8 hours ago
Remote, United States +2 moreStaff+
Base Salary
$180k - $270k/yr
Responsibilities
- Bring up and validate new hardware architectures across the Modular software stack, including kernels, compiler infrastructure, runtime, graph execution, and model serving.
- Write, port, and optimize Mojo kernels for novel accelerator architectures while establishing correctness and improving performance.
- Investigate correctness and performance issues involving kernel execution, memory movement, compiler lowering, graph decisions, synchronization, and vendor runtime behavior.
- Develop expertise in accelerator ISAs, execution models, memory hierarchies, synchronization mechanisms, compiler constraints, and vendor toolchains.
- Map AI operators and workloads such as matrix multiplication, convolution, and reductions onto target hardware.
- Build portability infrastructure, compiler integrations, tooling, testing, debugging workflows, and hardware-vendor integration tests.
- Benchmark and profile workloads, identify bottlenecks, and help mature platforms from initial bring-up to production-quality performance.
- Document emerging architectures through technical documentation, demos, design discussions, and engineering write-ups.
- Collaborate with internal compiler, runtime, kernel, and hardware teams and with external hardware vendors.
Requirements
- 5+ years of experience in high-performance computing, compiler engineering, accelerator software, systems engineering, or a closely related domain in industry or research.
- Strong C++ skills and experience contributing to complex, multi-component software systems.
- Hands-on experience with a heterogeneous programming model such as CUDA, SYCL, OpenCL, or a comparable accelerator programming environment.
- Experience writing or modifying GPU kernels, custom operators, accelerator code, or working with PyTorch at the C++/systems layer.
- Working knowledge of memory hierarchy, parallel execution, synchronization, data movement, and software performance relationships to hardware architecture.
- Experience debugging across software abstraction boundaries and reasoning systematically about correctness and performance.
- Bachelor’s degree and 4+ years, master’s degree and 3+ years, or PhD and 2+ years of relevant software engineering or related work experience.
- 2+ years of work experience with a programming language such as Python, Mojo, Rust, or C++.
- Strong communication and collaboration skills and the ability to learn unfamiliar hardware platforms, architecture manuals, ISA documentation, and vendor materials.
- Helpful qualifications include experience with DSPs, NPUs, AI ASICs, MLIR, LLVM, Triton, CUTLASS, CuTe, new accelerator bring-up, hardware vendors, hardware-specific performance tools, graph compilers, model serving, inference optimization, runtime concepts, and multi-architecture portability layers.
Benefits
- Comprehensive healthcare coverage, retirement and savings programs, employee stock purchase opportunities, paid time off, wellbeing resources, family support programs, and learning and development opportunities, varying by location.
- Competitive compensation may include RSU grants, annual target bonus, equity, and benefits.
- Remote work from home or work from offices in Los Altos, California, or Edinburgh, United Kingdom.
- New-hire onboarding is conducted in person at the appropriate office.
- Regular team onsites and local meetups, with expected travel two to four times per year.
- Candidates in the United States, Canada, and United Kingdom are welcome to apply.
About Modular
Modular builds an AI developer platform for training and especially inference/serving, centered on the MAX runtime and the Mojo programming language. Its tools accelerate and deploy models from frameworks like PyTorch and TensorFlow on CPUs and GPUs, for teams running on cloud or on‑prem infrastructure. Founded in 2022, the company operates remote‑first with an office in Los Altos, CA, and sells a commercial platform and enterprise support to organizations productionizing generative and classical ML.