# Ivx Cf Gpu Optimization

> Optimize GPU kernels, memory usage, and inference throughput for ML workloads. Use when profiling GPU code, optimizing CUDA kernels, or designing GPU-efficient models.

- Skill: `intelli-verse-x/ivx-cf-gpu-optimization` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add intelli-verse-x/ivx-cf-gpu-optimization`
- Raw SKILL.md: https://api.skillmd.com/api/skills/intelli-verse-x/ivx-cf-gpu-optimization/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: intelli-verse-x (https://skillmd.com/u/intelli-verse-x)
- Updated: 2026-09-08
- Page: https://skillmd.com/skills/intelli-verse-x/ivx-cf-gpu-optimization

---

# GPU Optimization Skill

## Purpose

Maximize GPU utilization and minimize inference latency.

## Techniques

1. **Quantization**: FP32 → FP16/FP8/INT8 (2-4x speedup)
2. **Kernel Fusion**: Combine operations to reduce memory traffic
3. **Continuous Batching**: vLLM-style batching for LLMs
4. **Speculative Decoding**: Draft model + verification
5. **KV Cache Optimization**: PagedAttention, prefix caching
6. **Tensor Parallelism**: Multi-GPU for large models
7. **Pipeline Parallelism**: Overlap compute and communication

## Profiling

- **PyTorch Profiler**: Kernel-level timing
- **Nsight Systems**: End-to-end trace
- **DCGM**: GPU-level metrics (utilization, memory, bandwidth)

## Targets

| Metric | Target |
|--------|--------|
| GPU utilization | > 70% |
| Memory bandwidth | > 60% of theoretical |
| Tensor core usage | > 50% |
| Inference latency | Within SLO budget |

## CF-Specific

- Use vLLM for self-hosted LLM serving
- Warm GPU pools for cold start elimination
- MIG for sharing H100s across services
- Spot instances for batch workloads
- Monitor with DCGM + Prometheus