IntelligentHQ
IntelligentHQ
Skip to content

High-Performance Computing & Research

Accelerated computing for science and engineering: the HPC SDK compilers and libraries, CUDA-X, CUDA-Q for quantum research, PhysicsNeMo for physics AI, and Grace CPUs with HGX and DGX systems.

Technology profiles for this category are in research.

Overview

Research computing centers run long simulations in climate, chemistry, materials and engineering, often in Fortran and C++ code built up over decades. Porting and scaling that code is the main task in this category.

The HPC SDK provides the NVC, NVC++ and NVFORTRAN compilers with OpenACC, CUDA Fortran and standard C++ parallel algorithms, plus cuBLAS, cuFFT and other math libraries, a CUDA-aware Open MPI, NCCL, NVSHMEM and the Nsight profilers. CUDA-Q is an open-source platform for hybrid quantum and classical programs that can simulate circuits on GPUs or run on quantum processors. For hardware, NVIDIA lists HGX and DGX platforms and Grace CPUs for HPC; the Grace CPU Superchip joins two CPUs over NVLink-C2C.

Codes dominated by serial logic gain little from GPUs. Profile before porting, and check the GPU Apps Catalog in case a GPU-ready version of your application already exists.1234

Problems it addresses

  • Porting legacy Fortran1

    NVFORTRAN supports OpenACC directives and CUDA Fortran, so code can move to GPUs step by step.

  • Scaling across nodes1

    A CUDA-aware Open MPI, NCCL and NVSHMEM handle communication between GPUs and nodes.

  • Quantum research without hardware2

    CUDA-Q GPU simulators let researchers develop and benchmark circuits without a quantum processor.

  • Costly repeated simulations5

    PhysicsNeMo trains surrogate models that approximate expensive solvers.

A typical workflow

  1. Profile the CPU code6

    Find the loops that dominate run time before porting anything.

  2. Port step by step1

    Add OpenACC directives or standard parallel algorithms and compile with NVFORTRAN or NVC++.

  3. Call tuned libraries1

    Replace hand-written math with cuBLAS, cuFFT, cuSOLVER or cuSPARSE.

  4. Scale out1

    Use CUDA-aware MPI and NCCL across nodes.

  5. Profile again1

    Check GPU use with Nsight Systems and Nsight Compute, both included in the HPC SDK.

AI Factory Efficiency Lab

Model token and infrastructure costs for your own numbers.

Open the lab

Next steps

  1. Check the GPU Apps Catalog for an existing GPU version of your application.

  2. Profile one representative run to find the loops worth porting.

  3. Estimate GPU count and power for a planned simulation campaign in the AI Factory Efficiency Lab.

Sources

  1. NVIDIA HPC SDK (opens in a new tab)NVIDIA · Vendor-reported
  2. NVIDIA CUDA-Q (opens in a new tab)NVIDIA · Vendor-reported
  3. NVIDIA High-Performance Computing (opens in a new tab)NVIDIA · Vendor-reported
  4. NVIDIA Grace CPU (opens in a new tab)NVIDIA · Vendor-reported
  5. NVIDIA PhysicsNeMo (opens in a new tab)NVIDIA · Vendor-reported
  6. NVIDIA developer tools overview (opens in a new tab)NVIDIA · Vendor-reported

Fill out the form below to request your copy.

Name(Required)