NVIDIA recently announced the launch of the NVIDIA OVX servers featuring the new NVIDIA L40S GPU. They are designed to accelerate demanding applications such as AI training, 3D design, video processing and generative AI workloads, according to an Aug. 8 news release.
“As generative AI transforms every industry, enterprises are increasingly seeking large-scale compute resources in the data center,” Bob Pette, vice president of professional visualization at NVIDIA, said in the release. “OVX systems with NVIDIA L40S GPUs accelerate AI, graphics and video processing workloads and meet the demanding performance requirements of an ever-increasing set of complex and diverse applications.”
These powerful GPUs will be integrated into OVX systems offered by major global system builders, including ASUS, Dell Technologies, GIGABYTE, HPE, Lenovo, QCT and Supermicro, to enhance AI and graphics performance for a wide range of industries, the release reported.
The new NVIDIA L40S GPU is a potent, all-purpose data center processor created to speed up the most computationally intensive, complex applications, including AI training and inference, 3D design and visualization, video processing and industrial digitalization with the NVIDIA OmniverseTM platform, the release said.
Generative AI is altering workflows and services across industries, including text, picture and video generation, chatbots, game development, product design and healthcare, and it is powered by the new GPU, according to the release.
Up to eight NVIDIA L40S GPUs with 48GB of RAM apiece can be used with NVIDIA OVX systems. The L40S, which incorporates fourth-generation Tensor Cores and an FP8 Transformer Engine and is based on the NVIDIA Ada Lovelace GPU architecture, offers more than 1.45 petaflops of tensor processing capability, the release said.
In comparison to the NVIDIA A100 Tensor Core GPU, the L40S allows up to 1.2 times more generative AI inference performance and up to 1.7 times more training performance for complex AI workloads with billions of parameters and many input modalities, such as text and video. The NVIDIA L40S GPU has 142 third-generation RT Cores that provide 212 teraflops of ray-tracing capability to handle high-fidelity professional visualization operations like real-time rendering, product design and 3D content creation, the release reported.
This makes it possible for creatives to produce photorealistic material and immersive visual experiences, according to the release.
The NVIDIA L40S features 18,176 CUDA cores, providing almost five times the single-precision floating-point performance of the NVIDIA A100 GPU to speed computationally intensive tasks such as engineering and scientific simulations, the release said. CoreWeave, a leader in large-scale, GPU-accelerated workloads, is one of the first cloud service providers to deliver L40S instances.
"With the explosion of generative AI, our customers across industries are seeking powerful compute offerings and scale to match the complexity of any workload — from interactive video to AI design and automation," Brian Venturo, chief technology officer at CoreWeave, said in the release.
CoreWeave is the first specialist cloud provider to offer these new resources for quick, effective and affordable accelerated computing to fuel the subsequent wave of generative AI applications, according to the release.
"NVIDIA L40S GPUs will further expand our broad portfolio of NVIDIA solutions," the company said in the release.
