GPU/Kernel Engineer

Acceler8 Talent; Palo Alto (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Acceler8 Talent is seeking a GPU Kernel Performance Engineer in Palo Alto to optimize kernels, analyze memory hierarchies, and drive system-level efficiency for AI workloads. You will own kernel development, profile bottlenecks, and work with ML infra teams to translate kernel gains into broader performance improvements.

The role is highly hands-on, with opportunities to shape GPU performance across architectures and accelerator stacks in a fast-moving startup environment.

Qualifications

• Hands-on experience developing, modifying, or optimizing GPU kernels using CUDA or comparable GPU/HPC programming technologies.
• Strong understanding of GPU architecture, including memory hierarchies, parallel execution, synchronization, and hardware utilization.
• Demonstrated ability to profile GPU workloads and identify underlying performance bottlenecks.
• Experience implementing performance improvements and measuring their impact using benchmarks or profiling tools.
• Ability to explain the relationship between GPU hardware constraints, software implementation choices, and measured performance.
• Strong programming fundamentals and experience debugging low-level performance issues.
• Willingness to work across GPU kernels, compiler/runtime behavior, and broader software systems.
• Strong technical problem-solving skills and the ability to work effectively in a fast-moving engineering environment.

Responsibilities

• Profile GPU workloads to identify performance bottlenecks, hardware limitations, and opportunities for improved efficiency.
• Optimize existing GPU kernels and develop new kernels where necessary to improve workload performance.
• Analyze GPU architectural constraints, including memory hierarchy, parallel execution, synchronization, and compute utilization.
• Investigate and optimize memory access patterns, kernel fusion, occupancy, and data movement.
• Develop reproducible performance benchmarks and measure improvements across representative AI workloads.
• Validate numerical correctness and performance consistency following kernel optimizations.
• Collaborate with ML infrastructure, serving, and applied machine learning teams to translate kernel-level improvements into system-wide performance gains.
• Investigate performance interactions across GPU kernels, compilers, runtimes, and supporting software infrastructure.
• Evaluate communication and memory bandwidth trade-offs across GPU and multi-GPU systems.

Skills

GPU kernel optimization

CUDA

GPU architecture

Profiling workloads

Performance tuning

Low-level debugging

System-level optimization

Programming fundamentals

Tools

CUDA toolkit

NVIDIA Nsight

Triton

CUTLASS

cuDNN

NVCC

PTX / SASS

Job description

An innovative AI infrastructure startup is seeking a GPU / Kernel Performance Engineer to optimize the performance, power efficiency, and cost of large-scale AI training and inference workloads.

The company is developing intelligent software that analyzes AI workloads and improves how they execute on modern accelerator hardware.

This is a highly hands-on engineering role offering direct ownership of GPU performance optimization, kernel development, and system-level efficiency improvements. You'll work closely with a small, highly technical team spanning GPU computing, ML infrastructure, and performance engineering.

The team is open to both dedicated GPU kernel specialists and broader performance engineers who have experience modifying and optimizing low-level GPU code.

Key Responsibilities

• Profile GPU workloads to identify performance bottlenecks, hardware limitations, and opportunities for improved efficiency.
• Optimize existing GPU kernels and develop new kernels where necessary to improve workload performance.
• Analyze GPU architectural constraints, including memory hierarchy, parallel execution, synchronization, and compute utilization.
• Investigate and optimize memory access patterns, kernel fusion, occupancy, and data movement.
• Develop reproducible performance benchmarks and measure improvements across representative AI workloads.
• Validate numerical correctness and performance consistency following kernel optimizations.
• Collaborate with ML infrastructure, serving, and applied machine learning teams to translate kernel-level improvements into system-wide performance gains.
• Investigate performance interactions across GPU kernels, compilers, runtimes, and supporting software infrastructure.
• Evaluate communication and memory bandwidth trade-offs across GPU and multi-GPU systems.

Required Qualifications

• Hands-on experience developing, modifying, or optimizing GPU kernels using CUDA or comparable GPU/HPC programming technologies.
• Strong understanding of GPU architecture, including memory hierarchies, parallel execution, synchronization, and hardware utilization.
• Demonstrated ability to profile GPU workloads and identify underlying performance bottlenecks.
• Experience implementing performance improvements and measuring their impact using benchmarks or profiling tools.
• Ability to explain the relationship between GPU hardware constraints, software implementation choices, and measured performance.
• Strong programming fundamentals and experience debugging low-level performance issues.
• Willingness to work across GPU kernels, compiler/runtime behavior, and broader software systems.
• Strong technical problem-solving skills and the ability to work effectively in a fast-moving engineering environment.

Preferred Qualifications

• Experience with GPU programming frameworks and libraries such as CUDA, Triton, CUTLASS, or cuDNN.
• Familiarity with low-level GPU instructions and optimization using PTX or SASS.
• Experience with GPU profiling tools such as NVIDIA Nsight Systems or Nsight Compute.
• Knowledge of roofline analysis, Tensor Cores, warp specialization, memory coalescing, and occupancy optimization.
• Understanding of GPU memory systems, including HBM, shared memory, caches, and memory bandwidth limitations.
• Experience with multi-GPU communication technologies such as NCCL, NVLink, or NVSwitch.
• Familiarity with AMD GPU programming using ROCm/HIP.
• Experience with TPU/XLA or performance optimization across multiple accelerator architectures.
• Background optimizing AI training, inference, or high-performance computing workloads.

Why Join?

• Work on core performance optimization technology for next-generation AI infrastructure.
• Take direct ownership of GPU kernel and system-level performance improvements.
• Collaborate with experienced engineers from leading AI hardware and accelerator development teams.
• Gain exposure to GPU architecture, compiler/runtime behavior, and large-scale AI infrastructure.
• Work within a small engineering team where individual contributions have a direct impact on product performance.
• Opportunity for exceptional graduates and early-career engineers to take on challenging, high-impact technical work.