1.4. Requirements for the GPU Accelerator in the Mechanical APDL Application

Your system must meet the following requirements to use the GPU accelerator capability in Mechanical APDL. For information on the most recently tested GPU cards, see the GPU Accelerator Capabilities PDF on the Platform Support section of the Ansys Website .

  • The machine(s) being used for the simulation must contain at least one GPU card.

  • A minimum 16GB of on-card memory is recommended in order to achieve meaningful acceleration in simulations that can use the GPU card.

  • To achieve optimal performance, only GPU cards with significant double precision performance (FP64) are recommended for use with the sparse direct solver and eigensolvers based on the sparse solver (for example, Block Lanczos or subspace). The following cards are recommended:

     GPU Information[a]Recommended Equation Solvers
    Card[b]Release YearFP64 and FP32 Performance (TFLOPS)Memory Capacity (GB)Memory Bandwidth (GB/sec)Sparse (Direct)Iterative (PCG, etc.)Mixed
    Data Center Cards
    NVIDIA B300[c]202575/1.2270 GB7,700 GB/sYYY
    NVIDIA B200202580/40186 GB7,700 GB/sYYY
    NVIDIA B1002025—[d]—[d]—[d]YYY
    NVIDIA H200202434/67141 GB4,800 GB/sYYY
    NVIDIA H100202230/6094 GB3,900 GB/sYYY
    NVIDIA L402022 1.4/9048 GB864 GB/sNYY
    NVIDIA A3020215.2/10.324 GB933 GB/sYYY
    NVIDIA A10020209.7/19.580 GB1,940 GB/sYYY
    NVIDIA A4020200.58/3748 GB696 GB/sNYY
    Professional Workstation Cards
    AMD Radeon AI PRO R970020251.5/47.832 GB640 GB/sNYY
    AMD Ryzen AI Max+ PRO 39520250.5/14.5128 GB256 GB/sNYY
    AMD Ryzen AI Max+ 39520250.5/14.5128 GB256 GB/sNYY
    AMD Radeon PRO W790020231.9/6148 GB864 GB/sNYY
    NVIDIA RTX PRO 600020251.9/12696 GB1,340 GB/sNYY
    NVIDIA RTX PRO 500020251/66.548 GB1,790 GB/sNYY
    NVIDIA A80020249.7/19.540 GB1,500 GB/sYYY
    NVIDIA RTX 5880 Ada2024 1.08/6948 GB 864 GB/sNYY
    NVIDIA RTX 5000 Ada20231.0/6532 GB576 GB/sNYY
    NVIDIA RTX 4500 Ada20230.6/3920 GB640 GB/sNYY
    NVIDIA RTX 6000 Ada20221.5/9148 GB960 GB/sNYY
    NVIDIA RTX A550020220.5/3424 GB768 GB/sNYY
    NVIDIA A1620210.1/4.516 GB200 GB/sNYY
    NVIDIA A1020211/3124 GB600 GB/sNYY
    NVIDIA RTX A500020210.5/2724 GB768 GB/sNYY
    NVIDIA RTX A450020210.3/2324 GB640 GB/sNYY
    NVIDIA RTX A400020210.3/1916 GB448 GB/sNYY
    NVIDIA RTX A600020200.6/3848 GB768 GB/sNYY
    Consumer Cards[e]
    NVIDIA GeForce RTX 5090 20251.6/10432 GB1790 GB/sNYY
    NVIDIA GeForce RTX 50802025 0.8/5616 GB960 GB/sNYY
    NVIDIA GeForce RTX 5070 Ti2025 0.7/4416 GB896 GB/sNYY

    [a] The hardware specifications in this table are based on publicly available product datasheets and vendor documentation. Specifications may change or contain inaccuracies. Verify all details with the official manufacturer resources before making decisions.

    [b] GPUs that are unreleased at the time of this software release are not listed in this table. These GPUs may still be compatible with this release, but their performance and functionality have not been validated.

    [c] Hardware specifications for NVIDIA B300 are sourced from the official NVIDIA product datasheet. For details, see https://resources.nvidia.com/en-us-dgx-systems/dgx-b300-datasheet.

    [d] The vendor has not declared the specification for this entry.

    [e] Consumer GPUs with 16 GB of memory offer limited performance for large-scale workloads. Their fixed memory capacity restricts acceleration compared to workstation or data center GPUs with higher memory configurations.

  • For NVIDIA GPU cards, note the following:

    1. The Mechanical APDL software installation provides the necessary CUDA 13.0.2 libraries to accelerate the FEA solution. No separate installation of CUDA is required.

    2. The GPU driver is not installed by Mechanical APDL. The driver version must be 581.42 or newer. For optimal performance on Windows, the TCC (Tesla Compute Cluster) driver mode is recommended when using Tesla series GPU cards. Some limitations exist when using this driver mode. Check your GPU card documentation for more details on how to set this driver mode and the existing limitations.

    3. The CUDA 13.0.2 libraries only support Visual Studio 2022 Professional.

  • For AMD GPU cards, note the following:

    1. The Mechanical APDL installation provides the necessary HIP/ROCm 6.4.2 libraries to accelerate the FEA solution. No separate installation of HIP/ROCm is required. Note that your machine must have at least SUSE Linux Enterprise Server 15 SP5 to use the ROCm libraries.

    2. The GPU driver is not installed by Mechanical APDL. The driver version must be 6.4.2 or newer. Some instructions are given below.

  • To utilize a GPU device that is not on the recommended list of cards, set the following environment variable:

    ANSGPU_OVERRIDE=1

    This is most beneficial when you wish to run on newer GPUs that were not available at the time of release of this version of the Ansys program. If you choose to use this environment variable, you should ensure that the GPU device that you wish to use is sufficiently powerful, in terms of both double-precision compute power and on-card memory, to achieve meaningful acceleration for your simulation. Using this environment variable with an underpowered CPU may actually decelerate your simulation.

    Support for AMD GPUs released after a specific version of the Mechanical APDL application is limited due to incompatibility between required software libraries and the GPU hardware. To check which GPU architectures are supported, refer to the AMD compatibility matrix version. You can find the GPU architecture in the device’s specification sheet.

  • For details on HPC licensing, see HPC Licensing in the Parallel Processing Guide.

  • For details on GPU benchmarks, see Ansys Mechanical Benchmarks.

  • For more information about using GPUs with the Mechanical APDL application, see the following resources on the Ansys website and innovation space:

Note:  On Windows, the use of Remote Desktop may disable the use of a GPU device. Launching Mechanical APDL through the Ansys Remote Solve Manager (RSM) when RSM is installed as a service may also disable the use of a GPU. In these two scenarios, the GPU accelerator capability cannot be used. Using the TCC (Tesla Compute Cluster) driver mode, if applicable, can circumvent this restriction.