ResearchTasksPaged Decode Attention ServingPaged Decode Attention ServingFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedBimodal B4 TraceFlashInfer1GPU Utilization Ratio1Jun 26, 2026Zipf B8 tracePersistentKV1.157GPU Ratio1Jun 26, 2026Uniform B8 tracePersistentKV1.053GPU Ratio1Jun 26, 2026Bimodal B8 tracePersistentKV1.299GPU Ratio1Jun 26, 2026Bucketed B1 tracePersistentKV1.576GPU Ratio1Jun 26, 2026