NVFP4 attention output projection with residual
FP4-quantized attention output projection followed by residual addition. Takes attention output (batch, seq_len, num_heads * head_dim), projects through FP4-quantized linear layer (o_proj), and adds residual connection. Uses NVFP4 block-scaled GEMM with 1x16 block scaling.
Reported evidence · last observed 2026-08-14. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencyhead_dim = 128 · proj_dim = 8192 · hidden_size = 8192 · sf_vec_size = 16 · proj_dim_half = 4096 · num_proj_blocks = 512 · num_attention_heads = 6416 cases
Not measured on H100 for this workload. Challenges →
166.5µs91% SOL1.00×Reported · license unknown · no source2026-08-14
Measured exactly what you asked. Reported by source; not independently reproduced.
266.5µs91% SOL1.00×Reported · license unknown · no source2026-07-23
Measured exactly what you asked. Reported by source; not independently reproduced.
376.7µs89% SOL1.15×Reported · license unknown · no source2026-07-13
Measured exactly what you asked. Reported by source; not independently reproduced.
476.7µs90% SOL1.15×Reported · license unknown · no source2026-08-17
Measured exactly what you asked. Reported by source; not independently reproduced.
577.8µs90% SOL1.17×Reported · license unknown · no source2026-08-01
Measured exactly what you asked. Reported by source; not independently reproduced.
678.0µs89% SOL1.17×Reported · license unknown · no source2026-07-19
Measured exactly what you asked. Reported by source; not independently reproduced.
782.0µs89% SOL1.23×Reported · license unknown · no source2026-07-13
Measured exactly what you asked. Reported by source; not independently reproduced.
8105.6µs86% SOL1.59×Reported · license unknown · no source2026-08-11
Measured exactly what you asked. Reported by source; not independently reproduced.
9110.2µs86% SOL1.66×Reported · license unknown · no source2026-08-04
Measured exactly what you asked. Reported by source; not independently reproduced.
10111.5µs85% SOL1.68×Reported · license unknown · no source2026-08-04
Measured exactly what you asked. Reported by source; not independently reproduced.
11131.3µs83% SOL1.98×Reported · license unknown · no source2026-07-13
Measured exactly what you asked. Reported by source; not independently reproduced.
12139.2µs83% SOL2.09×Reported · license unknown · no source2026-07-31
Measured exactly what you asked. Reported by source; not independently reproduced.
13314.9µs65% SOL4.74×Reported · license unknown · no source2026-07-22
Measured exactly what you asked. Reported by source; not independently reproduced.
14315.1µs65% SOL4.74×Reported · license unknown · no source2026-07-22
Measured exactly what you asked. Reported by source; not independently reproduced.
15513.7µs54% SOL7.73×Reported · license unknown · no source2026-07-13
Measured exactly what you asked. Reported by source; not independently reproduced.
161.01ms38% SOL15.27×Reported · license unknown · no source2026-07-13
Measured exactly what you asked. Reported by source; not independently reproduced.
171.70ms27% SOL25.60×Reported · license unknown · no source2026-08-10
Measured exactly what you asked. Reported by source; not independently reproduced.