Member of Technical Staff, Performance & Capacity

Measure fleet utilization, goodput, and training efficiency on real workloads Identify and close sources of wasted capacity Own the capacity model, including node counts, assumptions, and error bars Defend capacity decisions when real money is committed Own workload mix, preemptible fraction, and checkpointing economics over time Optimize parallel file systems, data locality, and interconnect behavior for multi-node GPU training Locate actual performance bottlenecks beyond profiler summaries Explain performance results to researchers in usable terms Requirements Five or more years with GPU and large-scale compute workloads Real multi-node experience with distributed training performance, interconnects, and collective communication System-level knowledge of GPU performance characteristics, including H100 and B200 workloads Deep understanding of AI training and inference performance Understanding of step time, MFU, communication/compute overlap, decode bandwidth, batching, and latency floors Experience owning a capacity decision with financial consequences Fluent knowledge of parallel file systems and their impact on training throughput Fluent knowledge of InfiniBand or RDMA-class networking behavior Ability to explain performance results clearly to researchers Work authorization/sponsorship may be relevant for the United States Nice to have: data center floor experience, thermal and power envelopes, failure domains, energy- or preemption-aware scheduling, cloud and specialist GPU provider sourcing, and scientific computing or HPC background Core Competencies Demonstrates expertise in optimizing GPU performance for large-scale compute workloads, with a strong focus on capacity modeling, parallel file systems, and interconnect behavior. Capable of clearly communicating complex performance results to researchers and making informed capacity decisions with financial implications. Highest-signal resume keywords GPU Performance Optimization Distributed Training Performance Capacity Decision Ownership Parallel File Systems Knowledge InfiniBand Networking Expertise Hard Skills GPU Workloads Large-Scale Compute Performance Bottleneck Analysis AI Training and Inference Communication/Compute Overlap Batching Latency Floors Error Bar Analysis Checkpointing Economics Data Locality Optimization Soft Skills Clear Communication Problem-Solving Industry Keywords Capacity Modeling Multi-Node GPU Training Training Throughput HPC Background Data Center Experience Tools & Technologies InfiniBand RDMA Networking Parallel File Systems #J-18808-Ljbffr Jobtailor