Of course, Nvidia was going to have at least one supercomputer; I mean it’s been making GPUs since Medium Knowledge for a million years. And I know that DLSS demands a safe amount of pre-training in order to make its funky speaker algorithms as robust and effective as they are. But this silly Dave only learned last week that Nvidia has a compromised supercomputer, “with thousands of our latest and greatest GPUs,” that has gone full-time for six years just to DLS quality.
During RTX Blackwell’s publishing day last week in glorious Las Vegas*, in the middle of those 2025, Brian Catanzaro, the vice president of Nvidia’s implemented unimplemented, leveled up to talk through DLSS and many tweaks and demanding situations that come with it.
As well as the game-changing switch from convolutional neural networks to the new transformer model for DLSS 4, the other thing to catch my ear was the aside Catanzaro made about how Nvidia trains its models.
“How is it that we’ve been able to make progress with DLSS over the years?” He asks. “You know, it’s been a six year, continuous learning process for us.
“In fact, we have a giant supercomputer in Nvidia, with several thousand of the last and most productive GPU, which takes position 24/7, 365 days a year for DLS. And that has been doing that for six years. “
Maybe it’s me being utterly naïve, but I didn’t realise the amount of resource that was being dedicated to making its upscaling solution better over time.
I have an idea that Nvidia can give its gangs a time committed to a multi-million dollar device so that education ends from time to time, yet I knew it lived without hiring in the brain of its own supercomputinerray
A supercomputer purely committed to crush photographs that are intelligent enough.
“What we do in this process,” Catanzaro continues, “is that we analyze chess. When DLSS style fails, it looks like ghosts, flicker, or blur. And, you know, we locate chess in a lot of games that we “watch” we watch. A and see to perceive what’s going on, why does the style make the wrong selection on how to draw the symbol there?”
“Then we locate tactics to develop our set of educational knowledge. Our educational knowledge sets are still developing. We fulfill examples of what the graphics look and what disorders are difficult to solve DLSS disorders.
“We put those in our training set, and then we retrain the model, and then we test across hundreds of games in order to figure out how to make DLSS better. So, that’s the process.”
And it is an infernal process. And according to my delight with the new transformer style in DLSS 4, it is only improving. Not only for the new RTX 50 series.
The best CPU for games: the best Intel and AMD chips. The best motherboard for games: the right plates. The best graphics card: your best pixel pusher awaits you. The best SSD for games: enters the game first.
* Las Vegas, with all honesty, is decidedly irregular, especially for an inveterate introvert as me, however, you will have to say that or will not let you out of the airport. They received strict surveillance.
Dave has been gaming since the days of Zaxxon and Lady Bug on the Colecovision, and code books for the Commodore Vic 20 (Death Race 2000!). He built his first gaming PC at the tender age of 16, and finally finished bug-fixing the Cyrix-based system around a year later. When he dropped it out of the window. He first started writing for Official PlayStation Magazine and Xbox World many decades ago, then moved onto PC Format full-time, then PC Gamer, TechRadar, and T3 among others. Now he’s back, writing about the nightmarish graphics card market, CPUs with more cores than sense, gaming laptops hotter than the sun, and SSDs more capacious than a Cybertruck.
”Put Them on the Slides”: How Jensen Huang Was Invented and Then Announced the Generation of NVIDIA DLSS AI in Two Weeks
The Spider-Man 2 PC launch misfortunes aren’t satisfied that they don’t believe it, because it turns out that the direction powered through the GPU is taken from the performance
Kingdom eats: the weapons of liberation 2 are very, very bad, but that is what makes them so fun: