AI Accelerator, Sr. Principal Software Engineer - Kernel Library
Ampere Computing
Job details
Pay
$237,000 – $338,500 / yearly
Job type
Full-time
Work setting
Remote / Hybrid
Location
Portland, OR
Posted
Posted yesterday
Full job description
About the role
As a Software Engineer on Ampere's AI Accelerator team, you will develop high-performance kernels to optimize throughput, latency, and hardware resource utilization for state-of-the-art deep learning models used in real-world AI applications, with potential implications across industries such as robotics and automotive. You will identify kernel-development opportunities across a broad range of models and optimize memory bandwidth utilization, data movement, and parallel execution to deliver efficient workloads on accelerator hardware.
What You'll Achieve:
- Deliver high-performance kernels that improve the throughput and latency of production deep learning workloads.
- Turn model profiling results into targeted kernel and data-movement optimizations for accelerator hardware.
- Improve utilization of compute resources, memory bandwidth, and on-chip resources across different deep learning models.
- Help shape the kernel-development workflow, performance methodology, and software stack for the accelerator platform.
- Identify and prioritize the operations that dominate end-to-end model latency and throughput.
- Design specialized kernels for compute-intensive and memory-bandwidth-intensive operations.
- Develop reusable kernel components and optimization techniques for multiple models.
- Enable efficient execution of state-of-the-art models on the accelerator platform.
- Collaborate with compiler, runtime, kernel, architecture, hardware, and application teams...
Qualifications
- Experience with accelerator programming platforms or frameworks such as CUDA, ROCm, OpenCL, SYCL, or Triton, and familiarity with GPU, NPU, or other accelerator architectures.
- Familiarity with compiler technologies, graph compilers, code generation, or runtime systems.
- Experience developing complex, multi-stage kernels, such as FlashAttention, paged attention, block-sparse attention, MoE routing and dispatch, Mamba selective scan kernels, and fused Transformer decoding kernels.
- Experience developing or optimizing DSP workloads, such as audio and speech processing, image and video processing, wireless communications, radar, or sensor fusion.
Requirements
- Bachelor's degree in Computer Science, Computer Engineering, or a related technical field and 12 years of relevant experience; or a Master's degree and 8 years of relevant experience; or a PhD and 5 years of relevant experience.
- Demonstrated proficiency in Python and C/C++ for performance-sensitive software development.
- Experience developing or optimizing kernels for GPUs, AI accelerators, DSPs, or other parallel computing hardware.
- Ability to use profiling and benchmarking tools to identify performance bottlenecks and measure optimization impact.
- Understanding of the performance characteristics of deep learning operations, including matrix multiplication, convolution, attention, normalization, reduction, and data movement.
- Experience validating kernel correctness and numerical accuracy across different data types and problem sizes.
- Strong analytical and debugging skills, with the ability to investigate ambiguous technical problems and deliver robust solutions.
- Ability to quickly learn new architectures, frameworks, compilers, and workloads.
Ready to apply?
Apply on the company siteYou’ll be taken to an external site to complete your application.