Skip to content

Tensor Virus

Overview

The tensor_virus uses Half-Precision (FP16) mathematics to deeply saturate the Tensor and Matrix cores.

Execution Mechanics

  • It executes an FP16 hammer using a universal Fused Multiply-Add (FMA) loop.
  • To ensure continuous saturation across both AMD and NVIDIA architectures, it relies on universal hardware intrinsics.
  • The instructions continuously flip polarity to prevent values from converging, ensuring the datapaths cannot aggressively power-gate.

Target Subsystems

  • Primary Target: Tensor Cores and FP16 compute pipelines.