GPU AI Performance Engineer

We are looking for a GPU AI Performance Engineer to help design, optimize, and accelerate AI workloads on GPU-based platforms. This role focuses on improving the performance, memory efficiency, and scalability of modern AI models — especially LLMs, transformer workloads, MoE models, and computer vision pipelines — by working across the stack from model architecture to compiler/runtime integration and low-level kernel optimization.

Responsibilities

- Analyze and evaluate GPU architecture/microarchitecture for performance optimizations of AI workloads

- Work across the AI stack, from model graphs and inference runtimes down to GPU kernels and compiler IR

- Collaborate with hardware, software, and ML teams to identify performance bottlenecks and propose architectural or algorithmic optimizations

- Analyze AI workload characteristics and correlate their behavior across different GPU generations

- Contribute to architectural trade-off studies and influence GPU roadmap decisions with data-driven insights

Preferred Skills:

- Strong understanding of CPU/GPU architecture

- Experience in Python, C++, and ML frameworks (e.g., TensorFlow, PyTorch)

- Skills: C/C++ Programming Language, Scripting (Python/Perl), Assembly, Verilog/SystemVerilog

- Familiarity with AI inference runtimes and deployment stacks for model compilation, optimization, and execution on GPU/accelerator platforms

- Good understanding of common neural network layers and operations, including what they do and how they affect model behavior and performance

Nice to have:

- Experience with LLM inference engines such as llama.cpp

- Exposure to Triton, TTIR, TTGIR, or GPU compiler pipelines

- Knowledge of quantization formats and tradeoffs

- Understanding of MoE architectures

- Experience with GPU driver and compiler development

- Experience with OpenCL or Cuda development