GPU acceleration engineer

GPU Acceleration Engineer - Calculation Engine

Main Mission

Massively accelerate the sparse calculation engine of a UK SaaS B2B - Enterprise Planning & Analytics company by porting critical algorithms from Rust/C++ to GPU (CUDA). Transform currently impossible calculations (requiring thousands of years of CPU time) into operations achievable in minutes.

Context

UK SaaS B2B - Enterprise Planning & Analytics company manages planning models reaching 64 quadrillion cells with billions of time periods. Our Hyperblock/Polaris engine is currently limited by:

• Legacy CPU architecture (Java/Rust/C++)

• Memory constraints on massive sparse structures

• Prohibitive calculation times on complex scenarios

Objective: Achieve performance gains of 100x to 1000x via GPU offloading.

Main Responsibilities

GPU Offloading

• Port existing Rust/C++ algorithms to CUDA/GPU

• Identify and extract critical calculation paths to accelerate

• Optimize sparse matrix operations for GPU architecture

• Develop performant Rust ↔ CUDA wrappers

• Benchmark and validate performance gains

Memory Optimization

• Design GPU memory management strategies for massive datasets

• Implement efficient patterns for sparse structures

• Optimize CPU ↔ GPU memory transfers

• Manage GPU memory limitations on large-scale calculations

Technical Collaboration

• Work with engineering team on integration

• Document GPU porting patterns

• Participate in code reviews and design reviews

• Train the team on GPU best practices

Technical Stack

Languages (in order of importance)

• CUDA - Primary GPU development

• Rust - Source language for algorithms to port

• C++ - Legacy components and CUDA interoperability

• (Java - platform context, no dev required)

Key Technologies

• NVIDIA CUDA (toolkit, libraries: cuBLAS, cuSPARSE)

• Rust (ownership model, unsafe blocks, FFI)

• GPU Programming (kernels, memory hierarchy, optimization)

• Sparse Matrix Operations (compression, storage formats)

• Profiling Tools (nvprof, Nsight, perf)

Required Profile

Essential Skills

GPU & CUDA (Essential)

• Significant CUDA programming experience (3+ years)

• Mastery of GPU kernel optimization

• Deep knowledge of NVIDIA GPU architecture (memory hierarchy, warps, occupancy)

• Experience with sparse calculations on GPU (cuSPARSE or equivalent)

Rust (Essential)

• Production Rust development

• Mastery of ownership and borrowing system

• Experience with unsafe Rust and FFI (Foreign Function Interface)

• Ability to analyze and refactor existing Rust code

C++ (Required)

• Modern C++ (C++11/14/17)

• C++ ↔ CUDA integration

• Templates and metaprogramming (asset)

Algorithms (Required)

• Data structures for scientific computing

• Sparse matrix algorithms (CSR, COO, etc.)

• Performance optimization and profiling

• Parallelization and concurrency concepts

Highly Valued Experience

• Documented CPU → GPU porting projects

• HPC experience (supercomputers, GPU clusters)

• Memory optimization for large-scale datasets

• Scientific computing or numerical simulation

• Rust interop with other languages (C/C++/Python)

Working Arrangements

Location & Travel

• 100% remote (France/Europe base preferred)

• Occasional travel to London

  • Frequency: ~1 week/month for team sprints

  • Project kickoff + key reviews

  • Intensive collaboration sessions

Start date: As soon as possible