List your company

Matrix Multiplication Latency on Synthetic Matrix Multiplication Shapes

9.24Latency (µs)BWTA
5.663229.806653.9578.0934Dec 19, 2025Jan 5, 2026Jan 23, 2026Feb 10, 2026Feb 28, 2026Mar 18, 2026Apr 5, 2026

Evaluation results

MethodLinks
BWTAShape=W: [768, 768], A: [128, 768]2026.049.24
BWTAShape=W: [3072, 768], A: [128, 768]2026.0410.43
QTIP (r2)M=1, N=2048, K=20482025.1219.44
QUIP# (e8p)M=1, N=2048, K=20482025.1219.47
cuBLASM=8, N=2048, K=20482025.1219.79
cuBLASM=1, N=2048, K=20482025.1219.82
cuBLASM=4, N=2048, K=20482025.1219.99
AQLM (2x8)M=1, N=2048, K=20482025.1220.55
CodeGEMM (m1v4)M=1, N=2048, K=20482025.1220.66
CodeGEMM (m2v8)M=1, N=2048, K=20482025.1220.75
QUIP# (e8p)M=1, N=2048, K=81922025.1223.44
CodeGEMM (m1v4)M=1, N=2048, K=81922025.1223.81
QUIP# (e8p)M=1, N=4096, K=40962025.1223.96
CodeGEMM (m2v8)M=1, N=2048, K=81922025.1224.28
BWTAShape=W: [768, 3072], A: [128, 3072]2026.0424.38
CodeGEMM (m2v8)M=1, N=4096, K=40962025.1224.76
QTIP (r2)M=1, N=2048, K=81922025.1224.9
CodeGEMM (m1v4)M=1, N=4096, K=40962025.1224.97
QUIP# (e8p)M=1, N=8192, K=20482025.1225.52
CodeGEMM (m2v8)M=1, N=8192, K=20482025.1225.94
CodeGEMM (m1v4)M=1, N=8192, K=20482025.1226.7
QTIP (r2)M=1, N=4096, K=40962025.1226.74
QTIP (r2)M=1, N=8192, K=20482025.1227.08
cuBLASM=1, N=2048, K=81922025.1227.52
cuBLASM=1, N=4096, K=40962025.1228
cuBLASM=8, N=4096, K=40962025.1228.11
cuBLASM=4, N=4096, K=40962025.1228.54
cuBLASM=8, N=2048, K=81922025.1228.69
AQLM (2x8)M=1, N=8192, K=20482025.1228.83
AQLM (1x16)M=1, N=2048, K=20482025.1228.84
AQLM (1x16)M=1, N=8192, K=20482025.1228.84
cuBLASM=4, N=2048, K=81922025.1229.82
cuBLASM=1, N=8192, K=20482025.1230.57
AQLM (2x8)M=1, N=2048, K=81922025.1230.93
cuBLASM=4, N=8192, K=20482025.1231.31
cuBLASM=8, N=8192, K=20482025.1231.7
AQLM (2x8)M=1, N=4096, K=40962025.1232.28
CodeGEMM (m1v4)M=1, N=4096, K=143362025.1233.92
CodeGEMM (m1v4)M=1, N=8192, K=81922025.1235.45
QTIP (r2)M=4, N=2048, K=20482025.1236
CodeGEMM (m2v8)M=1, N=4096, K=143362025.1236.15
QUIP# (e8p)M=4, N=2048, K=20482025.1236.71
QUIP# (e8p)M=1, N=4096, K=143362025.1237.27
CodeGEMM (m1v4)M=1, N=14336, K=40962025.1237.51
CodeGEMM (m2v8)M=1, N=8192, K=81922025.1237.99
QUIP# (e8p)M=1, N=8192, K=81922025.1238.31
CodeGEMM (m2v8)M=1, N=14336, K=40962025.1238.85
