September 20th, 2022, marks the next day in computing hardware announcements. Today on the docket: NVIDIA. Graphics cards. And apparently, a whole lot of talk about artificial intelligence. Last month when AMD announced their new CPUs, the presentation was very compact and very focused: These are the new products, this is how we designed them, this is how they compare to our past work and to our competitors current products, here’s the price and release date and the surrounding hardware they need. NVIDIA’s presentation, was very very different from that. The presentation was about twenty minutes of hardware discussion, followed by well over an hour of discussing the industrial implications and applications of NVIDIA’s new hardware.
There is a lot to discuss and a lot more to unpack in NVIDIA’s announcements today, a lot of really cool “the future is now” type things (and every so often a moment of not-quite-self-aware product naming, shoutout to calling their new IOT monitoring and control suite Metropolis), but let’s prune it all down to just what’s interesting to gamers. Because that’s what we do here, we’re a bunch of gamers who build computer hardware for other gamers. And what’s the news in gaming! Newer, bigger, top-end graphics cards.
The big announcement is the long-anticipated NVIDIA GeForce RTX 4090 Graphics Card, coming out the gates on October 12th (three weeks from now!) at $1599, carrying 24GB GDDR6X memory, and promising between twice and four times the effective performance of the previous king, the RTX 3090 Ti. That 4x performance boost comes from a series of architectural and processing improvements (including, most stunningly, Shader Execution Reordering, which brings the CPU theory of out-of-order execution into the realm of rendering) (… and also sending 30% more power into the core). Alongside the RTX 4090, there will also come the NVIDIA GeForce RTX 4080 Graphics Card, in both 16GB and 12GB variants, for $1299 and $899 respectively.
However, in contrast with AMD’s announcement last month, that’s basically all we got for hardware. We did not get thorough spec sheets, only a price point and a memory load, and comparative promises of performance (2x performance-per-watt, 2x rasterized graphics performance, 2x ray-tracing performance, and overclockable up to over 3GHz but they didn’t actually say the base clock speeds at any point).
What we did get, was several actually really cool gaming features. The first is a new version of NVIDIA’s Deep Learning Super Resolution (DLSR or DLSS, they go back and forth on which name to use). The previous versions of DLSS worked by taking a lower-resolution render image (say, 1080p) and using AI-trained interpolation to upscale that image to the resolution seen on screen (say, 1440p or 4k). The new form, DLSS 3, will seek to further improve gaming framerates by wholesale generating entire new frames, through AI interpolation of current and future frames without needing to run through the normal graphics calculation pipeline. (This announcement also came with another misuse of the phrase “Quantum Leap”, stop using that phrase to describe your big new product it doesn’t mean that…)
The second is mods. Game mods. They call it Omniverse RTX, and it’s a plugin package to allow anyone to take existing older games, import them into the NVIDIA Omniverse simulation environment, and then re-export a mod package to rebuild the game in RTX. They showcased this in The Elder Scrolls 3: Morrowind, which was impressive, but the real stunning one was announcing Portal, fully remastered in RTX, which they’re slating to release as a free DLC. And honestly, it looks really cool, Portal was an amazing game and upscaling it from 2007 graphics into RTX does look really good.
The gaming portion of the presentation opened and closed with footage from an NVIDIA internal game called RacerX, which looks to be a racing game focused on RC cars driving around a fully simulated world around them, at a scale of household, city, and nature park. The point that Jensen drove home the most in this game’s discussion was that every aspect of the game world is being actively simulated, rather than pre-rendered or statically built, and the simulation is being run on a single RTX graphics card.
And that simulation wording set the stage for the rest of the hour and a half long presentation. After the initial 20 minutes of gaming news, NVIDIA turned to spending the next hour on a huge discussion of their work on artificial intelligence, and their real goal: to give the word “Metaverse” an actual meaning. And NVIDIA’s meaning of Metaverse is the creation of a digitally simulated and emulated copy of every object in the physical world, so that AI-controlled robotic entities (industrial robots, self-driving vehicles, etc.) can learn to interact with the real world by first spending long spans of time learning in a virtual world. And then this also goes the other way, of letting the AI generate entities in the virtual world, and letting humans interact with them. There’s a whole lot to unpack in all of this, but I’m just going to be amused by some video clips produced by Siemens AG which basically look like they’re playing the game Satisfactory but in real life with real world factories.
NVIDIA had a lot of news and a lot to say for the wider world, but for the gaming world, the news is this: RTX 4090 and RTX 4080 graphics cards, a new version of DLSS, and the ability to, in theory, make an RTX recreation of any game, which they’re prominently demonstrating with Portal RTX. And the first of the new graphics cards will launch on October 12th, 2022. Between AMD’s new Ryzen 7000 CPUs launching next week (September 27th), NVIDIA’s new GPUs on October 12th, and Intel’s new 13th gen Raptor Lake CPUs still not yet announced but rumored for release on October 20th, we are in for an exciting end of the year.