High-performance compression
Error-bounded CPU and GPU compressors target extreme throughput, high compression ratios, and scalable data reduction.
ZCCL advances high-performance compression and communication for AI and scientific applications, with co-designed systems that process and move data efficiently.
Our work advances each area independently and brings them together in end-to-end systems for AI and scientific workloads.
Error-bounded CPU and GPU compressors target extreme throughput, high compression ratios, and scalable data reduction.
Collective communication methods reduce data movement and support large distributed AI and scientific workloads.
Compressed-domain and homomorphic techniques connect both areas to eliminate avoidable work.
FSZ advances high-throughput GPU lossy compression, while hZCCL2 co-designs vectorized homomorphic compression with collective communication.
Explore publicationsCompression-accelerated collective communication for AI and scientific applications.
View repository → CPU compressionAn extremely fast error-bounded lossy compressor with OpenMP support.
View repository → GPU compressionA GPU compressor designed for high throughput and high compression ratios.
View repository →