The writing has been on the wall from the moment Nvidia introduced DLSS, and then followed up with DLSS Frame Generation. The company is all-in on AI, and ever since we got frame gen, hardware upgrades have stopped being the sole focus of every new release. Instead, DLSS and Multi-Frame Generation drive the hype.
AMD has FSR, of course. But its more straightforward approach is starting to look really appealing now, when you need a degree in Google search to figure out what your GPU is capable of.
I remember when we were all excited by hardware-level upgrades. More VRAM, more cores … Stuff that actually makes a difference in nearly every workload. But these days, the upgrades hinge heavily on AI, and DLSS is Nvidia's ultimate ace.
It's not that these GPUs no longer get hardware upgrades; obviously, they do. It's more that a lot of the advertised performance relies on DLSS being toggled on, and each new iteration of the tech—especially the frame gen bit—is more impressive than the previous one. Some call frame generation a crutch for poorly optimized games, or an easy out for Nvidia to focus more on software rather than hardware, but either way, DLSS is what it's all about. We're now up to DLSS 5.
The thing is though, that DLSS 5 is not just universally available to everyone who's ever laid their hands on an Nvidia card. Obviously not. And now, with the rising number of different iterations of Nvidia's signature tech, many GPU owners are left scratching their heads as they wonder whether their GPU is capable of a certain feature or not.
Right now, it's all pretty segmented. Super Resolution and Ray Reconstruction (including the second-gen transformer model that came in with DLSS 4.5), work on every RTX card from the 20-series until now. That's lovely and generous and all well and nice, but these days, most people think of frame gen when thinking of DLSS, and that's not universally available.
Frame generation needs an RTX 40 or 50, but Multi-Frame Generation (the 6x mode) is RTX 50 only. So is DLSS 5's neural rendering.
Nvidia's been gatekeeping DLSS since the beginning, but AMD used to be pretty open-minded about FSR, making the feature available on most of its newer cards. That changed with FSR 4.1, the machine-learning upscaler that launched as an RDNA 4 exclusive. And now, AMD's changed its mind yet again, expanding FSR 4.1 to RX 7000 cards.
RX 6000 owners still have that to look forward to next year, so that expansion is supposed to continue. Lighter models are reportedly in the works for RDNA 3 APUs. Unsurprisingly, like Nvidia, frame gen remains locked to RDNA 4, so this isn't a full backport.
There's a cost (of course). RDNA 3 lacks the hardware for the FP8 model that RDNA 4 runs, so it gets an INT8 version instead. Some reviewers report a performance hit based on that difference, and I'm not at all surprised. You can't easily (or at all) overcome major hardware differences.
Still, I was surprised that AMD chose to expand FSR 4.1 after initially locking it to a single generation. Full or not, the backport is great to see, and it does bring it a little more on par with Nvidia.
AMD does have other versions of FSR available, too, and some are much more widely available, to the point where even non-AMD cards can run them. But if we zoom out and look at just the best of the best, AMD's stack is less segmented right now.
From early framebuffers to AI-powered rendering — how well do you really know the GPU?
Which company released the GeForce 256 in 1999, marketing it as the world's first GPU?
Which company produced the Voodoo graphics card, one of the most iconic early 3D accelerators of the 1990s?
