Tpu Systolic Arrays

Use when the user asks about TPUs, systolic arrays, the MXU, XLA on TPU, JAX TPU performance, Pallas kernels, TPU padding or recompilation. Produces shape, dtype and sharding rules plus JAX code and profiling steps. Not for NVIDIA GPU kernels (use cuda-programming) or GPU multi-node training (use distributed-training).

45 stars 6 forksSoftware Development

Install

skills CLI (Claude Code, Cursor, Codex and more)
npx skills add j4flmao/agent-skills --skill tpu-systolic-arrays
Manual install
git clone --depth 1 https://github.com/j4flmao/agent-skills.git /tmp/agent-skills
mkdir -p ~/.claude/skills
cp -r /tmp/agent-skills/skills/hardware-acceleration/tpu-systolic-arrays ~/.claude/skills/tpu-systolic-arrays

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# TPU Systolic Arrays

## Purpose
A TPU spends most of its silicon on systolic matrix-multiply units (MXUs) fed by on-chip memory and HBM, controlled by a compiler (XLA) rather than by thousands of hardware threads. Performance comes from giving XLA large, static, well-aligned matrix multiplies in bfloat16 and from sharding across chips over the inter-chip interconnect. Understanding the systolic data flow explains the shape and dtype rules.

## Agent Protocol

### Trigger
"tpu", "systolic array", "mxu", "xla tpu", "jax tpu", "pallas", "tpu utilization", "tpu padding", "weight stationary", "pod slice".

### Input Context
- TPU generation and slice topology, framework (JAX, PyTorch/XLA, Keras), model shapes and dtypes.
- Profiler trace (XProf / TensorBoard profile) if diagnosing.

### Output Artifact
Shape/dtype/sharding recommendations + code + profiling steps. No file unless requested.

### Response Format
```
Bottleneck: {MXU-bound|HBM-bound|host/input-bound|recompilation|communication}
Evidence: {profile metric}
Change: {shape padding|dtype|sharding|kernel}
Expected: {metric to re-check}
```

### Completion Criteria
- [ ] No repeated recompilations in steady state.
- [ ] Matmul dimensions aligned to MXU tiles; bf16 inputs with f32 accumulation.
- [ ] Input pipeline keeps up (device not idle waiting on host).
- [ ] Profile confirms MXU utilization improved.

## How a Systolic Array Multiplies
Weight-stationary design (as described for TPU v1):
Read the full skill on GitHub

Details

Skill name
tpu-systolic-arrays
Path
skills/hardware-acceleration/tpu-systolic-arrays/SKILL.md
License
MIT
Repo updated
17 November 2026

FAQ

What does the Tpu Systolic Arrays skill do?

Use when the user asks about TPUs, systolic arrays, the MXU, XLA on TPU, JAX TPU performance, Pallas kernels, TPU padding or recompilation. Produces shape, dtype and sharding rules plus JAX code and profiling steps. Not for NVIDIA GPU kernels (use cuda-programming) or GPU multi-node training (use distributed-training).

How do I install the Tpu Systolic Arrays skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add j4flmao/agent-skills --skill tpu-systolic-arrays. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Tpu Systolic Arrays popular?

It lives in j4flmao/agent-skills, which has 45 GitHub stars and 6 forks.

Related skills

Agent Observability

45

j4flmao/agent-skills

Use when the user asks about agent observability, tracing agent reasoning, debugging an agent loop, OpenTelemetry GenAI spans for LLM calls, LLM cost tracking per agent, agent audit logs or hallucination/eval monitoring in production. Produces instrumentation config, trace schema, dashboards, alerts and cost attribution. Not for generic service tracing without LLMs (use devops-opentelemetry).

DevOps & Cloud

AI Agent Architectures

45

j4flmao/agent-skills

Use when the user asks to build an AI agent, design an agent that calls tools and keeps memory, write an agent system prompt or pick an agent framework (Claude Agent SDK, OpenAI Agents SDK, LangGraph). Produces an architecture brief covering loop, tools, memory, guardrails and evals, routing depth to child skills. Not for workflow patterns (use ai-agentic-workflows).

AI, LLMs & Agents

AI Agent Core Loop

45

j4flmao/agent-skills

Use when the user asks about the agent loop, ReAct vs plan-and-solve, re-planning, Reflexion, an agent stuck in a loop, max iterations or termination conditions. Produces a loop spec or patch with state, success predicate, caps and no-progress detection. Not for overall agent design (use ai-agent-architectures) or tool schemas (use tool-calling-principles).

AI, LLMs & Agents

Unity

45

j4flmao/agent-skills

Use when the user asks about Unity 6, MonoBehaviour, the player loop, URP, HDRP, Addressables, Netcode for GameObjects, Unity physics or Unity Profiler. Produces Unity project architecture, gameplay code and performance fixes. Not for DOTS, Entities and Burst in depth (use unity-csharp) or Unreal (use unreal).

Software Development

Symfony Backend

45

j4flmao/agent-skills

Use when the user asks about Symfony, Doctrine ORM entities, Symfony Messenger, autowiring, security voters, Flex recipes or WebTestCase. Produces Symfony 7 services, entities and tests. Not for Laravel (use php-laravel) or Laminas (use php-zend).

Software Development

Unity Csharp

45

j4flmao/agent-skills

Use when the user asks about Unity DOTS, Unity ECS with Entities, ISystem, IJobEntity, Burst jobs, the C# Job System, NativeArray or GC-free Unity C#. Produces Entities, Jobs and Burst code with correct safety rules. Not for general Unity project architecture (use unity) or engine-agnostic ECS design (use ecs-pattern).

Software Development

Browse the most popular Claude skills, all software development skills, or write your own with the skills generator.