{"id":"W1979527452","doi":"10.1109/ispass.2009.4919648","title":"Analyzing CUDA workloads using a detailed GPU simulator","year":2009,"lang":"en","type":"article","venue":"","topic":"Parallel Computing and Optimization Techniques","field":"Computer Science","cited_by":1660,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Parallel computing; CUDA; Thread (computing); Graphics; Microarchitecture; Instruction set; SIMD; Computer architecture; General-purpose computing on graphics processing units; Memory bandwidth; Multithreading; Supercomputer; Operating system","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005379987,0.0008084078,0.00053406,0.0006729025,0.0004316245,0.0006249106,0.00151136,0.0006195094,0.001330086],"category_scores_gemma":[0.002192661,0.0004809079,0.000499506,0.001314879,0.0003567201,0.0006161944,0.0003294849,0.0008432375,0.0002857895],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001320827,"about_ca_system_score_gemma":0.001165236,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02048216,"about_ca_topic_score_gemma":0.01428069,"domain_scores_codex":[0.9995683,0.00008877058,0.00003975605,0.00007537455,0.0001460825,0.00008168029],"domain_scores_gemma":[0.9985818,0.0005752367,0.00009281106,0.0002633093,0.0004082962,0.00007852995],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002576029,0.0003314215,0.01565147,0.000123761,0.0001081518,0.0002679022,0.0003078566,0.945747,0.01671453,0.002134364,0.004054589,0.01430145],"study_design_scores_gemma":[0.00002274355,0.00006100013,0.002325328,0.000002978299,0.000006733434,0.00001591809,0.00003017083,0.9930692,0.003379627,0.0002046631,0.0008720754,0.00000955852],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9684415,0.000114251,0.0203543,0.0001216081,0.00004239429,0.0001924964,0.002591847,0.003436787,0.004704875],"genre_scores_gemma":[0.9685688,0.0001436689,0.02638294,0.00006627198,0.000008342399,0.0001795059,0.003241042,0.0003179596,0.001091493],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02048216,"threshold_uncertainty_score":0.04072589,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02341184931854888,"score_gpt":0.2874748231294658,"score_spread":0.2640629738109169,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}