ABOUT
Eliminating Floating-Point Error
VLA achieves exact GPU arithmetic 13,848x faster than CPU methods. Cross-GPU reproducible. Patent pending.
The Problem
For 40 years, scientific computing has faced an impossible tradeoff: fast (GPU, wrong) or correct (CPU, slow).
Floating-point arithmetic on GPUs produces different results on different hardware. The same code, same data, different answers. Papers can't be replicated. Audits fail. The Patriot missile bug killed 28 soldiers.
Until now, the only solution was to use slow CPU exact methods (Python Decimal, mpmath) that take days instead of minutes.
Our Solution: VLA
VLA (Verified Lossless Arithmetic) is a drop-in replacement for PyTorch operations that guarantees exact results with zero floating-point error.
13,848x
faster than CPU exact methods
100%
cross-GPU reproducible
Same checksum on RTX 4060, Tesla T4, or H100. Bit-identical. Guaranteed.
The Team
Kyle Clouthier
Founder
Kyle founded Clouthier Simulation Labs to make exact GPU arithmetic accessible to every engineer, researcher, and student — not just those with access to expensive HPC clusters.
The Company
Clouthier Simulation Labs
A Canadian technology company building GPU-accelerated precision arithmetic tools with patent-pending technology.
Location
Petawawa, Ontario, Canada
Status
Y Combinator S2026 Applicant
IP Protection
Patent pending