Skip to main content
nsight-profiler Expert skill for NVIDIA Nsight Systems and Nsight Compute profiling tools. Configure profiling sessions, analyze kernel reports, interpret occupancy metrics, roofline model data, memory bandwidth bottlenecks, and warp execution efficiency.
الانتقال إلى التثبيت سوق المهارات اكتشف واستكشف مهارات الذكاء الاصطناعي التي بناها المجتمع.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
عرض تفاصيل Prompt يتجاوز الأمر المباشر Prompt المخصّص للمراجعة. افحص المصدر قبل تشغيله.
npx skills add https://github.com/a5c-ai/babysitter --skill nsight-profilerيبقى الأمر في سطر واحد. مرّر أفقيًا لمراجعته كاملًا قبل النسخ.
تفضّل نسخة محلية؟ نزّل الملفات المتاحة حاليًا لدى SkillsMP.
تحميل Zip جاري التحميل... المهن ذات الصلة SOC
استنادا إلى تصنيف SOC المهني
Reference for querying the Atlas knowledge graph through its MCP tools — the SECONDARY enrichment/comparison layer that adds best-practice context to systems you have ALREADY scanned from your real sources (`az`, repos, dirs). Use when you need to look up nodes, edges, kinds, clusters, stats, or wiki pages in Atlas to compare against your real inventory. (atlas graph, query atlas, atlas mcp, search the graph, graph neighbors, atlas record, atlas kinds, enrichment layer)
Atlas turns your STATED NEED into a real systems atlas by SCANNING your actual sources (Azure via `az`, git repos, local dirs) and process/data mining them, THEN enriching against the Atlas knowledge graph. Use this skill when asked to inventory/map your real systems, scan your cloud + repos + directories, mine the real processes or data they contain, or collect their real constraints/gotchas. (atlas, scan my systems, inventory our azure account, map my repos, real systems atlas, process mining, data mining, collect nuances, system discovery)
assimilate-popular-workflows This skill should be used when the user asks to "find skills in the wild", "assimilate popular workflows", "discover SKILL.md files in repos", "research external skills", "find workflow patterns", "survey the skill landscape", "what skills exist out there", or wants to investigate public repositories for extractable processes, babysitter plugins, and reusable procedural insights. Searches GitHub for SKILL.md files, classifies repos by archetype, and maintains structured research under docs/reference-repos/.
name nsight-profiler description Expert skill for NVIDIA Nsight Systems and Nsight Compute profiling tools. Configure profiling sessions, analyze kernel reports, interpret occupancy metrics, roofline model data, memory bandwidth bottlenecks, and warp execution efficiency. allowed-tools Bash(*) Read Write Edit Glob Grep WebFetch metadata {"author":"babysitter-sdk","version":"1.0.0","category":"performance-profiling","backlog-id":"SK-002"} graph {"domains":["domain:scientific-computing"],"specializations":["specialization:gpu-programming"],"skillAreas":["skill-area:cuda-kernels","skill-area:compute-shaders","skill-area:compiler-implementation"],"roles":["role:computational-scientist","role:ml-engineer"]}
nsight-profiler
You are nsight-profiler - a specialized skill for NVIDIA Nsight Systems and Nsight Compute profiling tools. This skill provides expert capabilities for performance analysis and optimization of GPU applications.
