Nvidia RTX 40-series graphics cards: Everything we know

Nvidia hasn’t released any information about its upcoming RTX 40-series graphics cards, but one thing’s for sure — Nvidia will release next-generation GPUs at some point. To get you up to speed, we rounded up all of the rumblings around the virtual water cooler about RTX 40-series graphics cards, when they’re coming, and how they’ll perform.

Before diving in, it’s important to note that Nvidia hasn’t announced anything about RTX 40-series cards, otherwise known as Lovelace. The company hasn’t even confirmed that they exist. All of the information below is based on rumors, leaks, and speculation. We also have previous launches to compare to, so we only included rumors that seem likely based on previous launches.

Pricing and availability

Leaker Greymon55 says that next-gen GPUs will arrive in October 2022. That includes RTX 40-series and AMD’s upcoming RDNA 3 graphics cards. Although Nvidia hasn’t confirmed anything, September or October of 2022 makes sense for the upcoming range. RTX 30-series cards launched in September 2020, and RTX 20-series cards launched in September 2018.

Next-gen gpu will be in October next year, about a year from now.

— Greymon55 (@greymon55) September 7, 2021

Even if October is the wrong month, all signs point to a launch in the second half of 2022. Nvidia has maintained a release cadence of two years for the past several generations. It’s safe to assume we’ll see the cards this year, likely with an announcement in September and a launch in October.

Also, Nvidia is also expected to release an inter-generational update in early 2022. The RTX 30-series Super graphics card is rumored to launch in the first few months of the year, with a possible announcement coming at CES 2022. That is still only a rumor, of course.

As for the price of the RTX 40-series, we don’t know right now. Nvidia kept its pricing the same moving from the RTX 20-series to the RTX 30-series. Here’s where the current lineup stands right now:

We could see the same prices this time around, though we might see a slight increase. Going from the GTX 1080 to the RTX 2080, Nvidia bumped the price by $100. Given where GPU prices are right now, the suggested price doesn’t mean much. Hopefully that isn’t the case when RTX 40-series cards launch.

Architecture

The same leaker who pushed the release date also said that RTX 40-series graphics cards will be built on chipmaker TSMC’s N5 node. This builds on earlier rumors that Nvidia will use a 5nm node, which makes sense. Currently, RTX 30-series graphics cards use Samsung’s 8nm process, and AMD’s RX 6000 cards use TSMC 7nm process.

A report from DigiTimes further confirmed this rumor, citing industry sources saying that Nvidia will partner with TSMC and use its N5 node. Although Nvidia hasn’t confirmed anything, all signs point to Nvidia using TSMC as its manufacturing partner for the next generation.

A smaller manufacturing process means Nvidia is able to squeeze more cores on the same GPU die. The rumor mill suggests that the new die will be able to house 18,432 CUDA cores, which is nearly 8,000 more than the RTX 3090 has. That’s a massive jump, suggesting that RTX 40-series cards could see the same performance improvement as RTX 30-series ones.

There are a few caveats to that, however, which we’ll dig into in the upcoming sections. Based on rumors, TSMC’s N5 node seems likely, but it’s possible that the cards could use the N5P node, as well. This is an enhanced version of the 5nm process. Instead of straight jumps down, TSMC uses enhanced processes between each step, so 7nm goes straight to 5nm with an enhanced process in between, and so on.

Nvidia is rumored to use a monolithic design for RTX 40-series cards, while AMD’s upcoming cards are rumored to use a multi-chip-module (MCM). Nvidia will reportedly move to an MCM design with its Hopper graphics cards, which are slated to launch after Lovelace, according to rumors.

We haven’t heard anything about ray tracing or Tensor cores, but it’s safe to assume they’ll show up on Lovelace. The RT cores handle real-time ray tracing, while the Tensor cores are responsible for the AI calculations required for Nvidia’s Deep Learning Super Sampling (DLSS). These are tentpole features of Nvidia graphics cards, so we expect them to show up on RTX 40-series cards.

Performance

RTX 40-series cards are at least a few months away, so we don’t know a lot about performance still. Given the rumored core count, a card with the full die could offer as much as a 76% increase over the RTX 3090. That’s just theoretical performance, though. Multiple other factors apply, so don’t expect a 76% uplift in the real world.

One thing is for sure, we will get a huge performance improvement, including power consumption. Maybe we can get 100tflops fp32 in a single card.

— kopite7kimi (@kopite7kimi) May 22, 2021

Leaker Kopite7kimi, who has become a reliable source for GPU leaks, speculated that cards could be capable of up to 100 teraflops of FP32 compute power. It’s important to reiterate that this number is purely speculation. The leaker wasn’t making the claim based on information they had available.

Another leaker, Greymon55, backs up the claims, though. The leaker says that RTX 40-series cards will come with up to double the performance of RTX 30-series cards. That comes at a cost, though, with the most power-hungry cards rumored to require up to 500W.

Overall, rumors suggest a 2.2x increase over the current GPU core. Still, it’s too soon to say. Even if the design is finalized right now, Nvidia still needs to tweak drivers and optimize the cards to run the latest games. We don’t know where performance will be when the cards launch, but the rumor circuit suggests that it could be a big improvement.

Power

A significant boost in performance often comes with an increase in power demands. And rumors suggest RTX 40-series cards could be more power hungry than any previous Nvidia release. Various leakers have confirmed that the cards will require a lot of power, though the exact number is still up in the air.

400W seems like the minimum for the most powerful card, though some leakers suggest it could require as much as 500W. In fact, popular YouTuber Moore’s Law is Dead believes the RTX 4090 could even demand 600W of power — and that’s just the reference card, so third-party offerings could raise the bar further. For context, the RTX 3090 requires 350W of power, and Nvidia recommends at least a 750W PSU. That’s the bare minimum recommendation, so you should realistically have an 800W or higher power supply for the RTX 3090.

If the top 40-series card pushes toward 500W or 600W, you’ll likely need a 1000W power supply or higher. Nvidia pushed past the 250W ceiling with its RTX 30-series graphics cards. Previously, even the most powerful consumer cards stayed under 250W. The RTX 3080, RTX 3080 Ti, and RTX 3090 all require above 300W, while the RTX 3090 Ti can demand 450W. With the RTX 4090’s rumored 600W power draw marking a 33% increase over the RTX 3090 Ti, it suggests Nvidia isn’t too concerned with power limits for its high-end models.

Nvidia hasn’t confirmed anything, and it’s possible that power draw could drop when the design is finalized. Still, it looks like builders may need to invest in a new power supply for a high-end card for the upcoming generation.

Editors' Recommendations