SoylentNews

Nvidia Announces the RTX 4090, 4080 (16 GB and 12 GB), and More

takyon writes:

At Nvidia's GPU Technology Conference (GTC), the company announced its first "Lovelace" GPUs for consumers: the RTX 4090 ($1600), RTX 4080 16 GB ($1120), and RTX 4080 12 GB ($900). The graphics cards are made with TSMC's N4 process, and support AV1 encoding and DLSS 3 upscaling.

The RTX 4090 comes with 24 GB of GDDR6X VRAM, and launches on October 12. Performance of the 4090 should be at least 60-70% higher than the RTX 3090 Ti, or higher in some cases (raytracing performance should be better than doubled). However, Nvidia is claiming up to quadruple the performance when using DLSS 3, which will not be made available on RTX 20/30-series GPUs due to an apparent requirement of fourth generation Tensor Cores and a newer version of "Optical Flow Accelerator". The new version of Deep Learning Super Sampling (DLSS) can generate entire frames, similar to video interpolation.

The RTX 4080 variants differ in both core counts and VRAM capacity, leading to a significant performance gap between them, and will launch sometime in November. The 4080 16 GB has nearly 27% more CUDA cores and 46% higher memory bandwidth than the 4080 12 GB. The GPUs also use different dies (AD103 and AD104). This could lead you to believe that Nvidia has turned the xx70-class card into a "4080" in order to sell it at a higher price.

Nvidia also announced the Jetson Orin Nano, a system-on-module capable of 20-40 trillion operations per second, starting at $200. CNX Software has a table showing order-of-magnitude improvements in some cases over the Jetson Nano and Jetson TX2 NX.

Nvidia has cancelled its DRIVE Atlan SoC for ~2025 driverless cars, replacing it on the roadmap with a more powerful "DRIVE Thor" SoC with 2 petaflops of floating point inference performance (FP8).

AMD will announce its Radeon RX 7000 graphics cards on November 3.

NVIDIA Reveals its Next-gen Chipset for Autonomous Vehicles

upstart writes:

NVIDIA reveals its next-gen chipset for autonomous vehicles:

[...] NVIDIA says that Drive Thor can unify all the various functions of vehicles — including infotainment, the digital dashboard, sensors, parking and autonomous operation — for greater efficiency. Vehicles with the chipset will be able to run Linux, QNX and Android simultaneously. Given the vast processing power that autonomous vehicle operations require, automakers can even use two of the Drive Thor chipsets in tandem by employing a NVLink-C2C chip interconnect technology to have them running a single operating system.

In addition, NVIDIA claims that the SoC marks a significant leap forward in "deep neural network accuracy." The chipset has a transformer engine, a new addition to the NVIDIA GPU Tensor Core. "Transformer networks process video data as a single perception frame, enabling the compute platform to process more data over time," NVIDIA says. It noted that the SoC can boost inference performance of transformer deep neural networks by up to nine times, "which is paramount for supporting the massive and complex AI workloads associated with self driving."

---

Original Submission #2  Original Submission #1