II.
Workflow overview
Reference · liveworkflow:gpu-kernel-benchmarking
GPU Kernel Benchmarking overview
Benchmarks GPU compute kernels across hardware targets — measuring throughput, latency, memory bandwidth, and occupancy; comparing against baseline; profiling hotspots; and publishing results to a performance dashboard. Excludes kernel development.
Attributes
displayName
GPU Kernel Benchmarking
workflowKind
development
triggerType
event-driven
typicalCadence
per-optimization
complexity
single-team
description
Benchmarks GPU compute kernels across hardware targets — measuring
throughput, latency, memory bandwidth, and occupancy; comparing
against baseline; profiling hotspots; and publishing results to
a performance dashboard. Excludes kernel development.
Outgoing edges
applies_to_domain2
- domain:scientific-computing·DomainScientific Computing
- domain:ml-ops·DomainMLOps
involves_role3
- role:performance-profiler·RolePerformance Profiler
- role:implementer·RoleImplementer
- role:data-scientist·RoleData Scientist
performed_by_org_unit2
- org-unit:research-engineering·OrgUnitResearch Engineering
- org-unit:ml-team·OrgUnitML Team
requires_skill_area1
- skill-area:observability-pipeline·SkillAreaObservability Pipeline
triggers_responsibility1
- responsibility:performance-budget-tracking·ResponsibilityPerformance budget tracking
Incoming edges
None.