Ejecuta cualquier Skill en Manus
con un clic

Ejecuta cualquier Skill en Manus con un clic

apu-memory-tuner

Estrellas12

Forks6

Actualizado23 de junio de 2026 a las 23:37

Inspects and tunes the shared-vs-dedicated memory split on AMD Ryzen APUs with unified memory (UMA) so larger LLMs and image-gen models fit on the iGPU, or so reserved GPU memory is returned to the CPU. Use when the user mentions Ryzen AI, Strix Halo / Strix Point / Krackan / Phoenix / Hawk Point, Ryzen AI Max, gfx1150 / gfx1151 / gfx1152, integrated Radeon, iGPU memory, UMA Frame Buffer Size, AMD Variable Graphics Memory, VGM, GTT, GART, TTM, pages_limit, amd-ttm, amd-debug-tools, "shared GPU memory", "dedicated GPU memory", carve-out, "not enough VRAM", "out of VRAM", "GPU OOM", llama.cpp on iGPU, ROCm on APU; or asks how much memory the iGPU can use, how to give the iGPU more memory, how to balance memory between CPU and GPU on UMA, or how to change the BIOS UMA reservation. Read-only diagnostics work everywhere; tuning runs automatically on Linux via `amd-ttm` and prints guided BIOS steps on Windows. Do not use for discrete Radeon cards, Intel iGPUs, or Apple Silicon -- it is APU-only.

Instalación

Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.

Ejecutar en Manus

Fuente

amd

amd/skills

Abrir repositorio de GitHub Ver repositorios del creador

Descarga

Ejecutar en Manus

Ocupaciones relacionadasSOC

Basado en la clasificación ocupacional SOC

Administradores de redes y sistemas informáticosOcupaciones informáticas y matemáticas·SOC 15-1244

Explorador de archivos

6 archivos

SKILL.md

readonly

Más de este repositorio

mismo repositorio

local-ai-app-integration

amd/skills

Integrates local AI capabilities into applications using Embeddable Lemonade. Use when the user wants to add local AI, offline AI, private AI, on-device AI, a local LLM, local chat, embeddings, image generation, speech-to-text, or text-to-speech to an app; replace or supplement OpenAI, Anthropic, Ollama, or other cloud AI APIs with a local backend; bundle AI inference into an app installer; or mentions Lemonade, `lemond`, embeddable lemonade, Ryzen AI, NPU/iGPU/dGPU inference, or auto-optimizing local AI.

2026-07-2012

local-ai-use

amd/skills

Routes image generation, text-to-speech, and speech-to-text through a local Lemonade Server so the agent stops paying for cloud calls on those modalities. Use when the user wants to save tokens, save cost, or save money using local AI; default to local, offline, on-device, or private image generation, transcription, or text-to-speech in this workspace; stop using DALL-E, Whisper-as-a-service, ElevenLabs, or other paid multimodal APIs; route the agent's image, TTS, or STT tool calls to a local model; or mentions Lemonade Server, OmniRouter, SD-Turbo, kokoro, Whisper, Ryzen AI, NPU/iGPU/dGPU inference, or "use local for images but cloud for chat". Run once per workspace; the rule it installs handles every later request.

2026-07-1712

tracelens-analysis-orchestrator

amd/skills

Orchestrates modular PyTorch profiler trace analysis with TraceLens: generates perf reports, prepares category data, runs system-level and compute-kernel subagents in parallel, validates outputs, and writes a prioritized stakeholder report (analysis.md). Use when the user asks to follow the analysis orchestrator, run the agentic analysis workflow, analyze a trace, compare two traces, or mentions standalone or comparative TraceLens analysis.

2026-07-1612

serving-llms-on-epyc

amd/skills

Serves a language model on an AMD EPYC CPU host using vLLM with the zentorch backend, in a container (Docker or Podman) or a conda env. Use whenever the user wants to run, serve, deploy, start, host, or launch an LLM on AMD EPYC, Zen CPU, "vLLM on CPU", "zentorch serving", or "serve a model without a GPU". Use for "serve Qwen on EPYC", "start a CPU vLLM endpoint", "run an OpenAI server on my EPYC box", or similar. Handles the full single-instance flow: detect the CPU (incl. EPYC generation), validate the runtime/env, check vLLM supports the model (via vLLM's registry, not a modality blocklist), check it fits host RAM, size CPU threads/KV/NUMA from the hardware, confirm the plan with the user, launch, and poll until the endpoint is responsive. Single instance, single socket (pinned to one socket + its memory; vLLM scales poorly across sockets). Does NOT debug failures and does NOT retry -- it reports and stops. Do not use for GPU/Instinct (use serving-llms-on-instinct) or multi-node.

2026-07-1312

magpie-kernel-evaluator

amd/skills

Performs GPU kernel correctness and performance evaluation and LLM inference benchmarking with Magpie. Analyzes single or multiple kernels (HIP/CUDA/PyTorch), compares kernel implementations, runs vLLM/SGLang benchmarks with profiling and TraceLens, and runs gap analysis on torch traces. Creates kernel config YAMLs, discovers kernels in a project, and queries GPU specs. Use when the user mentions Magpie, kernel analyze or compare, HIP/CUDA kernel evaluation, vLLM/SGLang benchmark, gap analysis, TraceLens, creating kernel configs, or discovering GPU kernels.

