While Intel's Habana Gaudi offers somewhat competitive performance and comes with the Habana SynapseAI software package, it still falls short compared to Nvidia's CUDA-enabled compute GPUs. This, paired with limited availability, is why Gaudi hasn't been as popular for large language models (LLMs) like ChatGPT.
Now that the AI rush is on, Intel's Habana is seeing broader deployments. Amazon Web Services decided to try Intel's 1st Generation Gaudi with PyTorch and DeepSpeed