QUIP# (e8p)M=1, N=14336, K=40962025.1238.91
cuBLASShape=W: [768, 768], A: [128, 768]2026.0441.32
CodeGEMM (m1v4)M=4, N=2048, K=20482025.1241.92
AQLM (2x8)M=4, N=2048, K=20482025.1243.31
QTIP (r2)M=1, N=4096, K=143362025.1243.85
CodeGEMM (m2v8)M=4, N=2048, K=20482025.1244.04
QTIP (r2)M=1, N=8192, K=81922025.1249.86
QTIP (r2)M=1, N=14336, K=40962025.1251.3
QUIP# (e8p)M=4, N=2048, K=81922025.1251.91
CodeGEMM (m1v4)M=4, N=2048, K=81922025.1252.57
QUIP# (e8p)M=4, N=4096, K=40962025.1253.92
CodeGEMM (m2v8)M=4, N=2048, K=81922025.1256.21
CodeGEMM (m1v4)M=4, N=4096, K=40962025.1257.79
QTIP (r2)M=8, N=2048, K=20482025.1257.87
AQLM (2x8)M=1, N=4096, K=143362025.1258.7
QTIP (r2)M=4, N=2048, K=81922025.1259.03
QUIP# (e8p)M=8, N=2048, K=20482025.1259.44
AQLM (1x16)M=1, N=2048, K=81922025.1260.47
CodeGEMM (m2v8)M=4, N=4096, K=40962025.1260.58
QUIP# (e8p)M=4, N=8192, K=20482025.1260.7
AQLM (2x8)M=1, N=8192, K=81922025.1262.5
QTIP (r2)M=4, N=4096, K=40962025.1262.74
AQLM (1x16)M=1, N=4096, K=40962025.1263.13
CodeGEMM (m2v8)M=4, N=8192, K=20482025.1263.97
AQLM (2x8)M=1, N=14336, K=40962025.1264.76
CodeGEMM (m1v4)M=4, N=8192, K=20482025.1265.36
QTIP (r2)M=4, N=8192, K=20482025.1266.18
CodeGEMM (m1v4)M=8, N=2048, K=20482025.1269.72
AQLM (2x8)M=8, N=2048, K=20482025.1273.03
AQLM (1x16)M=4, N=2048, K=20482025.1274.67
AQLM (1x16)M=4, N=8192, K=20482025.1274.67
CodeGEMM (m2v8)M=8, N=2048, K=20482025.1275.18
AQLM (2x8)M=4, N=8192, K=20482025.1276.15
CodeGEMM (m1v4)M=1, N=8192, K=286722025.1276.5
CodeGEMM (m1v4)M=1, N=28672, K=81922025.1276.71
cuBLASShape=W: [3072, 768], A: [128, 768]2026.0480.17
AQLM (2x8)M=4, N=2048, K=81922025.1282.18
CodeGEMM (m2v8)M=1, N=8192, K=286722025.1286.2
cuBLASM=1, N=4096, K=143362025.1286.31
CodeGEMM (m2v8)M=1, N=28672, K=81922025.1286.48
cuBLASM=8, N=4096, K=143362025.1286.49
cuBLASM=4, N=4096, K=143362025.1286.51
cuBLASM=1, N=14336, K=40962025.1288.67
cuBLASM=4, N=14336, K=40962025.1289.08
cuBLASM=8, N=14336, K=40962025.1289.29
AQLM (2x8)M=4, N=4096, K=40962025.1289.76
QUIP# (e8p)M=8, N=2048, K=81922025.1289.91
CodeGEMM (m1v4)M=8, N=2048, K=81922025.1290.73
CodeGEMM (m1v4)M=4, N=4096, K=143362025.1292.61
QUIP# (e8p)M=8, N=4096, K=40962025.1293.43
cuBLASM=8, N=8192, K=81922025.1295.45
cuBLASM=1, N=8192, K=81922025.1296.4
CodeGEMM (m1v4)M=4, N=8192, K=81922025.1298.66