2026-06-2312

rocm-doctor

amd/skills

Diagnoses why ROCm, the HIP SDK, PyTorch, or llama.cpp is broken on an AMD GPU on Linux or Windows, and either applies a low-risk fix with consent or hands back the exact next step. Also routes Lemonade, LM Studio, and Ollama issues to the right upstream channel. Use when the user reports that ROCm or HIP isn't working, torch.cuda.is_available() is False Ryzen AI, rocminfo or hipInfo can't see the GPU, or hits hipErrorNoBinaryForGpu, HSA_STATUS_ERROR_INVALID_ISA, invalid device function, missing amdhip64_6.dll, vcruntime140_1.dll, or libamdhip64.so, cannot open /dev/kfd, ROCk module not loaded, an Adrenalin driver too old for the HIP SDK, or a ROCm wheel that doesn't recognize gfx1151, gfx1150, or gfx1103; or mentions HSA_OVERRIDE_GFX_VERSION, HIP_VISIBLE_DEVICES, PYTORCH_ROCM_ARCH, render-group permissions, amdgpu blacklist, Secure Boot, iGPU/dGPU collisions, or multi-GPU hangs. Do not use for non-AMD GPUs, performance tuning, or ROCm-on-WSL2.

2026-06-2312

name

apu-memory-tuner

description

APU Memory Tuner

Help the user inspect and tune the shared-vs-dedicated memory split on AMD Ryzen APUs with unified memory architecture (UMA). The user states intent ("I want to run a 70B model", "I just want my games to be smooth"); this skill picks the numbers, explains the trade-offs, and applies the change where it can.

What's actually going on

On a UMA APU (Strix Halo, Strix Point, Krackan, Phoenix, Hawk Point, etc.) the CPU and integrated Radeon share one physical pool of DRAM. There is no separate VRAM chip. Two logical knobs cut that pool up:

Knob	Where set	What it does	Reversible?
Dedicated VRAM (a.k.a. UMA Frame Buffer / carve-out / GART)	BIOS, requires reboot	Permanently reserves DRAM for the GPU. CPU can't see it.	Only via BIOS.
Shared GPU memory (a.k.a. GTT)	Kernel/driver-managed	Dynamic, OS-reclaimable cap on how much system RAM the GPU may map at a time.	Yes; Linux: `amd-ttm`. Windows: no user knob.

Key insight the rest of this skill rests on: "VRAM" and shared RAM run at the same speed on UMA, because they're the same DRAM. So the right default for AI workloads is small VRAM + large GTT, not the other way around.

When to use

Use this skill when the user wants to:

Run a model that doesn't fit ("out of VRAM", "GPU OOM" on an iGPU).
Inspect the current memory split before changing anything.
Move the slider toward more shared memory (LLMs, large image gen) or toward more dedicated VRAM (gaming, predictable framebuffer).
Revert any prior changes.

Do not use it for: discrete Radeon cards (no GTT/UMA on those), Apple Silicon (different architecture entirely), or NVIDIA/Intel GPUs.

Prerequisites

State these to the user up front so a missing one doesn't surface as a mystery script error halfway through:

GPU architecture: AMD APU with integrated Radeon. Officially tuned for RDNA3.5 (gfx1150 / gfx1151 / gfx1152); RDNA3 / RDNA2 APUs work but with conservative profile numbers.
OS: Linux (kernel + amd-ttm path) or Windows (guided BIOS path). macOS is not supported.
Linux kernel: see the live AMD matrix linked from reference.md. The detection script enforces a conservative floor (mainline 6.18.4 / Ubuntu HWE 6.17.0 / Ubuntu OEM 6.14.0).
Linux extras: pipx install amd-debug-tools for the amd-ttm CLI. The skill prints this command but never installs it silently. Reading the BIOS carve-out from dmesg typically requires sudo.
Windows extras: AMD Adrenalin Software 25.x or newer if the user wants the Variable Graphics Memory slider as an alternative to BIOS.
Reboot tolerance: any tuning change requires a reboot. Don't propose this skill mid-workload.

Silent footguns to surface when relevant:

HSA_OVERRIDE_GFX_VERSION — users running ROCm/PyTorch on an APU often set this to convince ROCm the iGPU is a supported target. It does not affect memory tuning, but if the user reports HIP error: invalid device after raising GTT, this env var is usually the cause, not the GTT change.
Windows Win32_VideoController.AdapterRAM is a 32-bit field capped at 4 GiB. If the user's reported "dedicated VRAM" is exactly 4096 MB, that's the WDDM truncation, not the real BIOS reservation. The real value lives in Task Manager > Performance > GPU.

The four-step flow

Run these in order. Each one is read-only until step 4.

