Nvidia claims 1 million times better path tracing performance is coming in future gaming GPUs — says current GPUs are already 10,000x faster than Pascal

News
By published

Thanks to rapid AI progress.

Nvidia claims 1 million times better path tracing performance in future gaming GPUs
(Image credit: Future / CD Projekt Red, Remedy Entertainment, Microsoft)

Despite increasing competition from Intel and AMD, Nvidia's RTX lineup remains the best hardware for ray tracing and path tracing in games. Ever since Turing, RTX 20 series, the company has made significant strides — mostly leveraging AI and neural rendering — to increase graphical fidelity without compromising performance. Now, at GDC 2026, it's claiming that the future holds an even more impressive milestone.

Driving Innovation and RTX Advances with John Spitzer, VP of Developer and Performance Technology - YouTube Driving Innovation and RTX Advances with John Spitzer, VP of Developer and Performance Technology - YouTube
Watch On

That's largely due to a focus on hardware-accelerated neural rendering enabled by dedicated RT and Tensor cores that handle machine learning inside Nvidia GPUs. Features like DLSS are entirely reliant on AI; the ability to piece together frame data more accurately in both upscaling and frame-gen situations is only possible due to models trained on Nvidia's supercomputers.

Article continues below

Spitzer says that Moore's Law is dead and that silicon advancements alone wouldn't be enough to generate photorealistic visuals in his lifetime. Nvidia wants to achieve a level of graphical fidelity that's indistinguishable from real life, but that would require a "hundred or thousand times more computational power" — this is where AI becomes the catalyst.

Nvidia path tracing roadmap

(Image credit: Nvidia)

In the future, AI advances will take gaming GPUs to 1,000,000 times better path tracing performance when compared to the RTX 10 series. Newer, faster, more efficient hardware blocks will basically make neural rendering the default going forward, as already claimed by Nvidia CEO Jensen Huang. Games would "look like a film" while still running smoothly due to multiple frames being interpolated in real-time by AI.

None of this is a revelation — of course, things are supposed to get better over time — but the wait might not be too long. The next-gen Rubin GPUs from Nvidia, slated to launch sometime between 2027 and 2028, could usher in this 1-million-times better path tracing reality. The list of games supporting path tracing is already growing at a rapid pace, with Resident Evil Requiem being the latest addition.

As such, the presentation also included some bits about new path tracing technologies, such as ReSTIR (recent spatiotemporal resampling algorithms) and RTX Mega Geometry. To showcase this, Nvidia brought a tech demo for Witcher 4 with over two trillion triangles in the scene, depicting realistic foliage and lighting simultaneously. Make sure to check out the video linked above for more details.

Google Preferred Source

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Hassam Nasir
Hassam Nasir
Contributing Writer

Hassam Nasir is a die-hard hardware enthusiast with years of experience as a tech editor and writer, focusing on detailed CPU comparisons and general hardware news. When he’s not working, you’ll find him bending tubes for his ever-evolving custom water-loop gaming rig or benchmarking the latest CPUs and GPUs just for fun.

27 Comments Comment from the forums
  • Faiakes
    So I'll ask again, will these GPUs be available to gamers or only to Ai farms and movie studios?
    Reply
  • coolitic
    I don't think their very large numbers has quite the "wow" impact these days, so much as it does the "inviting ridicule" one.
    Reply
  • usertests
    coolitic said:
    I don't think their very large numbers has quite the "wow" impact these days, so much as it does the "inviting ridicule" one.
    It would take some interesting math to get 100x from Blackwell within the next 2 years. They're already at 6x frame gen, does that count?
    Reply
  • Kindaian
    Not really if one counts the too high prices of hardware nowadays. So yes, they may be able to get new hardware to do amazing things, but nope, nobody but the 1% of the gamer universe will be able to afford it (if that much).
    Reply
  • moon2
    Less breathless regurgitation of nonsense. More analysis.

    10,000x is a garbage comparison. 1,000,000x is a garbage target.

    If you're baselining on decade old hardware which had no hardware support for path ray tracing and saying things got 10,000x better all you're actually saying is "we added hardware support for this new technique". If you're predicting a further 100x increase in performance and also changing the implementation from accurate to heuristic, then it's also jank.

    This is why I read anandtech by preference.
    Reply
  • DS426
    1) May be a decent technical achievement in an obscure way but it's marketing B.S.
    3) Comparing to hardware that doesn't have comparable circuitry (tensor cores) is apples to oranges, so why do it? *slow clap*
    2) Won't be able to get GPU's soon
    Reply
  • bit_user
    moon2 said:
    Less breathless regurgitation of nonsense. More analysis.

    10,000x is a garbage comparison. 1,000,000x is a garbage target.
    It's coming from the same company that always feeds us outrageous numbers! It's a good thing they actually have the best hardware, or else the ridicule would be merciless!

    They even coined the term "Huang's Law" to tout their progress on AI computing:
    https://www.tomshardware.com/news/nvidia-outlines-jensen-huangs-law-of-computing
    Reply
  • CelicaGT
    moon2 said:
    Less breathless regurgitation of nonsense. More analysis.

    10,000x is a garbage comparison. 1,000,000x is a garbage target.

    If you're baselining on decade old hardware which had no hardware support for path ray tracing and saying things got 10,000x better all you're actually saying is "we added hardware support for this new technique". If you're predicting a further 100x increase in performance and also changing the implementation from accurate to heuristic, then it's also jank.

    This is why I read anandtech by preference.
    NVIDIAs slides are as bad as Apples now. Unfortunately the venn diagram for PC gamers vs. PC enthusiasts intersects far less than one might think. Most PC gamers lap this garbage up, not knowing any better. The enthusiasts are in here shaking their heads in disappointment. Knowledge is power, most consumers are clueless. Jensen et al know this.
    Reply
  • DingusDog
    A million times sounds pretty silly but after upgrading from Ampere to Blackwell recently and seeing the increase in RT core performance and efficiency, I guess it's kind of believable. One example is VSR, my Ampere card used to chug watts and get pretty warm while the Blackwell card does it without breaking a sweat.
    Reply
  • JTWrenn
    The only interesting parts in the videos to me are that they really have put efforts into managing memory usage better. It sucks they will use that to try to justify cards with too little memory but that is where we are at.

    All the performance numbers are made up edge case silliness, but at least they recognize the mempocalypse...that they have been a big part in causing.
    Reply