CUDA Kernel Engineer - High-Performance Transformer Inference

Fuse Energy Supply; Greater London; On-site; GBP 63,000 - 103,000; Full time

Benefits offered by this job

Equity sign-on bonus

Biannual bonus

Tech stipend

Breakfast and dinner allowance

Job summary

Fuse Energy Supply in the United Kingdom is seeking an experienced CUDA specialist to write and optimize production kernels for transformer inference workloads, with a focus on latency and throughput.

You will profile and tune kernels, apply fusion, manage memory hierarchies, and implement mixed-precision and quantisation-aware code. Experience with multi-GPU, high-performance libraries, and reading PTX/SASS is a plus.

Qualifications

• 4+ years writing production CUDA code with shipping performance-critical kernels.
• Deep understanding of GPU microarchitecture, warps, occupancy, and memory hierarchy.
• Strong CUDA C++ skills, including streams and asynchronous execution.
• Hands-on profiling to diagnose compute-bound vs. memory-bound bottlenecks.
• Experience with kernel fusion, memory coalescing, and avoiding warp divergence.
• Experience writing quantised and mixed-precision kernels.
• Solid grasp of parallel algorithm design and numerical precision tradeoffs.
• Nice to have: transformer/attention-style kernels or autoregressive decoding.
• Nice to have: building high-performance GPU libraries from scratch.
• Nice to have: background in HPC or latency-critical perf engineering.
• Nice to have: exposure to multi-GPU or multi-node kernel-level optimisation.
• Nice to have: comfortable reading PTX/SASS to validate kernel efficiency.

Responsibilities

• Write and optimise custom CUDA kernels for core transformer inference operations.
• Profile kernels to identify and eliminate bottlenecks in occupancy, memory throughput, and warp divergence.
• Apply kernel fusion to reduce memory round-trips and launch overhead across inference pipelines.
• Optimise memory access patterns and manage the memory hierarchy for maximum bandwidth utilisation.
• Implement quantisation-aware kernels and mixed-precision arithmetic to reduce latency and memory footprint.
• Build and tune caching mechanisms for efficient autoregressive decoding.
• Tune kernel launch configurations for target GPU architectures.
• Benchmark kernels against existing baselines and drive measurable throughput and latency improvements.
• Write tests for CUDA code to catch performance and correctness regressions.
• Maintain internal CUDA libraries and contribute to our team coding standards and documentation.

Skills

CUDA

CUDA C++

GPU microarchitecture

Profiling

Kernel fusion

Memory coalescing

Warp divergence

Quantised kernels

Mixed-precision

Parallel algorithms

Transformer kernels

High-performance libraries

HPC experience

Multi-GPU knowledge

PTX/SASS reading

Tools

SASS

NodeJS

CUDA

Job description

Fuse Energy Supply in the United Kingdom is seeking an experienced CUDA specialist to write and optimize production kernels for transformer inference workloads, with a focus on latency and throughput.

You will profile and tune kernels, apply fusion, manage memory hierarchies, and implement mixed-precision and quantisation-aware code. Experience with multi-GPU, high-performance libraries, and reading PTX/SASS is a plus.