The GTC event (GPU Technology Conference) of Nvidia just finished, and the keynote was packed with news. The new RTX 40-series graphics cards were obviously the headline announcements, kicking off a whole new generation of graphics cards for PC gamers.
But as usual, Nvidia’s announcement went far beyond consumer games, touching on industries ranging from new products in the world of robotics and self-driving cars to breakthroughs in the fields of medicine and science.
The big announcement that kicked off the keynote, of course, was three new graphics cards. The RTX 4090 is the flagship model, of course, which Nvidia says is up to 2x faster than the previous generation (RTX 3090 Ti) at traditional raster gaming. This massive card is based on Ada Lovelace of course, which is the basis for everything advertised on GTC. It comes with 16,384 CUDA cores and 24GB of GDDR6X memory. And, above all, the same 450 watts of power as the RTX 3090 Ti.
The RTX 4090 will be available on October 12 and will cost $1,599.
|RTX 4090||RTX 4080 16GB||RTX 4080 12GB|
|Memory||24GB GDDR6X||16GB GDDR6X||12GB GDDR6X|
|Increase clock speed||2520MHz||2505MHz||2610MHz|
|bus width||384 bit||256 bit||192 bit|
The RTX 4080 takes a step down from there, coming in two different configurations. The 16GB model will cost $1,199, and the 12GB model will cost $900.
These GPUs are a bit more manageable, at 320w or 285w, respectively. However, the RTX 4090 has more than twice as many CUDA cores as the RTX 4090 12GB model, which is a huge gap. There is a big difference in specs between these two RTX 4080 models, so it will be interesting to see how they stack up in terms of performance.
Alongside the new GPUs, Nvidia also announced the third generation of RTX, which includes DLSS 3. Like moving from DLSS to the second generation, this appears to be a substantial evolution of the technology. This time around, machine learning can predict actual frames, not just pixels, resulting in even more increased frame rates in games.
How fast? Well, Nvidia says up to four times faster, thanks to the optical flow accelerator. This technology actually bundles three different Nvidia innovations into one, including DLSS Super Resolution, DLSS Frame Generation, and Nvidia Reflex.
The other great feature of DLSS 3 is its ability to boost CPU-limited games. Nvidia showed off a demo of Microsoft Flight Simulator, which fits the bill in an extreme way. Seeing DLSS 3 increase frame rates from 64fps to 135fps is really impressive, especially in these types of games.
Nvidia says that DLSS 3 will be four times faster than conventional rendering, overall. DLSS 3 is due out in October, and Nvidia says more than 35 games and apps will support it at launch.
Lastly, Nvidia announced Portal RTX, a revamped version of the popular PC game, now with DLSS 3 and ray tracing effects.
Nvidia has been trying to get a foot in the door for the future of self-driving cars for years, but its latest product feels like a significant step in the right direction. Nvidia Drive Thor is their next generation superchip, based on the Hopper GPU architecture, paired with the Nvidia Grace CPU.
Drive Thor is the first automated platform to have an inference transformer engine.
Nvidia says Drive Thor will be available for automakers’ 2025 car models as the true successor to Drive Orin, which is currently in production. Thor takes the place of Drive Atlan, which had been announced last year.
In addition to the new chip, Nvidia also showed off Drive Concierge, a comprehensive in-vehicle infotainment system. It replaces the instrument cluster on the typical car dashboard with what Nvidia calls a “digital cockpit.” The system is compatible with Android Automotive, which allows systems to be customized by car manufacturers. Notably, the announcement comes just a few months after Apple’s landmark next-generation CarPlay announcement, which does the same for its digital instrument cluster.
Of course, in other parts of the vehicle, Drive Concierge also offers passengers features like video conferencing, streaming video, digital assistants, cloud gaming via GeForce Now, and full car views.
Of course, the entirety of the design was created in Omniverse, and Nvidia says that design in Omniverse allows car markers to fine-tune all of these aspects of the car long before it’s physical reality.
Omniverse Cloud was announced earlier this year as Nvidia’s full suite of cloud services for people building the future of the metaverse, without the need for all that performance on your real computer. New additions to the suite of services include the robotics simulation app, Nvidia Isaac Sim, as well as the autonomous vehicle simulation, Nvidia Drive Sim.
Interestingly, the Omniverse Cloud was mentioned alongside something called the Nvidia Graphics Delivery Network (GDN), which is the distributed data center that powers the Omniverse Cloud. Built on the same capabilities as GeForce Now, the company’s cloud gaming service, the GDN is the network that brings all this high-performance, low-latency graphics to anyone who needs it.
Nvidia talked at length about all the ways Omniverse and digital twins are being used, even stating that every future software-driven product must have a digital twin for testing purposes.
Nvidia announced the Jetson Orin Nano modules, the latest addition to the Jetson family of small computers built to speed up AI and robotics processes. These new “system-on-modules” claim to bring up to 40 trillion operations per second (TOPS) of AI performance, 80 times the performance over their predecessor, the Orin NX modules.
Orin Nano modules will be available starting in January, with the 8GB model starting at $199.
- Nvidia has announced two new large language models, the NeMo API for natural language AI applications and BioNeMo for applications in chemistry and biology.
- Lowe’s has announced that it will make its library of over 600 photorealistic 3D product assets free to other Omniverse creators. The company also discussed its exploration of using interactive digital twins and a Magic Leap 2 AR headset to give employees “superpowers.”
- The German rail network Deutsche Bahn has announced that it is using digital twins in the Omniverse to develop its future rail system.
- The second generation of Nvidia OVX, powered by the L40 GPU, aimed at building “complex industrial digital twins”, has been announced. L40 uses 3rd generation RT cores and 4th generation Tensor cores for these intense Omniverse workloads, and has already been outfitted for companies like BMW and Jaguar Land Rover. These new OVX systems will be available from Lenovo, Inspur and Supermicro in early 2023.
- Nvidia has announced that the H100 Tensor Core GPU has entered full production and is set to release its first Hopper-based products in October.
- In the medical world, Nvidia has announced IGX, a combined hardware and software platform designed specifically for use cases like robot-assisted surgery and patient monitoring.
- Nvidia also demonstrated how IGX also has applications in the industrial world, specifically in creating safe autonomous factories where human-machine collaboration is involved.
- A partnership between Nvidia and the Broad Institute of MIT and Harvard that brings GPU-accelerated Clara Parabricks software to the Terra biomedical data platform, enabling researchers to speed up tasks such as genome sequencing by up to 24 times. Nvidia says it is contributing its own deep learning model to “help identify genetic variants that are associated with disease.”
- Nvidia and Booz Allen have announced an “expanded collaboration” to use AI through GPU acceleration of their cybersecurity platform, based on Nvidia’s Morpheus framework.