The opportunity
Cerebras Systems builds the world's largest AI chip, 56 times larger than GPUs. This architecture allows Cerebras to deliver industry-leading training and inference speeds; over 10 times faster than GPU-based hyperscale cloud inference services.
What you'll do
Own and evolve performance models and modeling methodologies for: next-generation accelerator and system architectures.
Build and extend analytical, simulation-based or trace-driven models across: workloads, architectural features and product generations.
Analyze important AI workloads, from individual kernels through end-to-end: inference and training execution, to determine where time, bandwidth, compute and capacity are spent.
Identify hardware and software bottlenecks and quantify opportunities to: improve latency, throughput, utilization and energy efficiency.
Evaluate proposed architectural features and determine their expected: performance return across representative workloads.
Study how models and kernels map onto the underlying compute, memory and communication architecture.
What they're looking for
- + years of experience in performance analysis, performance modeling or: architecture exploration for CPUs, GPUs, AI accelerators or other high-performance computing systems.
- Strong understanding of hardware architecture developed through hardware,: compiler, kernel, runtime or system-performance work.
- Experience developing analytical, simulation-based or trace-driven: performance models using Python, C++ or similar environments.
- Solid understanding of processor architecture, memory systems, interconnects,: parallel execution and hardware resource constraints.
- Ability to move between kernel-level behavior and end-to-end application or system performance.
- Experience profiling workloads, forming performance hypotheses and validating them with quantitative evidence.
- Understanding of how software mapping and programmability affect realized hardware performance.
- Ability to communicate modeling assumptions, uncertainty, bottlenecks and recommendations clearly.
