The last few weeks in tech coverage were dominated by Nvidia's RTX 50 series GPUs, especially the company's tall claims regarding its AI-powered DLSS 4 advancements. Nvidia has been at the forefront of leveraging AI to unlock new levels of performance on consumer and professional GPUs. With its DLSS tech still leading AMD's FSR and Intel's XeSS, it was only a matter of time before the competition stepped up.

AMD has been playing catch-up with Nvidia for a long time now. With its upcoming RDNA 4 architecture, the company finally decided to go all-in on AI, including dedicated AI hardware for the first time. The company also announced that machine learning will play a big role in FSR 4 to power frame generation. Nvidia's Multi Frame Generation (MFG) is now out in the open, and all eyes are on AMD and what it brings to the market in March.

3 AMD had to catch up with Nvidia and Intel

Got to go with the flow

AMD's ray tracing and upscaling performance might have lagged behind that of Nvidia's more mature tech, but it was also superseded by Intel in terms of dedicated AI hardware. For years, AMD's RDNA architecture has relied mostly on general-purpose shader cores to power ray tracing and upscaling, unlike dedicated RT cores and Tensor cores on Nvidia's RTX GPUs, and XMX cores on Intel's Arc GPUs.

With RDNA 4, the company decided to build the architecture from the ground up to be able to compete better with the competition. This involves implementing dedicated hardware cores for powering more advanced ray tracing, upscaling, and frame generation on its RX 9000 series GPUs. This will be a major architectural shift for Team Red; needless to say, we can expect some sweeping improvements.

Performance leaks and early impressions of FSR 4 have shown truly promising results, and it seems AMD is confident of its work with RDNA 4. The company isn't competing in the high-end segment this time around, but to stand out from both Nvidia's mid-range cards and Intel's budget offerings, AMD needed a game-changer. It seems betting everything on dedicated AI cores is what the company finally settled on.

2 AI/ML is essential to overcome the limits of native rendering

You can only shrink transistors so much

With every new generation of GPUs (and CPUs), it is expected manufacturers will opt for a smaller process node to boost performance. Of course, other factors play a role, too, but the density of the process node still determines a lot of the gen-on-gen performance boost. We've already seen Nvidia's Blackwell GPUs struggle to offer any revolutionary gains over Ada Lovelace since both architectures are essentially using the same TSMC 4nm node.

Therefore, Nvidia's reliance on AI-generated frames makes sense, since something has to give if companies need to overcome the limits of Moore's Law. An IEEE report from 2016 predicted that transistors would stop shrinking in 2021, which closely, if not completely, aligns with what we're seeing right now. AMD had to move in the same direction to avoid getting left behind.

Hence, this new normal of leveraging software advancements reliant on specialized hardware seems to be the future. The end of traditional rendering feels closer than ever before, and AMD's adoption of AI for RDNA 4 and FSR 4 only feels like a natural progression.

1 AMD already committed to AI over a year ago

AI - AMD Intelligence?

It's not like AMD's plans to bring AI to FSR 4 were crystallized overnight. According to senior leadership at AMD, the company had pivoted to AI-based FSR in 2023. It started with the intention of improving the performance and battery efficiency of AMD's mobile APUs used in gaming handhelds. While the latest Ryzen Z2 series is still based on RDNA 3.5, AMD's next lineup will most likely be based on RDNA 4 and FSR 4.

The PlayStation 5 Pro's custom AMD hardware already uses some hardware elements from RDNA 4, specifically the cores that handle ray tracing. Its PSSR upscaler also leverages AI to boost framerates. Hence, it's reasonable to expect that AMD had committed to going AI-based long before leaks and reports about RDNA 4 started appearing on the internet, meaning FSR 4 had to be AI-powered.

AMD is waiting till March to unveil its RX 9070 XT and RX 9070 GPUs, presumably to fine-tune its drivers and nail down the pricing. The company has already commented that its GPUs are achieving all the performance and power targets in the lab, and it's confident about providing attractive offerings in the mid-range segment. With all three GPU manufacturers now fully onboard the AI train, it seems we've entered a new era of GPU rendering.

The times, they are a-changin'

You might not like the new era of "fake frames" or "AI trickery" on luxury graphics cards, but it looks like they are the future. Nvidia was the first to showcase it, but it surely won't be the last. Manufacturers have to overcome the limits of traditional rendering someday, and AI seems to be the answer. Whether this pivot will benefit users or introduce new visual and performance drawbacks needs more time to answer.