CUDA Engineer: High-Perf GPU Kernels for Inference

Fuse Energy; United States; On-site; USD 140,000 - 210,000; Full time

Benefits offered by this job

Biannual bonus

Equity eligibility

Fully expensed tech

Private health insurance

Meals allowance

Job summary

Fuse Energy seeks a CUDA Engineer to write and optimise low-level GPU code powering our transformer inference workloads. You will design custom CUDA kernels, tune memory patterns, and push maximum throughput across our GPU fleet.

Responsibilities include kernel fusion, mixed-precision arithmetic, and efficient autoregressive decoding. You will profile, test, and maintain CUDA libraries while collaborating with a fast-moving AI-enabled energy team.

Qualifications

• 4+ years writing production CUDA code.
• Deep understanding of GPU microarchitecture: warps, occupancy, registers and memory hierarchy.
• Strong CUDA C++ skills, including streams and asynchronous execution.
• Experience with kernel fusion, memory coalescing and avoiding warp divergence.
• Experience writing quantised and mixed-precision kernels.

Responsibilities

• Write and optimise custom CUDA kernels for core transformer inference operations.
• Profile kernels to identify bottlenecks in occupancy, memory throughput and warp divergence.
• Apply kernel fusion to reduce memory round-trips and launch overhead.
• Optimise memory access patterns and manage memory hierarchy for max bandwidth.
• Implement quantisation-aware kernels and mixed-precision arithmetic to reduce latency.
• Build and tune caching mechanisms for autoregressive decoding.
• Tune kernel launch configurations for target GPU architectures.
• Benchmark kernels against baselines and drive throughput improvements.
• Write tests for CUDA code to catch performance and correctness regressions.

Skills

CUDA C++

GPU profiling

Kernel optimization

Quantisation & mixed precision

Parallel algorithm design

HPC latency engineering

Multi-GPU / multi-node

Job description

Fuse Energy seeks a CUDA Engineer to write and optimise low-level GPU code powering our transformer inference workloads. You will design custom CUDA kernels, tune memory patterns, and push maximum throughput across our GPU fleet.

Responsibilities include kernel fusion, mixed-precision arithmetic, and efficient autoregressive decoding. You will profile, test, and maintain CUDA libraries while collaborating with a fast-moving AI-enabled energy team.