SparkleDock slashes molecular docking time 100x on GPUs
New SparkleDock framework achieves 9.7x speedup on A100 GPUs and 2 orders of magnitude at scale
A team from the Chinese Academy of Sciences and Japan's RIKEN institute has developed SparkleDock, a breakthrough framework for high-fidelity molecular docking that leverages GPU supercomputing to achieve unprecedented speedups. The system addresses a major bottleneck in biomolecular research by redesigning Glowworm Swarm Optimization (GSO) to expose fine-grained parallelism at the agent level, while restructuring energy scoring computations into Tensor Core-friendly operations. This enables efficient processing of irregular pairwise interactions through structured matrix operations.
The results are dramatic: SparkleDock delivers 9.7x faster performance than LightDock on NVIDIA A100 GPUs and 18.9x on H100 GPUs. At scale, running on 512 GPUs, it reduces docking time from hours to just seconds—enabling large-scale virtual screening that was previously impractical with flexible docking approaches. The team achieved this through a performance-model-driven scheduling system that handles load balancing and out-of-core scaling across multiple GPUs, fundamentally changing how researchers can approach drug discovery and protein interaction studies.
- SparkleDock achieves 9.7x-18.9x speedup over LightDock on single A100/H100 GPUs
- Redesigned GSO with Tensor Core-compatible energy scoring enables efficient irregular computation
- On 512 GPUs, reduces docking time from hours to seconds—enabling large-scale virtual screening
Why It Matters
Accelerates drug discovery pipelines by 100x, potentially cutting years off development timelines for new therapies