NVIDIA's GPU Roadmap Teases When To Expect GeForce RTX 50 Series GPUs

NVIDIA Voyager Hero
What's driving demand for NVIDIA GPUs? Intel and AMD have the #1 and #2 supercomputer spots, respectively, and prices on the company's gaming graphics cards are falling as PC gamers are underwhelmed by mediocre generational improvements since Ampere, launched in 2020. Despite those factors, the company is selling GPUs faster than ever, and the reason is because they're really, really good at AI training.

NVIDIA just held a press briefing to talk about just how great its GPUs are for these types of tasks, and there were some interesting details in said briefing, like the fact that a cluster of 3,584 Hopper GPUs were able to train a GPT-3-based AI model in just 11 minutes—an incredible speed considering the immense number of parameters that go into those kinds of large language AI models.

nvidia slide roadmap

The thing we're here to talk about today doesn't have anything to do with AI, but rather, a specific slide (reproduced above) that NVIDIA showed during the briefing. The slide seems to be a roadmap of sorts for NVIDIA's next few years of architectures, and it clearly shows that the company's next gaming GPUs will arrive in 2025 alongside a second-generation Grace Superchip.

That's interesting enough on the face of it (and directly contrary to previous rumors), but it becomes rather curious when you think about it awhile. NVIDIA's been releasing new GPU architectures on a two-year cadence for a good long while now. Ever since release of the GeForce GTX 980 in 2014, it's been a fresh batch of GeForce GPUs roughly every two years. Waiting until 2025 before we see the successor to Ada Lovelace—possibly codenamed Blackwell—will be a three-year gap.

farcry6 rtx4060 benchmark
From our GeForce RTX 4060 review.

That's not exactly the end of the world for NVIDIA nor for PC gamers. NVIDIA's last-generation Ampere-based GeForce RTX 3000 family already demolishes the current-generation game consoles, and the next-gen GeForce RTX 4000 parts are even faster—at least, from the x70 tier on up, anyway. The 'delay' (such that it is) also gives more time for NVIDIA's last- and current-generation parts to filter out of the supply chain.

We're impressed with the Ada Lovelace architecture, but the graphics cards offered using it have largely been either disappointingly-weak or frustratingly-expensive. For a lot of people looking for a graphics card upgrade right now, the best option is to buy a deeply-discounted last-generation GPU. That's not a great state for the market, but it absolutely justifies a late launch of a new series.