GPU Performance Engineer — Open-Weight AI, Hybrid Dublin

HireHive; Dublin; Hybrid; EUR 120,000 - 180,000; Full time

Benefits offered by this job

Hybrid working

25 days paid annual leave

Free inference tokens

Job summary

TensorX in Dublin is hiring multiple GPU Performance Engineers to optimize inference engines for latency targets. You will patch open-source engines, tune kernels, and bring up new models on day zero within a high-concurrency production environment.

You will work with the Inference Team on CUDA kernels, memory management and KV cache behavior, striving to maximize per-GPU throughput while maintaining correctness and observability.

Qualifications

• Ability to read inference engine source to identify root causes.
• Strong GPU architecture understanding and CUDA fundamentals.
• Working knowledge of transformer inference and KV cache behavior.
• Proficient in Python; comfortable reading C++ and CUDA.
• Measure-first mindset with controlled experiments and timing checks.
• Honest about results, including failed experiments.
• Eager to learn; stack evolves weekly with new tech.
• Familiar with AI-assisted development tools such as Claude Code/Codex.
• Clear communicator able to explain decisions to technical and non-technical audiences.

Responsibilities

• Engine tickets end to end: reproduce, patch, test and roll out fixes.
• Diagnosis across GPU memory, scheduler, container, router and traffic shape.
• Improve per-GPU goodput by locating bottlenecks and validating gains on production traffic.
• Patch SGLang and vLLM; upstream fixes when appropriate.
• Modify and optimize GPU kernels (attention/indexer, FP8/FP4 paths).
• Model bring‑up on day zero with parallelism layouts and cache config testing.
• Tune cache behavior and router coordination; document changes.
• Maintain pre-production gate; provide benchmarks for repeatability.
• Communicate findings clearly for both engineers and customers.

Skills

GPU architecture

CUDA fundamentals

Transformer inference

Python

C++

Kernel development

KV cache

Benchmarking

AI tooling

Education

BSc/MSc/PhD in CS/Engineering or ML

Tools

SGLang

vLLM

TensorRT-LLM

CUDA

Job description

TensorX in Dublin is hiring multiple GPU Performance Engineers to optimize inference engines for latency targets. You will patch open-source engines, tune kernels, and bring up new models on day zero within a high-concurrency production environment.

You will work with the Inference Team on CUDA kernels, memory management and KV cache behavior, striving to maximize per-GPU throughput while maintaining correctness and observability.