Ray Tracing and Upscaling Aren't Optional Anymore
Real-time ray tracing and DLSS-style AI upscaling have gone from marketing gimmicks to baseline expectations on new GPUs.
Not that long ago, “ray tracing” was basically a marketing sticker slapped on a GPU box, something that tanked your frame rate for a lighting effect you’d notice for five minutes before turning it off. That’s not really true anymore. Walk through the feature list of any new graphics card launch this year and ray tracing support plus some flavor of AI upscaling is just… there. Expected. Table stakes, the way anti-aliasing or anisotropic filtering used to be.
Nvidia deserves a lot of the credit here, mostly because DLSS actually solved the problem ray tracing created. Real-time ray tracing is brutally expensive computationally — you’re simulating how individual light rays bounce around a scene instead of faking it with baked lighting and clever shader tricks. Early implementations made that trade-off obvious: prettier reflections and shadows, much worse performance. DLSS flipped the math by rendering at a lower internal resolution and using a trained neural network to reconstruct something close to native detail. Suddenly you could have your ray-traced cake and still hit 60fps.
What’s interesting is how fast the rest of the industry has had to respond. AMD’s FSR takes a different approach — no dedicated tensor cores required, works across a wider range of hardware — and it’s improved noticeably since its first version. Intel is reportedly building similar upscaling tech into its upcoming Arc GPUs. Whatever you think of the underlying tech, the direction is unanimous: nobody wants to ship a modern GPU without an answer to “how do we make ray tracing not murder your frame rate.”
The timing is almost cruel
Here’s the frustrating part. This is genuinely the best era for GPU rendering techniques in a long time, and almost nobody can actually buy the hardware at a reasonable price. Between the ongoing chip shortage and crypto-mining demand eating up supply, GPU prices have been absurd for over a year now. You can read all the glowing reviews you want about how good ray-traced lighting looks in Cyberpunk 2077 or Control, but if the card costs double MSRP (if you can find it at all), the feature checklist is somewhat academic.
It does make you wonder about the incentive structure going forward. If AI upscaling becomes the expected way to make ray tracing viable, that pushes more value into software and less into raw silicon — which is interesting for smaller GPU makers who can’t out-manufacture Nvidia but might be able to out-train a neural upscaler. Intel entering the discrete GPU market next year with its own upscaling tech baked in is going to be a real test of that theory.
For now, if you’re shopping for a GPU (good luck), it’s worth checking not just the ray tracing hardware specs but whether the upscaling software ecosystem around a given card is actually mature. The silicon promises are only half the story; the drivers and the trained models behind them are quickly becoming the other half.