Nvidia wants to be a central cloud. Although its history is in graphics cards for gaming enthusiasts, it has recently focused on data center GPUs for artificial intelligence, learning inference, inference, and visualization. Today, at its GTC conference, the company announced the latest configuration of its RTX server for Hollywood studios and other users to quickly generate visual content.
A full RTX server pod can support up to 1280 Turing. GPU on 32 RTX blade servers. This represents 40 GPU per server, each server occupying 8U space. The GPUs here are Quadro RTX 4000 or 6000 GPUs, depending on the configuration.
"NVIDIA RTX servers, which include fully optimized software stacks available for rendering, games, VR and AR, and Optix RTX professional viewing applications, can now provide improved cinematic quality graphics through ray tracing. a much lower cost of electricity for cluster rendering based on the same performance, "notes the company in its announcement today.
All this power can be shared by multiple users and the back-end storage and network interconnection is powered by Mellanox technology, purchased by Nvidia earlier this month for $ 6.9 billion. This acquisition and today's news clearly show how important the data center has become for Nvidia's future.
Systems manufacturers like Dell, HP, Lenovo, Asus and Supermicro will offer their customers RTX servers, all validated by Nvidia, and supporting the company's software tools to manage the workloads that run on them.
Nvidia also points out that these servers would work very well to run AR and VR applications at the edge, and then serve the images to customers on 5G networks. I think it's a bit too buzzwords, and consumer interest in RA and VR remains questionable, while 5G networks also remain far removed from traditional systems. Nevertheless, these servers have a role to play in feeding cloud game services, for example.