菜单
Configure NVIDIA MIG (Multi-Instance GPU) partitions on the DGX Station GB300, including enabling MIG mode, choosing a profile layout, creating instances, and retrieving MIG UUIDs. Use when the user asks to partition the GB300, set up MIG, run multiple models in isolation on one GPU, or reconfigure existing MIG instances.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Diagnose common DGX Station GB300 issues — CUDA crashes, wrong-GPU targeting, vLLM/SGLang container bugs, MIG state problems, NVLink/Fabric Manager errors, X/Vulkan failures, HuggingFace auth, and port conflicts. Use when the user reports a GPU error, inference server crash, MIG problem, or any unexplained DGX Station failure.
Deploy an SGLang inference server on an NVIDIA DGX Station GB300 with the cu130 container, RadixAttention prefix caching, and structured JSON output support. Use when the user asks to serve a model with SGLang, start an SGLang endpoint, or needs structured-output inference on DGX Station.
Deploy a vLLM inference server on an NVIDIA DGX Station GB300 with validated container, GPU targeting, and tuning parameters. Use when the user asks to serve a model with vLLM, start a vLLM endpoint, or set up OpenAI-compatible inference on DGX Station.
Teaches the analyst agent how to write correct, robust Python analysis code for FHIR clinical data using pandas, matplotlib, and scipy.
Prepare a complete clinical case summary for a patient from FHIR endpoints. Use when asked to summarize a patient, compile a case, or prepare for tumor board.
How to delegate clinical tasks to specialist agents. Always use sub-agent runtime with explicit agentId — never ACP. Never call FHIR via web_fetch.
| name | mig-configure |
| description | Configure NVIDIA MIG (Multi-Instance GPU) partitions on the DGX Station GB300, including enabling MIG mode, choosing a profile layout, creating instances, and retrieving MIG UUIDs. Use when the user asks to partition the GB300, set up MIG, run multiple models in isolation on one GPU, or reconfigure existing MIG instances. |
| metadata | {"publisher":"nvidia","hardware":"DGX Station GB300"} |
Configure MIG (Multi-Instance GPU) partitions on the DGX Station GB300.
Find the GB300 GPU index. Run:
nvidia-smi --query-gpu=index,name --format=csv,noheader
Check current MIG state:
nvidia-smi -i <GB300_INDEX> -q | grep -i "MIG Mode"
If MIG is already enabled, show current instances:
nvidia-smi mig -lgi -i <GB300_INDEX>
nvidia-smi mig -lci -i <GB300_INDEX>
If the user wants to reconfigure, destroy existing instances first (step 6).
If MIG is not enabled, enable it. All GPU processes must be stopped first:
# Check for running GPU processes
sudo fuser -v /dev/nvidia*
# Enable MIG
sudo nvidia-smi -i <GB300_INDEX> -mig 1
# Verify
nvidia-smi -i <GB300_INDEX> -q | grep -i "MIG Mode"
Show available profiles and help the user choose a layout:
nvidia-smi mig -lgip -i <GB300_INDEX>
Common GB300 MIG profiles:
| ID | Profile name (driver-dependent) | Approx. memory | Use case |
|---|---|---|---|
| 19 | 1g.35gb (59x) · 1g.31gb (61x) | ~30 GB | Small models (7-8B), dev/test |
| 20 | 1g.35gb+me · 1g.31gb+me | ~30 GB | Same + media extensions |
| 15 | 1g.70gb | ~68 GB | Slightly larger inference |
| 14 | 2g.70gb | ~68 GB | Medium models (14-30B) |
| 9 | 3g.139gb (59x) · 3g.126gb (61x) | ~137 GB | Large models (70B quantized) |
| 5 | 4g.139gb · 4g.126gb | ~137 GB | Large models, more compute |
| 0 | 7g.278gb (59x) · 7g.251gb (61x) | ~276 GB | Full GPU as single instance |
Profile names depend on your driver version; the profile IDs do not. Always read the exact names and sizes on your box with
nvidia-smi mig -lgip -i <GB300_INDEX>, and create instances by ID. (Driver 59x reports the…35gb/139gb/278gbnames; 61x reports…31gb/126gb/251gbfor the same IDs.)
Suggest layouts based on the user's workload (use the stable IDs). Examples:
3g + two 1g.70gb → IDs 9,15,151g → IDs 19,19,197g → ID 0MIG layouts are constrained by fixed memory-slice placement, not just total memory — never sum nominal GB and assume any combination fits. A
3g + 2g + 2glayout (9,14,14) is not realizable, for example, because the second2ghas no legal placement after a3g. Andnvidia-smi mig -lgipFree/Totaltracks compute (GPC) slices, so it overstates the number of instances you can actually create (QA observed only 3 creatable1ginstances on a 61x Station even thoughFree/Totalreported 7). Always validate a specific layout withnvidia-smi mig -lgippbefore relying on it.
Ask the user what models they want to run before suggesting a layout.
Create (or recreate) instances:
If reconfiguring, destroy existing instances first:
sudo nvidia-smi mig -dci -i <GB300_INDEX>
sudo nvidia-smi mig -dgi -i <GB300_INDEX>
Then create the new layout:
sudo nvidia-smi mig -cgi <PROFILE_IDS> -C -i <GB300_INDEX>
Get the MIG device UUIDs:
nvidia-smi -L
Note the MIG-<uuid> entries — these are used to target specific MIG instances.
Show the user how to use MIG devices:
# Bare metal
export CUDA_VISIBLE_DEVICES=MIG-<uuid>
# Docker
docker run --gpus '"device=MIG-<uuid>"' ...
Report the final layout to the user with UUIDs and suggested docker commands for each instance.
If the user wants to return to full-GPU mode:
# Stop all workloads using MIG instances first
sudo nvidia-smi mig -dci -i <GB300_INDEX>
sudo nvidia-smi mig -dgi -i <GB300_INDEX>
sudo nvidia-smi -i <GB300_INDEX> -mig 0
Do not run
nvidia-fabricmanageron DGX Station. It has a single GB300 over NVLink-C2C (no NVSwitch fabric), so Fabric Manager is not installed andsystemctl start nvidia-fabricmanagerfails with "Unit not found." NVLink-C2C re-initializes automatically after MIG is disabled. If MIG mode is stuck in a "pending" state, reset the GPU instead:sudo nvidia-smi -i <GB300_INDEX> --gpu-reset.