NVIDIA announced its latest GeForce RTX 40 series graphics cards last night, Beijing time. The first graphics cards to be announced this time are the high-end models of the 40 series, namely RTX 4080 (divided into 12GB and 16GB models) and RTX 4090. The domestic prices of the graphics cards are RMB 7199, 9499 and 12999 respectively.
In addition to bringing the latest graphics cards, NVIDIA also announced a new open source project - CV-CUDA. CV-CUDA is an open source library for building accelerated end-to-end computer vision and image processing pipelines that process images faster and more efficiently at scale.
Most internet traffic today comes from video, and more and more video is enhanced by AI special effects and computer vision tools. But using traditional computer vision tools for preprocessing and postprocessing consumes more time and computing resources.
CV-CUDA can accelerate AI effects, such as relighting, repositioning, blurring backgrounds, and super-resolution, through CV-CUDA to accelerate computer vision workloads. CV-CUDA also provides developers with more than 50 high-performance computer vision algorithms, as well as a development framework that can easily implement custom kernels and zero-copy interfaces to remove bottlenecks in AI pipelines.
Therefore, higher throughput and lower computational cost can be achieved with the help of CV-CUDA. According to the official statement, it can handle 10 times the number of video streams in the past at the same cost.
Although this is a brand new open source project, CV-CUDA still relies on the implementation of CUDA with NVIDIA's proprietary APIs and closed-source software/drivers, just like many other NVIDIA projects designed to accelerate GPU computing .
CV-CUDA can be integrated into C/C++, Python applications, and into existing deep learning frameworks such as PyTorch. According to the official plan, NVIDIA will release CV-CUDA in the form of early access in December, while planning to launch a beta version in March next year.