OPEN GPU BENCHMARKSBlog, guides, and reviews
PostBy Open GPU Benchmarks

Why I'm Starting Open GPU Benchmarks

Why I started opengpubenchmarks.com in 2026 to provide open, community-driven GPU benchmarks - and what comes next.

It’s 2026 and if you spend any time online, you’ve probably heard: “PC gaming is dead.”

I don’t buy it.

PC gaming isn’t dead. It’s just in a weird, rough patch right now high hardware prices, confusing product stacks, drivers frame generation, upscalers that matter more than ever, and performance that changes wildly depending on OS, power limits, and form factor.

I started opengpubenchmarks.com because I wanted a place to actually prove that with data, and to make that data open for anyone to use.

The goal: open, usable numbers

There are plenty of great benchmark reviews out there, but almost none of them give you the raw data in a format you can do something with.

One of my core goals with this site is to provide open GPU benchmarks and stats to the PC gaming community or really, to anyone who just likes a bunch of numbers.

Everything official lives as JSON. I love that about JSON and structured data: someone else can take it and build something I never even thought of. A comparison tool, a performance-per-dollar chart, a Linux vs. Windows deep dive. If the data is open, the ideas and crazy cool stuff can come from anywhere.

Official numbers vs. community numbers

I know benchmarking is messy. There are a ton of variables you should lock down if you want reproducible results. That’s exactly what my test bench is for.

The test bench is the source of truth. It’s locked down, documented, and consistent. Those are the official numbers on the site.

But I also know I can’t cover every card. I don’t own every card. No single person does.

So I’m building an interesting middle ground: a place where you can submit your own benchmarks and have them incorporated into a larger dataset.

Your submission won’t overwrite official results, but it will add context. If we have 50 community submissions for an RX 9070 XT on 1080p Ultra, that starts to tell a real story even with variance. It allows you to do real comparisons. It also allows you be honest with yourself on if your card is really that good.

If you want to contribute, there will be a submission guide and schema so your data can be validated and merged.

The dataset I actually want to see

Selfishly, I’m building the dataset I wanted for myself.

I want to be able to answer questions like:

  • How does this game run on Linux vs. Windows on the exact same hardware?
  • What’s the real gap between a handheld, a thin-and-light laptop, and a desktop with the same GPU class?
  • How much does VRAM actually matter in titles at 1440p vs. 4K?

I figured if I want to see that data, other people probably do too.

The long-term vision is a giant, queryable dataset where you can slice by OS, form factor, resolution, and settings not just by GPU model.

How AI is used on this site

I want to be transparent about this:

1. Code: Most of the site scaffolding is built with AI assistance. You can see that in the commits and PRs. It’s been a huge accelerator.

2. Test data: AI does generate synthetic benchmark data, but that data is never published as official. It lives in test folders and is only used to test charts, parsers, and the site itself.

3. Official benchmark data: Every official benchmark number is from a run I personally did, or a run where I was able to personally validate the source and methodology. No AI-generated performance numbers make it into the main dataset.

4. Writing: I write the bulk of the content. I do use AI tools to clean up grammar or restructure a paragraph so it flows better — like this post. If an article or guide is ever fully generated by AI, it will be clearly marked in the metadata as ai_generated: true. I don’t plan to have many of those, but for some straightforward tutorials, it just makes more sense to publish an AI-generated guide than to rewrite the same thing manually.

Still figuring stuff out and v1 is live!

Today I tagged v1.

This isn’t the finish line, it’s the starting line. The site will keep changing. I’m still figuring a lot of this out in public.

For v1, I started where I had to: Cyberpunk 2077 benchmarks on an RTX 5080 from my main rig. It’s one data point, but you have to start somewhere.

I have a backlog of cards, laptops, and handhelds I want to get uploaded next including some older GPUs that are still super relevant for budget builds.

What I hope happens next

My hope is to build a small community around this. People who care about reproducible benchmarks, open data, and actually helping other gamers make informed decisions.

If nobody shows up, that’s okay too. It was a cool little project that I was only able to figure out with the help of generative AI.

And realistically: this isn’t going to become my full-time job. I don’t have any illusions about that. But hardware is expensive. If affiliate links and other small experiments can offset the cost of picking up a new video card or two that I don’t have to pay for out of pocket, that would be pretty sweet.

If that sounds interesting to you. I’d love to have you around.

Thanks for reading the first post. Let’s see where we can take this.

Links & How to Contribute