gpu-kernels

Benchmark GPU versus CPU performance for sparq columnar operations using wgpu compute kernels.

8|1|Updated Jun 7, 2026
One-click install
npx skills add https://github.com/sparq-org/sparq --skill gpu-kernels
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: gpu-kernels
Source: https://github.com/sparq-org/sparq/tree/main/skills/gpu-kernels
Command: npx skills add https://github.com/sparq-org/sparq --skill gpu-kernels

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires sparq-gpu, and includes scripts (resource) components.

What problem does it solve?

This Skill evaluates the performance of GPU-based operations against CPU-based operations for sparq's hot-path operations, helping to determine whether GPU acceleration is beneficial.

Core Features & Use Cases

  • Performance Measurement: Compare the performance of GPU and CPU for various operations like FILTER + count, hash-join probe, and GROUP BY COUNT+SUM.
  • Opt-in Measurement Prototype: An opt-in prototype for evaluating GPU performance.
  • Use Case: Use this Skill to measure whether a GPU outperforms a CPU for sparq's columnar primitives, considering the cost of data transfer between host and device.

Quick Start

To get started with the gpu-kernels skill, execute the following command: cargo run --release -p sparq-gpu -- query data.ttl turtle 'SELECT ?s ?o WHERE { ?s http://schema.org/name ?o } LIMIT 10'

Frequently Asked Questions about gpu-kernels

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

FAQPage Schema
How do I benchmark GPU versus CPU performance for columnar query operations?

To benchmark GPU versus CPU performance for columnar query operations, this Skill executes wgpu and WGSL compute kernels to measure hot-path primitives like FILTER, hash-join probe, and GROUP BY COUNT+SUM operations.

Does data transfer overhead negate GPU acceleration benefits for compute kernels?

Data transfer overhead between host and device is a critical factor. This Skill evaluates GPU acceleration benefits by explicitly weighing the compute speedup against the data transfer cost for columnar operations.

What do I need to run wgpu WGSL compute kernels for sparq performance measurement?

You need the sparq-gpu crate installed to execute the wgpu and WGSL compute kernels required for sparq performance measurement. You can run the prototype benchmarking scripts using the cargo release command.

Can I measure hash-join probe performance on the GPU using sparq?

Yes, you can measure hash-join probe performance on the GPU using sparq. The Skill evaluates specific hot-path operations including hash-join probes against CPU baselines to determine acceleration viability.

When should I not use GPU acceleration for sparq operations?

You should not use GPU acceleration for sparq operations when the benchmarking results indicate that data transfer overhead between host and device outweighs the compute performance gains of the GPU.