Technology1 min read

Show HN: 1.499x speedup and 28.5% dynamic energy drop on A40 via BF16 tiling

By · Published by Everything Blog

In short

Kristian Lindholm and the AI Orchestra achieved notable performance and power improvements by applying BF16 tiling to simulated Ising grid workloads on NVIDIA hardware. Tests across standard and scaled grid dimensions revealed consistent decreases in runtime alongside more than a quarter reduction in energy use per sweep. Container permission restrictions on the test machine prevented direct hardware counter profiling, forcing researchers to infer expected memory bandwidth savings from data-type size reductions. All results underwent rigorous accuracy verification, showing identical state matching across every sub-lattice phase.

Key points

  • Validation confirmed exact numerical parity with zero site disagreements across all six s…: Validation confirmed exact numerical parity with zero site disagreements across all six sub-lattice phases compared to reference states.
  • Container security restrictions blocked hardware performance counters, requiring memory t…: Container security restrictions blocked hardware performance counters, requiring memory traffic reductions to be inferred from data footprint changes.
  • The methodology and performance receipts are strictly restricted from being deployed in m…: The methodology and performance receipts are strictly restricted from being deployed in military, targeting, or kinetic weapon applications.

Sources checked by our analysis pipeline: 1

Raw GPU benchmark outputs and dense context logic boot files for multi-agent execution loops.

sm_86

): 1.499x speedup at L=4608 (10.204 ms vs 15.291 ms), 28.5% dynamic energy reduction (186.735 vs 261.164 mJ/sweep), and 0 site disagreements across all 6 sub-lattice phases.nsys

) kernel execution and CUDA API timing summary on NVIDIA A40 (2026-11-16).

Authors: Kristian Lindholm and the AI Orchestra (Coldwater Tractors Inc)

Date: November 16, 2026 | Architecture: NVIDIA A40 (Ampere sm_86

, 48 GB GDDR6, 300 W TDP)

Use Restriction: Receipts and methodology open for civilian, academic, and commercial research. Strictly prohibited for kinetic, military, weapons, or targeting applications.

Prior to compiling or renting GPU compute, every architectural and numerical claim was subjected to a multi-model adversarial falsification loop ("Red-Team Audit"):

L ∈ {384, 768, 4608, 9216}

; updated from 4096/8192

via Patch 5 so L

is divisible by 3 × TILE = 96

for seamless 3-color torus wrapping), physics parameters (2D 7-point hexagonal Ising lattice, J = +1

, T = 3.0

, stateless PCG32 PRNG), and pass/fail gates were locked before first execution.FP64 std::exp

)…

Technology