skills.rest
maruakshay
maruakshayCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

gpu-infrastructure-security

Review GPU infrastructure for shared tenancy risks and VRAM residue.

4|Updated Apr 27, 2026
Search Tags:#security#infrastructure#gpu#multitenancy#vram#imds
One-click install
npx skills add https://github.com/maruakshay/mii-ai-security --skill gpu-infrastructure-security
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: gpu-infrastructure-security
Source: https://github.com/maruakshay/mii-ai-security/tree/main/skills/gpu-infrastructure-security
Command: npx skills add https://github.com/maruakshay/mii-ai-security --skill gpu-infrastructure-security

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

GPU infrastructures often run multiple workloads on shared GPUs, leading to VRAM residues, non-isolated CUDA contexts, and potential credential exposure. This Skill provides guidance to identify, mitigate, and validate isolation gaps to reduce cross-tenant data leakage and privilege escalation in GPU compute environments.

Core Features & Use Cases

  • VRAM isolation checks: verify memory is cleared before releasing GPU allocations and workloads initialize VRAM explicitly.
  • Credential isolation and IMDS hardening: ensure credentials aren’t leaked via instance metadata and limit cloud credentials exposure.
  • Deployment governance: enforce tenancy policies, auditing, and patching cadences for GPU drivers and firmware.
  • Use Case: In a multi-tenant cloud cluster, apply these controls to prevent a following workload from reading remnants of a previous job’s activations or weights.

Quick Start

Audit GPU tenancy to ensure VRAM isolation and IMDS hardening on all GPU nodes.

Frequently Asked Questions about gpu-infrastructure-security

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I prevent cross-tenant GPU data leaks in a multi-tenant cloud cluster?▼

Prevent cross-tenant GPU data leaks by enforcing explicit VRAM clearing before releasing allocations and isolating CUDA contexts. This ensures subsequent workloads cannot read remnants of previous job activations or weights.

What is VRAM residue and how does it expose data in shared GPU environments?▼

VRAM residue occurs when shared GPUs do not explicitly clear memory between workloads. This leaves previous job activations and weights in memory, allowing following workloads to read the leaked data.

How do I audit GPU infrastructure for tenancy isolation and credential exposure?▼

Audit GPU infrastructure by verifying VRAM clearing, checking IMDSv2 hop limits, minimizing node IAM permissions, and validating CUDA context isolation. This confirms workloads are isolated and credentials are protected.

Does IMDSv2 with hop limit secure credentials on multi-tenant GPU nodes?▼

IMDSv2 with hop limit secures credentials on multi-tenant GPU nodes by preventing unauthorized metadata access. This limits cloud credential exposure via instance metadata that could be accessed by non-isolated CUDA contexts.

What are the limitations of relying solely on CUDA contexts for GPU isolation?▼

Relying solely on CUDA contexts leaves vulnerabilities like VRAM residue and instance metadata exposure. Effective GPU isolation requires dedicated GPUs, explicit VRAM clearing, and IMDSv2 enforcement to prevent privilege escalation.

When do I need to enforce explicit VRAM clearing for GPU workloads?▼

Enforce explicit VRAM clearing for GPU workloads whenever operating in multi-tenant cloud clusters where shared GPUs process sequential jobs. This prevents cross-tenant data leakage from residual activations or weights.

Related Skills

Explore More Software Engineering Skills→
coreweavecoreweave
7

add-gpu-nodes-to-gke

Automate GPU node pool creation, driver setup, and Slurm configuration on GKE.

Official
Advanced
adryanmoldokkr32-pixeladryanmoldokkr32-pixel

canary

Monitor live applications for post-deploy errors, latency spikes, and page failures.

Community
Advanced
aws-samplesaws-samples
18

agentic-eks-bootstrap

Automate AWS EKS cluster setup with GPU node pools and observability.

Official
Advanced
ChanghwanKChanghwanK

devops:gpu-analysis

Analyze IDC GPU cluster health and VRAM usage via kubectl and Grafana DCGM metrics.

Community
Advanced