AMD GPU Performance Engineer

MindSource; San Francisco (CA); On-site; USD 140,000 - 190,000; Full time

Job summary

MindSource is seeking an experienced GPU Performance Engineer to optimize vLLM on AMD ROCm-based GPUs. You will focus on GPU kernel development, memory layout, and low-level performance tuning to push inference throughput.

You will collaborate with open-source AI infrastructure and contribute to performance benchmarks, profiling, and kernel fusion techniques, helping scale vLLM across AMD GPUs in production environments.

Qualifications

• Experience optimizing GPU kernels for AI/ML workloads.
• Strong knowledge of LLM inference techniques and GPU execution.
• Ability to profile, benchmark, and tune performance across AMD ROCm.
• Familiarity with attention, GEMM, KV cache, or fused kernels.

Responsibilities

• Develop and optimize GPU kernels for vLLM on AMD ROCm.
• Profile, benchmark, and tune inference performance at scale.
• Collaborate with open-source AI infra to improve kernel implementations.

Skills

GPU kernel dev

LLM inference optimization

Profiling & benchmarking

GPU architecture & memory

Attention / GEMM / KV cache

Tools

ROCm

Triton

PyTorch

Job description

We're looking for an experienced GPU Performance Engineer to help make vLLM a high-performance inference engine across AMD GPUs.

You'll work on cutting-edge AI infrastructure, optimizing GPU kernels and inference performance using AMD's ROCm ecosystem.

Key Skills:

✅ GPU kernel development and performance tuning

✅ LLM inference: vLLM, SGLang, or TensorRT-LLM

✅ Profiling, benchmarking, and performance optimization

✅ GPU architecture, memory optimization, and kernel execution

✅ Experience with attention, GEMM, KV cache, or fused kernels

Open-source contributions to vLLM, ROCm, Triton, or PyTorch are a plus.

This is an opportunity to solve challenging GPU performance problems and contribute to open-source AI infrastructure used across the ecosystem.