Overview
This skill enables AI-powered GPU profiling operations including:
Configure and execute Nsight Systems profiling sessions
Analyze Nsight Compute kernel reports
Interpret occupancy metrics and SM utilization
Parse and visualize roofline model data
Identify memory bandwidth bottlenecks
Analyze warp execution efficiency
Generate optimization recommendations from profiler data
Compare kernel performance across different configurations
Prerequisites
NVIDIA Nsight Systems 2023.1+
NVIDIA Nsight Compute 2023.1+
CUDA Toolkit 11.0+
GPU with compute capability 7.0+ (for full profiling features)
Capabilities
1. Nsight Systems Profiling System-wide performance analysis:
nsys profile -o report ./cuda_program
nsys profile -t cuda,nvtx,osrt -o report ./cuda_program
nsys profile --gpu-metrics-device=all -o report ./cuda_program
nsys profile -d 10 -o report ./cuda_program
nsys export -t sqlite,json report.nsys-rep
nsys stats report.nsys-rep
2. Nsight Compute Profiling Detailed kernel analysis:
ncu -o profile ./cuda_program
ncu --kernel-name myKernel -o profile ./cuda_program
ncu --set full -o profile ./cuda_program
ncu --set roofline -o profile ./cuda_program
ncu --section MemoryWorkloadAnalysis -o profile ./cuda_program
ncu --import baseline.ncu-rep --diff ./cuda_program
3. Occupancy Analysis Analyze and optimize occupancy:
ncu --section Occupancy -o occupancy ./cuda_program
// Query occupancy in code
int numBlocks;
int blockSize = 256;
cudaOccupancyMaxActiveBlocksPerMultiprocessor(
&numBlocks, myKernel, blockSize, sharedMemSize);
float occupancy = (numBlocks * blockSize) /
(float)deviceProp.maxThreadsPerMultiProcessor;
printf("Theoretical Occupancy: %.2f%%\n", occupancy * 100);
4. Roofline Model Analysis Performance bound analysis:
ncu --set roofline -o roofline ./cuda_program
Below memory roofline: Memory bound
Below compute roofline: Compute bound
At peak: Optimal utilization
5. Memory Bandwidth Analysis Identify memory bottlenecks:
ncu --section MemoryWorkloadAnalysis \
--section MemoryWorkloadAnalysis_Chart \
--section MemoryWorkloadAnalysis_Tables \
-o memory ./cuda_program
Global Load/Store Throughput
L1/L2 Cache Hit Rate
Shared Memory Bandwidth
Memory Transactions per Request
6. Warp Execution Analysis
ncu --section WarpStateStatistics -o warp ./cuda_program
ncu --section SchedulerStatistics -o scheduler ./cuda_program
Warp Cycles Per Issued Instruction
Eligible Warps Per Active Cycle
Active Warps Per Scheduler
Stall Reasons (memory, sync, execution)
7. Kernel Comparison
ncu -o baseline ./program_v1
ncu --import baseline.ncu-rep --diff ./program_v2
ncu --import baseline.ncu-rep \
--import optimized.ncu-rep \
--page diff --csv > comparison.csv
8. Performance Recommendations
ncu --section SpeedOfLight \
--section SpeedOfLight_RooflineChart \
-o speedoflight ./cuda_program
ncu --import profile.ncu-rep --page details --csv > details.csv
Common Profiling Workflows
Workflow 1: Initial Performance Assessment
nsys profile -t cuda -o system_overview ./program
nsys stats system_overview.nsys-rep
ncu --launch-skip 10 --launch-count 5 -o hot_kernels ./program
ncu --kernel-name hotKernel --set full -o detailed ./program
Workflow 2: Memory Optimization
ncu --section SourceCounters \
--section MemoryWorkloadAnalysis \
--kernel-name targetKernel \
-o memory_analysis ./program
ncu --metrics l1tex__t_sectors_pipe_lsu_mem_global_op_ld.sum,\
l1tex__t_requests_pipe_lsu_mem_global_op_ld.sum \
-o coalescing ./program
Workflow 3: Occupancy Optimization
ncu --section Occupancy \
--section LaunchStatistics \
-o occupancy ./program
for bs in 64 128 256 512 1024; do
ncu --section Occupancy -o occ_$bs ./program --block-size $bs
done
Process Integration This skill integrates with the following processes:
performance-profiling-analysis.js - Performance analysis workflowoccupancy-optimization.js - Occupancy optimizationwarp-efficiency-optimization.js - Warp efficiencygpu-memory-optimization.js - Memory optimization
Output Format When executing operations, provide structured output:
{
"operation" : "kernel-profile" ,
"tool" : "nsight-compute" ,
"kernel" : "matrixMultiply" ,
"metrics" : {
"duration_us" : 125.4 ,
"achieved_occupancy" : 0.78 ,
"theoretical_occupancy" : 1.0 ,
"compute_throughput_pct" : 65.2 ,
"memory_throughput_pct" : 89.3 ,
"roofline" : {
"arithmetic_intensity" : 12.5 ,
"achieved_gflops" : 4500 ,
"peak_gflops" : 8000 ,
"bound" : "compute"
}
} ,
"recommendations" : [
"Increase block size to improve occupancy" ,
"Consider loop unrolling to reduce instruction overhead"
] ,
"artifacts" : [ "profile.ncu-rep" , "summary.csv" ]
}
Dependencies
Nsight Systems 2023.1+
Nsight Compute 2023.1+
CUDA Toolkit 11.0+
Constraints
Full profiling requires root/admin privileges
Some metrics only available on specific GPU architectures
Profiling adds overhead; results may differ from production
Nsight Compute profiles one kernel invocation at a time by default
