Spark itself does not use GPUs, and is agnostic to what GPUs exist on a cluster, scheduled by the resource manager, and used by an application. In practice, virtually all GPU-related use cases (for deep learning for example) use CUDA, and this is NVIDIA-specific. Certainly, RAPIDS is from NVIDIA.
On Wed, Feb 16, 2022 at 7:03 AM 15927907...@163.com <15927907...@163.com> wrote: > Hello, > We have done some Spark GPU accelerated work using the spark-rapids > component(https://github.com/NVIDIA/spark-rapids). However, we found that > this component currently only supports Nvidia GPU, and on the official > Spark website, we did not see the manufacturer's description of the GPU > supported by spark( > https://spark.apache.org/docs/3.2.1/configuration.html#custom-resource-scheduling-and-configuration-overview). > So, Can Spark also support GPUs from other manufacturers? such as AMD. > Looking forward to your reply. > > ------------------------------ > 15927907...@163.com >