2 papers
cs.DC2026
Sketching the Error, Not the Product: Post Hoc Fault Recovery for Half Precision GPU Matrix Multiplication
Pranav Napolean, Vikas Srivastava, Napolean Periathambi
Silent data corruption (SDC) from defective accelerators now interrupts large scale training, yet deployed mitigations act on whole nodes. Algorithm based fault tolerance (ABFT) fo…
cs.DC2026
Syndrome Decoding for Silent Data Corruption in Quantized Integer GPU Arithmetic
Pranav Napolean, Vikas Srivastava, Napolean Periathambi
Quantized neural network inference runs integer matrix multiplications on GPU tensor cores, and the INT32 accumulators inside those cores have neither parity nor ECC. A transient f…