Skip to main content
  1. Home
  2. Computing
  3. News

These leaked GPU prototypes make the RTX 5090 look like a toy

Add as a preferred source on Google
Four power connectors on a prototype of the RTX 4090.
Chiphell

Nvidia’s best graphics cards are notorious for being power-hungry beasts, but some freshly leaked prototypes show that the company had much bigger plans. Various engineering samples surfaced on the Chiphell forums, including a prototype of the RTX 4090 with not one, not two, but four 16-pin power connectors. Will such designs ever see the light of day?

The leaker shared pictures of board prototypes going all the way back to the GTX 10 series, with the latest GPU being the RTX 4090. However, the leaker claims that the RTX 5090 was also prototyped with up to four power connectors, although such a card wasn’t among the pictured boards.

Recommended Videos

Still, let’s be real: Four 16-pin power connectors is a pretty insane number. Looking at the AD102 prototype, which corresponds to the RTX 4090 model, we’re seeing those four 16-pin connectors alongside a 45-phase power delivery. One 12VHPWR connector means a maximum 600-watt power consumption, so using four of those would’ve brought that figure up to 2,400 watts. That’s more than even the most high-end gaming PC consumes right now, and more than any power supply could possibly handle.

A prototype like this implies that Nvidia may have tried to dream (much) bigger, which links up with various whispers of a possible Titan GPU. Such a card, which many leakers claim was actually made for the Ada Lovelace lineup and then never used, would’ve ramped up both the overall specs and the power consumption.

A prototype of the GA102 GPU.
Chiphell

Would the RTX 4090 (or 5090) Titan use four 12VHPWR connectors, though? No way. The most we’ve ever seen in a consumer GPU were overclocked variants of Nvidia’s flagships, made by the company’s board partners, which sported two. It’s hard to imagine any GPU requiring that much power, nor would it be viable. We’ve heard of many cases of the 12VHPWR burning up — even in the RTX 5090 — so adding three more is hardly practical.

Older GPU models were featured too, with slightly less extreme power configurations. The Ampere generation was well represented: There was the GA102 Ampere board with four 8-pin connectors, but also the GA104 with three. Ampere cards ended up using 12-pin connectors outside of the prototype stage, but clearly, Nvidia was toying around with different designs too.

Keep in mind that we can’t take these prototypes at face value. They look legitimate, but they didn’t come directly from Nvidia. Still, assuming these are all real, it’s interesting to see that Nvidia may have tried to create a 2,400-watt GPU. Good thing those plans never came to fruition.

Monica J. White
Monica is a computing writer at Digital Trends, focusing on PC hardware. Since joining the team in 2021, Monica has written…
Meta’s detection tool fails to identify photos generated by its own Muse Image AI
Meta has created an invisible watermarking tool called Content Seal that is embedded in all images generated by the Muse Image AI.
Meta AI identification tool.

Earlier this week, Meta announced two new AI products, namely, Muse Image and Muse Video. As the name suggests, these are generative AI tools for making photos and video clips using natural language text prompts. Soon after their rollout commenced, these tools sparked controversy because Meta had automatically opted in Instagram users, allowing others to use their publicly posted media and convert them into remixed AI content. But it appears that Meta courted another loss on its side of the court.

What's the problem?

Read more
Your Google AI Studio apps can finally have polished, presentable web links
AI Studio web apps can now use personalized subdomains
google ai studio logos

Google AI Studio has made building a web app surprisingly easy. You can describe what you want, refine the design through prompts, and publish the result without setting up a traditional development environment. An awkward point of friction comes after deployment, when the finished app still has to live behind a long, forgettable Cloud Run link.

Google is now cleaning up that final step. AI Studio lets you assign a deployed web app a personalized address under the “ai.studio” domain, such as “your-app-name.ai.studio.” A recognizable URL should make the project look more presentable in a portfolio, client demo, social post, or internal project page.

Read more
You can now check if a Google ad was made using AI
Google will auto-label its own AI ads, but third-party AI ads still rely on advertisers to come clean.
google-ads-ai-label

Ever looked at an ad and wondered if a real person made it or if it was AI generated in seconds? Google is now giving you a way to find out.

The company just announced a new AI transparency label that tells you whether an ad was created or edited using generative AI tools. The label lives inside Google's My Ad Center, and it is rolling out across Google Search, YouTube, and Discover globally.

Read more