Scaling up HBM Efficiency of Top-K SpMV for Approximate Embedding Similarity on FPGAs
This improves similarity-search performance for applications like recommendation systems, though it is incremental as it adapts existing FPGA techniques to a specific bottleneck.
The paper tackled the inefficiency of Top-K SpMV for similarity-search on sparse embeddings in general-purpose systems by introducing an FPGA design using reduced precision and novel packet-wise CSR compression, achieving 100x speedup over CPU, 2x over GPU, and 14.2x higher power-efficiency.
Top-K SpMV is a key component of similarity-search on sparse embeddings. This sparse workload does not perform well on general-purpose NUMA systems that employ traditional caching strategies. Instead, modern FPGA accelerator cards have a few tricks up their sleeve. We introduce a Top-K SpMV FPGA design that leverages reduced precision and a novel packet-wise CSR matrix compression, enabling custom data layouts and delivering bandwidth efficiency often unreachable even in architectures with higher peak bandwidth. With HBM-based boards, we are 100x faster than a multi-threaded CPU implementation and 2x faster than a GPU with 20% higher bandwidth, with 14.2x higher power-efficiency.