FHElium

Appearance

Benchmarks

N = 65,536 · Scale 250

ScalingThroughput
Log Y

PT × CT matrix multiplication

64 × 64 · Entry depth 0
Not measured

CT × CT matrix multiplication

64 × 64 · Entry depth 0
Not measured

Nonlinear polynomial evaluation

Entry depth 0
Not measured

NTT

Entry depth 0
Not measured
View

Key switching

Entry depth 0
Not measured
View

RNS arithmetic

Entry depth 0
Not measured
View

Benchmark runs

5 runs
2026-09-07

NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition

Linux · PyTorch 2.13.0+cu130

3740 cases in original run
2026-09-08

NVIDIA GeForce RTX 4060 Ti

Linux · PyTorch 2.13.0+cu130

920 cases in original run
2026-09-08

NVIDIA RTX A6000 · 300 W

Linux · PyTorch 2.13.0+cu130

920 cases in original run
2026-09-09

NVIDIA H100 NVL · 400 W

Linux · PyTorch 2.13.0+cu130

920 cases in original run
2026-09-08

AMD Ryzen Threadripper PRO 7975WX 32-Cores

Linux · PyTorch 2.13.0+cu130

920 cases in original run