HetCCL: Accelerating LLM Training with Heterogeneous GPUs
This addresses a practical bottleneck for organizations expanding GPU clusters with mixed hardware, enabling cost-effective and high-performance training without application changes.
The paper tackles the problem of inefficient collective communication across heterogeneous GPUs from multiple vendors in large language model training, presenting HetCCL, a library that enables RDMA-based communication without driver modifications and matches homogeneous performance while scaling in heterogeneous environments.
The rapid growth of large language models is driving organizations to expand their GPU clusters, often with GPUs from multiple vendors. However, current deep learning frameworks lack support for collective communication across heterogeneous GPUs, leading to inefficiency and higher costs. We present HetCCL, a collective communication library that unifies vendor-specific backends and enables RDMA-based communication across GPUs without requiring driver modifications. HetCCL introduces two novel mechanisms that enable cross-vendor communication while leveraging optimized vendor libraries, NVIDIA NCCL and AMD RCCL. Evaluations on a multi-vendor GPU cluster show that HetCCL matches NCCL and RCCL performance in homogeneous setups while uniquely scaling in heterogeneous environments, enabling practical, high-performance training with both NVIDIA and AMD GPUs without changes to existing deep learning applications.