Senior System Software Engineer – Data Center GPU Compute Diagnostics

We are seeking a senior system software engineer to work on next-generation Data Center GPU diagnostics for rack-scale AI supercomputer systems. Our charter is to build applications and compute workloads that test and heavily stress GPU compute engines, HBM memory, cache hierarchy, PCIe/NVLink interfaces, power delivery, and thermal behavior, and to use those applications in silicon/system bring-up along with packaging such tools for manufacturing and customer use. The best candidates will have strong experience writing low-level diagnostic, performance, or stress software for complex hardware systems, ideally including experience with GPUs, CUDA kernels, GEMM-style workloads, NCCL communication patterns, CPUs, NICs or high-speed interconnects such as PCIe.  

Excellent interpersonal skills are required as this role will involve mentoring other engineers and collaborating with hardware architecture, silicon validation, manufacturing and field teams. In addition, the engineer will extensively use their knowledge of operating systems, computer architecture, GPU memory, voltage/frequency behavior, thermal limits, high-speed buses, and modern AI development and analysis tools to efficiently validate and test next-generation  processors and systems. Join an exciting, rewarding and fast paced environment! 

What you'll be doing: 

• Working closely with hardware architecture, driver, manufacturing and field teams through product development lifecycle of rack-scale AI systems. 

• Responsible for crafting CUDA/C++ diagnostic workloads and software infrastructure required for new chip development, validation, productization, and field triage. 

• Designing and implementing GPU compute tests that stress Tensor Cores, SMs, L2/cache hierarchy, HBM memory, and related power/thermal operating points. 

• Developing and tuning GEMM-style diagnostic workloads, including tests combined with additional load in NVLink, PCIe or CPU subsystems. 

• Developing and integrating higher-level AI workload tests, including PyTorch-based large model workloads to stress GPUs, memory, interconnects, thermals, and system software under realistic rack-scale AI use cases. 

• Assessing new hardware features and architecting manufacturing and field diagnostic tests using pre-beta GPU drivers, low-level diagnostic software, and system telemetry. 

• Debugging failures involving ECC, HBM behavior, thermal limits, voltage/frequency margining and PCIe/NVLink errors. 

What we need to see: