Skip to main content
  1. Home
  2. Computing
  3. Virtual Reality
  4. News

A mystery AMD graphics card just beat an Nvidia RTX 2080 Ti by almost 30%

Add as a preferred source on Google
OpenVR benchmark table
Image used with permission by copyright holder

An unnamed and entirely unknown AMD graphics card has appeared in the OpenVR benchmark paired up with an unnamed AMD CPU. That mystery is only compounded by the result, as together they managed to beat every Nvidia RTX 2080 Ti on the table, including heavily overclocked ones, by a sizeable margin. Is this the “Big Navi,” GPU that team red fans have been waiting for? It sure seems like it.

Since the release of AMD’s RX 5700 graphics cards in July 2019, speculation about when it would launch a new high-end graphics card has gathered traction. There have been rumors of AMD staff testing an alleged “Nvidia Killer” graphics card, and CEO Lisa Su has repeatedly stated that we can expect AMD to return to the top of the graphics card performance spectrum. But with no announcement of anything close to “Big Navi,” at CES 2020, many suspected we’d be waiting a long time to learn more. With this leaked result, perhaps not.

Recommended Videos

Discovered by Reddit user MuchCharles, the mysterious AMD graphics card appeared in the OpenVR benchmark results testing on an HTC Vive headset at 90Hz, with a resolution of 1,512 x 1,680. It managed a score of 103.32. For comparison, the most heavily overclocked Nvidia RTX 2080 Ti at those specifications managed a score of just 88.10.

Although the CPU was listed as unknown by the benchmark, VideoCardz broke down the code to discover that it is most likely an AMD Ryzen 7 4800H, paired up with a supremely powerful graphics card. Or two.

There are some caveats here, such as OpenVR being a very new benchmark and is not as widely used as other testing software, like UL’s 3DMark, so it probably doesn’t have the broadest section of systems for comparison. Engineering samples do have a tendency to throw testing software off, too. And there’s also a slight chance that we’re looking at a super-powerful Nvidia graphics card, but that the benchmark detected the 4800H’s onboard graphics as the GPU and named it as such.

But the creator of the benchmark, certainly thinks the GPU is AMD and draws some other interesting conclusions.

They make it clear that they don’t know any more than us, and can’t look into the results any more than the score and listed hardware. That said, they point out that standard 2080 Ti graphics cards tend to get around a score of 80, and the 88.1 score of the top 2080 Ti is a heavily overclocked version. That suggests that this unnamed Radeon GPU is around 17% faster than the top 2080 Ti, and close 30% faster than a stock 2080 Ti. That puts it outside the realms of anything we’ve ever seen from AMD or Nvidia.

The OpenVR author highlights that it might be possible for a heavily overclocked, liquid-nitrogen-cooled Titan RTX or 2080 Ti to achieve such a score, but that seems unlikely since the benchmarker was using a standard (albeit engineering sample) laptop CPU. If they were opting for maximum possible performance, they’d have used a desktop CPU and performed a similar overclock on that component too, surely?

Considering it’s extremely unlikely for someone to have access to both an engineering sample of an unreleased AMD CPU, and an Nvidia GPU at the same time, it is probable that what we’re looking at is an engineering sample of an unannounced AMD graphics card. That means it very well could be Big Navi. It could be a pair of 5700 XTs in crossfire too… but that seems less likely, as crossfire is effectively useless in modern gaming.

If it is Big Navi, and AMD has a GPU that can beat a 2080 Ti by almost a third, that’s huge news for team red fans. It might not be enough to take the performance crown when Nvidia reveals its Ampere GPUs — which will also be based on 7nm — later this year. But this would still be by far the most capable graphics card AMD has ever released, and give gamers and enthusiasts an option when it comes to high-end performance.

It might even do something about the crazy prices we’ve seen at the top of the GPU scale for the past few years.

Jon Martindale
Jon Martindale covers how to guides, best-of lists, and explainers to help everyone understand the hottest new hardware and…
Moonshot tells Kimi AI to code away, but leave the hacking alone
Moonshot is clipping Kimi AI’s hacking wings before letting it loose
Moonshot AI

Chinese AI startup Moonshot AI is deliberately limiting the hacking capabilities of its latest artificial intelligence model, highlighting a growing dilemma for the industry: how do companies make AI more powerful without making cyberattacks easier to carry out?

The company introduced Kimi K3.1 Pro on October 24, describing it as a highly capable, open-weight model with around 3 trillion parameters. Unlike conventional AI services that operate entirely through a provider's servers, open-weight models allow developers to download the underlying model files, run them on their own systems, and modify how they work.

Read more
Google changes its Workspace privacy promise on ads, and opting out limits Personal Intelligence
Smarter AI answers, with a side order of ads.
AI mode in Google Search

There’s no such thing as a free recommendation. Google’s latest policy tweak is a clear reminder. Google has long promised that your Gmail, Drive, and Google Photos content never helps it personalize ads. 

However, since October 16, 2026, its privacy policy adds a catch. The promise lapses if you let other ad-showing Google services use that content. That means using Personal Intelligence, an opt-in tool that lets Google’s AI search read your connected apps, does exactly that (via Google).

Read more
OpenAI is locking its most advanced AI models offline after a string of security scares
OpenAI’s fake internet is a safety fix and an admission.
openai-anthropic-logo-on-screen

Part of me is relieved that an AI lab is hitting the brakes. The other part wishes it hadn’t taken real incidents to get there.

OpenAI said Friday that its most advanced models will now train without live internet access. Instead, they’ll practice on a cached, simulated copy of the web.

Read more