B200 2B fp16 — complete.

Metric Value

N 2,000,000,000

GPU NVIDIA B200 (191.5 GB)

Build 1794.2s (Pass 1: 637s, Pass 2: 1157s)

Serving 132.00 GB (Z=[2B, 32] fp16)

Compression 46.5× fp64, 23.3× fp32

Query p50 60.89 ms

Query p99 62.58 ms

R@10 mean 98.0%

R@10 min 90.0%

VRAM serving 131 GB

VRAM query peak 142 GB

CPU RAM post 9 GB

Total wall 2693.6s (~45 min)