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

Nvidia RTX 40-series GPUs may consume a crazy amount of power

Add as a preferred source on Google

A flurry of recent rumors suggests that Nvidia’s upcoming RTX 40-series graphics cards will be even more power-hungry than what’s currently available. Leakers peg the power consumption in the range of 400W to 500W for the flagship card, which is higher than even the obscenely powerful RTX 3090.

3DCenter, who has previously covered the roller coaster of GPU prices in Europe, nailed down multiple leakers claiming the card will use at least 400W of power. That’s certainly not out of the question, as the RTX 3090 already requires 350W of power. Assuming Nvidia wants to push even more power out of the upcoming range, a 400W+ power requirement could be possible.

A close-up image of Nvidia's RTX 3080 Ti graphics card.
Image used with permission by copyright holder

Nvidia hasn’t announced anything about the RTX 40-series yet, so it’s likely that developers are still tweaking the final design. Kopite7kimi, one of the leakers who claimed a 400W+ power limit and is known for Nvidia leaks, said the upcoming range will be built on chipmaker TSMC’s 5nm node, breaking from the 8nm Samsung process Nvidia used on RTX 30-series graphics cards.

Recommended Videos

The next-generation architecture, tentatively named Lovelace, is rumored to arrive in late 2022 or early 2023. The rumor mill suggests that the graphics core powering the range will be capable of housing up to 18,432 CUDA cores, which is nearly 8,000 more than the RTX 3090.

AMD’s upcoming cards are rumored to require equally as much power. The RDNA 3 range is also rumored to consume between 400W and 500W of power with TSMC’s 5nm process. During a recent investors call, AMD CEO Lisa Su confirmed that 5nm is the goal and that the GPUs are on track for a 2022 launch.

Unlike Lovelace, RDNA 3 cards are rumored to use a multi-chip-module (MCM) GPU package. Essentially, the upcoming range is rumored to utilize multiple dies on the same package, unlike the RTX 40-series’ traditional monolithic design.

A diagram of an MCM on RDNA 3.
Image used with permission by copyright holder

Nvidia is rumored to be working on its own MCM design, currently named Hopper. Originally, rumors pegged Hopper as the successor to the current Ampere range, though recent speculation suggests Nvidia is locked on delivering Lovelace sooner.

Both new generations are rumored to offer up to a 2.5x improvement over the current generation. As for where they’ll fall in relation to each other, it’s too soon to say.

As is the case with all early rumors and speculation, you shouldn’t take this information as law. We’re still far out from launch, so AMD and Nvidia are more than likely still finalizing the design and tweaking specs to meet their price, power, and performance targets.

Based on what we know so far, however, a higher power draw will likely be something PC builders need to deal with. Nvidia pushed past the 250W ceiling with the RTX 3080, RTX 3080 Ti, and RTX 3090, surpassing even the most powerful cards from the generations that proceeded them. It’s too soon to say for sure, but you might need to invest in a new power supply when these cards finally arrive.

Jacob Roach
Former Lead Reporter, PC Hardware
Jacob Roach is the lead reporter for PC hardware at Digital Trends. In addition to covering the latest PC components, from…
As AI takes over schools, teachers are going back to basics to curb AI cheating
AI detectors kept getting it wrong, so schools are trying something that actually works.
ChatGPT on iPhone

Last year, teachers spent most of their energy trying to catch students using ChatGPT to write essays. This year, many have decided that's a losing battle, especially since the detectors meant to catch cheaters have been burning innocent students instead. 

A Bloomberg investigation found popular tools like GPTZero and Copyleaks wrongly flagged pre-ChatGPT college essays as AI-written, and one UC Davis senior was accused of cheating and handed a zero on his midterm, purely based on a detector's score, even though he wrote every word himself.

Read more
Meta’s Muse agent is getting closer, and it might take over your to-do list
Meta’s upcoming AI life manager does more than just answer questions.
Meta Logo

While most people's idea of AI is just a chatbot, agentic uses help build them as something more. Finding an email attachment, remembering a deadline, or following up on something that many people can take advantage of. Meta’s upcoming Muse agent is coming in as that solution.

Early screenshots shared by TestingCatalog show a personal assistant that can proactively handle everyday tasks and keep track of longer-term goals. The agent has reportedly moved from internal testing into a limited closed alpha, with invite-code functionality now appearing alongside its waitlist.

Read more
Chuwi reveals powerhouse Ryzen AI MAX+ PRO 495 workstation with a slick design at IFA 2026
Chuwi's new AI workstation is small in size but massive in brainpower.
Chuwi Ryzen AI MAX+ PRO 495 workstation

Running massive AI models has usually meant relying on the cloud or building a hefty rig. Chuwi's newest mini PC squeezes that kind of power into a box small enough to sit on your desk.

At IFA Berlin 2026, the company unveiled the UniBox AI495 Pro, a mini workstation built around AMD's flagship Ryzen AI Max+ PRO 495 chip, codenamed Gorgon Halo. AMD first teased this processor back in May 2026, and now Chuwi is putting it to work.

Read more