Nvidia · posted · checked
Senior System GPU Performance Engineer
C++PythonSQLOn-site · ChinaSenior
Who can apply
On-site · China
Location: China, Shanghai
NVIDIA has transformed computer graphics, PC gaming, and accelerated computing for more than 25 years. Today, our GPUs power advances in AI, Datacenter, Gaming, Robotics, Automotive, and scientific discovery. NVIDIA's Silicon Co-Design Group (SCG) takes GPU, SoC, and CPU programs from first power-on to high-volume production. We sit at the crossroads of architecture, design, marketing, operations, and productization across Datacenter, Gaming, Robotics, Automotive, and Embedded markets.
We are hiring a Senior System GPU Performance Engineer to maximize the performance and power efficiency of production GPU systems. You will connect workload behavior, silicon capability, software policy, and platform constraints to identify bottlenecks and productize improvements. This is not a benchmark-execution or validation-only role—you will own analysis from hypothesis through root-cause closure, plan-of-record integration, and confirmed product impact. Great work turns complex system data into faster, more efficient, and more predictable products.
What you'll be doing:
Own system-level GPU performance and power characterization from first silicon through production across representative applications, benchmarks, and product configurations.
Drive performance and power feature productization, translating measured behavior into firmware, driver, BIOS, platform, and silicon recommendations that meet product targets and speed-of-light schedules.
Design experiments, execute test plans, and build models that isolate bottlenecks across GPU compute, memory, interconnect, CPU interaction, power delivery, and thermal limits.
Analyze production-silicon data across process, voltage, temperature, workloads, and bins to quantify performance-per-watt trade-offs and identify causal optimization opportunities.
Lead multi-functional root-cause closure across architecture, design, validation, software/firmware, power and thermal, reliability, ATE, product management, manufacturing, and operations; own fixes through confirmation.
Establish reusable automation, visualization, and closed-loop methodologies that improve experiment coverage, analysis accuracy, debug velocity, and learning across future GPU programs.
Translate complex system signals into decision-ready options for executive leadership on feature readiness, product configuration, targets, and program risks.