cd /news/ai-infrastructure/show-hn-1-499x-speedup-and-28-5-dyna… · home › topics › ai-infrastructure › article
[ARTICLE · art-172859] src=cwtractors.com ↗ pub= topic=ai-infrastructure verified=true sentiment=↑ positive

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

A BF16-tiled kernel on the NVIDIA A40 delivered a 1.499x speedup over FP32 and cut dynamic energy by 28.5%, according to benchmark results published by Kristian Lindholm and the AI Orchestra (Coldwater Tractors Inc) on November 16, 2026. On a 4608 × 4608 lattice (21.23M sites), the BF16 kernel ran a median 10.204 ms versus 15.291 ms for FP32, while dynamic energy fell from 261.164 mJ per sweep to 186.735 mJ per sweep; on a 9216 × 9216 lattice (84.93M sites) the speedup was 1.477x and the energy saving 27.1%. The authors locked correctness gates before execution and reported zero deviation from an FP64 reference across randomized and uniform initial states at L=384 and L=768.

recommendation

Switch memory-bound kernel state from FP32 to BF16 tiling when a correctness gate against an FP64 reference passes; the source measured 1.499x speedup and 28.5% dynamic energy savings on an A40.

more in playbook: Inference and self-hosting
read3 min views4 publishedNov 17, 2026

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

sm_86): 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) reference across randomized (±1) and uniform ( s<sub>i</sub> = +1) initial states. nvcc -O3 -arch=sm_86 --ptxas-options=-v)

Kernel Variant Registers / Thread Shared Memory Spills Gate
skew_tiled_phase<bf16_t, 38> 15 registers 2,584 bytes 0 bytes PASS
skew_tiled_phase<float, 37> 36 registers 5,032 bytes 0 bytes PASS
skew_tiled_phase<int8_t, 37> 36 registers 1,258 bytes 0 bytes PASS
Grid & Initialization L2(h) /L2(dE) gpu_f32_diff dec_f64_rate |de|/|e| &|dm| Gate
L = 384 (Random ±1) 0.0e+00 /0.0e+00 0 0.00e+00 0.00e+00 /0.00e+00 PASS
L = 384 (Uniform +1) 0.0e+00 /0.0e+00 0 0.00e+00 0.00e+00 /0.00e+00 PASS
L = 768 (Random ±1) 0.0e+00 /0.0e+00 0 0.00e+00 0.00e+00 /0.00e+00 PASS
L = 768 (Uniform +1) 0.0e+00 /0.0e+00 0 0.00e+00 0.00e+00 /0.00e+00 PASS

cudaEvent Throughput (30 Interleaved Trials, 10 Sweeps)

Lattice Size (L × L) bf16_t Median (IQR) fp32 Median (IQR) int8_t Scalar Ref Speedup ( fp32 / bf16 )
4608 × 4608 (21.23M sites) 10.204 ms (0.005 ms) 15.291 ms (0.013 ms) 13.356 ms 1.499x
9216 × 9216 (84.93M sites) 40.679 ms (0.015 ms) 60.064 ms (0.020 ms) 53.072 ms 1.477x

P_idle = 73.02 W)

Lattice Size (L × L) Sweeps / Window E_dyn (fp32 ) E_dyn (bf16 ) Dynamic Energy Saved Total Board Saved
4608 × 4608 4,500 (~4.6–6.9 s) 261.164 mJ/sw 186.735 mJ/sw 28.5% (1.399x) ~29.9%
9216 × 9216 1,200 (~4.9–7.2 s) 1083.034 mJ/sw 788.288 mJ/sw 27.1% (1.371x) ~28.8%

ncu) ERR_NVGPUCTRPERM): bf16_t) vs. 4-byte ( float) representation and corroborated by the measured

[Gates 1-3 (rand_init, 2-sw) @ L=384] L2(h)=0.0e+00 L2(dE)=0.0e+00 gpu_f32_diff=0 dec_f64_rate=0.00e+00 |de|/|e|=0.00e+00 |dm|=0.00e+00 (|m|=0.0043)
[Gates 1-3 (unif_init, 2-sw) @ L=384] L2(h)=0.0e+00 L2(dE)=0.0e+00 gpu_f32_diff=0 dec_f64_rate=0.00e+00 |de|/|e|=0.00e+00 |dm|=0.00e+00 (|m|=0.9400)
[Gates 1-3 (rand_init, 2-sw) @ L=768] L2(h)=0.0e+00 L2(dE)=0.0e+00 gpu_f32_diff=0 dec_f64_rate=0.00e+00 |de|/|e|=0.00e+00 |dm|=0.00e+00 (|m|=0.0064)
[Gates 1-3 (unif_init, 2-sw) @ L=768] L2(h)=0.0e+00 L2(dE)=0.0e+00 gpu_f32_diff=0 dec_f64_rate=0.00e+00 |de|/|e|=0.00e+00 |dm|=0.00e+00 (|m|=0.9418)
L=4608 sweeps=10 | bf16 med=10.204ms (IQR 0.005) | fp32 med=15.291ms (IQR 0.013) | int8 med=13.356ms | speedup(fp32/bf16)=1.499x
L=9216 sweeps=10 | bf16 med=40.679ms (IQR 0.015) | fp32 med=60.064ms (IQR 0.020) | int8 med=53.072ms | speedup(fp32/bf16)=1.477x
GPU: NVIDIA A40
Baseline P_idle = 73.02 W
[NVML L=4608 sweeps=4500] E_dyn_fp32=261.164 mJ/sw | E_dyn_bf16=186.735 mJ/sw | Energy ratio=1.399x (28.5% reduction)
[NVML L=9216 sweeps=1200] E_dyn_fp32=1083.034 mJ/sw | E_dyn_bf16=788.288 mJ/sw | Energy ratio=1.371x (27.1% reduction)

Status: VERIFIED PASS (3/3 large sizes with statistical separation)

Target Architecture: NVIDIA A40 (48GB GDDR6, PCIe)

Performance Gain: ~1.05x at L=2048, 1.119x at L=8192 over optimal control.

For shop inquiries, technical documentation packets, or custom fabrication specs:

── more in #ai-infrastructure 4 stories · sorted by recency
── more on @nvidia a40 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/show-hn-1-499x-speed…] indexed:0 read:3min 2026-11-17 · —