compute.
NVIDIA's HPCG Gets Open-Sourced By The Firm As They Reiterate Their Commitment Towards The Collaborative Approach & AI Optimizations
Well, NVIDIA looks to shift to a more open-source nature since the firm previously announced the use of an open-source GPU kernel for the GeForce RTX 20 series and beyond in an attempt to promote development at their platform along with making a stride at Linux when it comes to having open-source assets. This time, the firm has decided to aid the AI industry by open-sourcing the NVIDIA HPCG, which is targeted towards AI hardware such as the Grace Superchips and the Hopper series architecture.
#NVIDIA #HPCG is now open-source!! 🔥https://t.co/AeD10moKaX
Kudos to Mohammad Almasri🟢 CPU-only (Grace, x86)
🟢 GPU-only (Ampere, Hopper)
🟢 Heterogeneous Grace Hopper superchip
🟢 Entirely relies on cuSPARSE and NVPL Sparse for extensive ops
🟢 Fully configurable— Federico Busato 🇺🇦 (@fedebusato) May 31, 2024
For those unaware, NVIDIA's HPCG is a benchmark designed for HPC applications that focuses on multiple mathematical operations to test out the performance of specific hardware across different scenarios. This not only allows Team Green to showcase the capabilities of their GPU, but the software package has played a significant role in the optimization of HPC systems. In the era where AI and HPC applications have seen a considerable rise, the open-sourcing of NVIDIA's HPCG means that the company is determined to see developments with the resource. Here are the main features, as detailed by the firm themselves (via Github):
Main Features
The NVIDIA HPCG benchmark exploits NVIDIA high-performance math libraries: cuSPARSE and NVPL Sparse to achieve the highest possible performance for Sparse Matrix-vector multiplication (SpMV) and Sparse Matrix triangular solvers (SpSV) on NVIDIA GPUs and Grace CPUs.
The NVIDIA HPCG benchmark supports highly configurable command line parameters to decide:
- Problem sizes for the GPU and Grace CPU
- 3D rank grid shape
- Execution modes: CPU-only, GPU-only and heterogeneous
- Point-to-point communication: MPI_Host (Send/Recv), MPI_Host_Alltoallv, MPI_CUDA_Aware, MPI_CUDA_Aware_Alltoallv, and NCCL
- NUMA-related configurations
Regarding compatibility, NVIDIA's HPCG is supported at Grace CPU systems along with Ampere and Hopper GPU architectures. The software only works with Linux as well, which is something that limits its scope. However, it's still an interesting move by NVIDIA, and it shows their commitment to open sourcing of assets, we do similar developments from the firm in the future as well.
About the author: Muhammad Zuhair is a hardware and technology reporter for Wccftech, specializing in the semiconductor industry and the complex interplay between technology, manufacturing, and geopolitics. His coverage focuses on the corporate strategies and technological roadmaps of industry giants like TSMC, NVIDIA, Samsung, and Intel. Zuhair's expertise lies in deconstructing complex topics such as fabrication nodes (e.g., 2nm process), the economic impact of policies like the CHIPS Act, and the strategic development of AI infrastructure from NVIDIA, AMD and Intel.
Follow Wccftech on Google to get more of our news coverage in your feeds.
