FP8 vision attention QKV projection
FP8-quantized QKV projection for vision transformer attention. Projects vision hidden states to query, key, and value tensors using a single fused linear layer with blockwise FP8 quantization (BlockWise1x128 for activations, BlockWise128x128 for weights).
Reported evidence · last observed 2026-08-19. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencyhead_dim = 96 · num_heads = 16 · hidden_size = 1536 · qkv_out_features = 460816 cases
Not measured on H100 for this workload. Challenges →
129.5µs74% SOL1.00×Reported · license unknown · no source2026-08-19
Measured exactly what you asked. Reported by source; not independently reproduced.
234.5µs70% SOL1.17×Reported · license unknown · no source2026-08-03
Measured exactly what you asked. Reported by source; not independently reproduced.
335.2µs70% SOL1.19×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
435.8µs69% SOL1.21×Reported · license unknown · no source2026-08-21
Measured exactly what you asked. Reported by source; not independently reproduced.
536.8µs69% SOL1.25×Reported · license unknown · no source2026-08-19
Measured exactly what you asked. Reported by source; not independently reproduced.
637.5µs68% SOL1.27×Reported · license unknown · no source2026-07-18
Measured exactly what you asked. Reported by source; not independently reproduced.
738.6µs68% SOL1.31×Reported · license unknown · no source2026-08-18
Measured exactly what you asked. Reported by source; not independently reproduced.
839.3µs67% SOL1.33×Reported · license unknown · no source2026-07-21
Measured exactly what you asked. Reported by source; not independently reproduced.
939.5µs67% SOL1.34×Reported · license unknown · no source2026-08-03
Measured exactly what you asked. Reported by source; not independently reproduced.
1039.6µs67% SOL1.34×Reported · license unknown · no source2026-08-11
Measured exactly what you asked. Reported by source; not independently reproduced.
1139.9µs67% SOL1.35×Reported · license unknown · no source2026-08-01
Measured exactly what you asked. Reported by source; not independently reproduced.
1259.7µs56% SOL2.02×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1360.2µs56% SOL2.04×Reported · license unknown · no source2026-08-23
Measured exactly what you asked. Reported by source; not independently reproduced.
1462.4µs55% SOL2.12×Reported · license unknown · no source2026-07-23
Measured exactly what you asked. Reported by source; not independently reproduced.
1571.0µs51% SOL2.40×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1682.2µs47% SOL2.78×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1792.5µs44% SOL3.13×Reported · license unknown · no source2026-07-23
Measured exactly what you asked. Reported by source; not independently reproduced.