[ ] 1. Detect platform and support level
[ ] 2. Show current configuration
[ ] 3. Pick a profile from the user's intent
[ ] 4. Apply (Linux) or print BIOS guidance (Windows); verify after reboot

Step 1: detect platform

python scripts/detect_platform.py

Add --json for parseable output. Exit codes:

Exit	Meaning	Next action
0	Supported AMD APU.	Continue to Step 2.
2	Wrong hardware (not an AMD APU, or unclassifiable).	Stop. Tell the user this skill can't help them.
3	AMD APU but a hard prerequisite is missing (Linux kernel too old).	Stop. Tell the user the prereq and stop. Do not attempt to upgrade the kernel from this skill.

The script reports the OS, CPU, GPU LLVM target (e.g. gfx1151), the generation bucket (RDNA3.5 / RDNA3 / RDNA2 / older), total RAM, and on Linux the kernel version vs. the minimums in the AMD doc.

Step 2: show current configuration

python scripts/show_config.py

Reports current dedicated VRAM, current shared-GPU cap, total RAM, and on Linux the raw pages_limit value plus a rocminfo sanity-check of what the runtime actually sees. Note any messages it prints — Linux often needs sudo for the dmesg read of the BIOS carve-out, and Windows' AdapterRAM field is capped at 4 GiB by WDDM (real value lives in Task Manager).

Step 3: pick a profile

Ask the user what they want, in workload terms, not numbers. Map their answer to one of these:

Profile	What it does	Use when the user says...
`large-models` (default)	GTT to ~75% of RAM, BIOS VRAM at the floor (0.5 GB).	"Run a big model", "fit Llama 70B", "I keep getting OOM on the iGPU", "give the GPU as much memory as possible".
`balanced`	GTT at the kernel default (~50%), 1 GB BIOS VRAM.	"I just do mixed dev work", "back to defaults", "don't waste RAM on the GPU".
`graphics`	GTT at default; BIOS VRAM raised to the larger of 8 GB or 25% of RAM.	"I'm gaming", "I want a predictable framebuffer", "stuttering in games".
`reset`	Revert all changes this skill made.	"Undo it", "go back to stock".
`custom`	Use the explicit `--gtt-gb` / `--vram-gb` the user passed.	"I want exactly N GB".

Default: if the user is here at all and didn't specify, they almost always want large-models -- that's the only profile that meaningfully changes the experience for the workload that brought them here (running a model that didn't fit). Use it unless they explicitly said gaming, said they want defaults, or said an exact number.

If you still can't tell, use the AskQuestion tool with the five options above labeled in plain English; do not invent a sixth.

Step 4: apply or guide

python scripts/apply_profile.py --profile <choice>

Add --dry-run first if the user wants to see the planned change before committing.

What happens on each OS:

Linux: writes the new GTT cap via amd-ttm --set <N> (which persists to /etc/modprobe.d/ttm.conf). Reboot is required; the script tells the user but never reboots automatically. If amd-ttm is missing, the script exits with a clear install hint (pipx install amd-debug-tools) — do not install it yourself without confirming with the user.
Windows: prints step-by-step BIOS instructions for the UMA Frame Buffer Size, plus a note about AMD Adrenalin's "Variable Graphics Memory" slider as an alternative on supported laptops. Nothing is written to disk or registry. The script never modifies BIOS for the user.

Then re-run python scripts/show_config.py after reboot to verify.

OS-specific reality check

Capability	Linux	Windows
Inspect dedicated VRAM	Yes (`dmesg`/`journalctl`, may need sudo)	Yes (Task Manager / dxdiag; AdapterRAM is truncated)
Inspect shared cap	Yes (`/sys/module/ttm/parameters/pages_limit`)	Yes (dxdiag "Shared Memory")
Change shared cap automatically	Yes (`amd-ttm`)	No — WDDM-managed, not user-tunable
Change dedicated VRAM automatically	No (BIOS only)	No (BIOS only; VGM via Adrenalin is the closest UI)

Net effect: on Windows, raising the BIOS UMA Frame Buffer Size is the only real way to give the GPU more memory. On Linux you almost always want to lower BIOS VRAM and raise GTT instead.

Safety rules

Never auto-reboot. Always tell the user the reboot is needed and let them do it.
Never touch BIOS programmatically. The script prints instructions; the user navigates the firmware menu.
Never silently install amd-debug-tools. Print the pipx install line and ask before running it.
Never set a profile whose validation fails. The script refuses GTT targets above 95% of RAM or VRAM targets above 50% of RAM.
Never claim a change took effect before the user has rebooted and re-verified with show_config.py.

Verification checklist

Mark this skill complete only when all are true:

python scripts/detect_platform.py exits 0.
python scripts/show_config.py reports the new values after the reboot following Step 4.
The user has tried the workload that motivated the change (loaded the model, launched the game) and confirmed the new headroom helps.
On Linux, cat /sys/module/ttm/parameters/pages_limit matches what apply_profile.py reported.

If any box is unchecked the change either didn't take effect or the user hasn't validated it yet — say so out loud rather than declaring success.

Reference

For the full glossary, the link to AMD's authoritative kernel-version / ROCm-compatibility matrix, per-OEM BIOS notes, profile math, and troubleshooting, see reference.md.