mrdbourke

mrdbourke / gpu-benchmarking

Public

Comparing different GPUs on various common ML and AI tasks.

28
2
100% credibility
Found Feb 08, 2026 at 17 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Scripts to benchmark NVIDIA GPUs like RTX 4090 versus DGX Spark on AI tasks including language model inference, training, image generation, and object detection.

How It Works

1
🔍 Discover GPU speed tests

You find a handy collection of tests to compare how fast different computer graphics cards handle AI tasks like chatting with smart assistants or creating pictures.

2
🛠️ Get your computer ready

You spend a little time installing a few simple tools on your machine so it can run the tests smoothly.

3
⚡ Launch your first speed test

With one command, you kick off a test and watch as it measures how quickly your graphics card processes AI thinking tasks – exciting to see real numbers appear!

4
🧪 Try tests for different tasks

You run more tests for things like training smart assistants, making images from descriptions, or spotting objects in photos, each one building a full picture of your card's strengths.

5
📊 Check out your personal results

Your tests save easy-to-read summaries showing speeds and times right on your computer, so you can see exactly how everything performed.

6
📈 Compare graphics cards side-by-side

You use the built-in tools to line up results from different cards like a speedy sports car versus a roomy family van, spotting winners for each task.

🏆 Pick the perfect graphics card

Now you know which card zooms through your favorite AI projects fastest, saving you time and helping you choose wisely for fun or work.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 17 to 28 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is gpu-benchmarking?

This Python repo lets you benchmark GPUs on real ML/AI workloads like LLM/VLM inference with llama.cpp or vLLM, fine-tuning small models like Gemma, image generation with Flux or Z-Image-Turbo, and object detection training. Run scripts to generate CSV results comparing speed across tasks—think RTX 4090 vs DGX Spark on token throughput or training epochs. It's built for quick hardware comparisons without quality checks, just raw performance metrics.

Why is it gaining traction?

Unlike synthetic benchmarks, it tests practical single-user scenarios with popular engines and models, making it dead simple to extend for comparing different AI models or LLMs on your setup. Developers dig the plug-and-play scripts, Docker integration for vLLM, and auto-generated reports/animations for sharing results. As a gpu benchmark app eyeing 2025 hardware like Grace Blackwell, it cuts through vendor hype with owner-tested data.

Who should use this?

ML engineers speccing local inference rigs for daily prompting or fine-tuning. Hardware tinkerers debating RTX vs workstation GPUs for solo workflows. Teams comparing github branches of inference setups or evaluating vLLM vs llama.cpp throughput before scaling.

Verdict

Grab it if you're shopping GPUs—solid baselines despite 19 stars and 1.0% credibility score from limited adoption. Docs guide setup well, but expect tweaks for your stack; it's mature enough for personal gpu benchmark 2025 runs, not production yet. (187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.