The bisectional bandwidth, at 70TB/s is 11 times that of its predecessor. Nvidia has a new NVLink switch that can scale up to 256 GPUs, 32 times the size of the previous NVLink domain.
Nvidia keynote 2021 cgi software#
This enhances security and is helpful in a number of situations, including licensed software protection, running sensitive data and algorithms on 3rd party platforms, and for Federated Learning where multiple participants train a shared model by using their own proprietary data.īecause Nvidia always considers data center scaling in its architectural planning, the H100 has a new 4th generation of NVLink for coherent GPU-to-GPU that can extend across chassis.
Nvidia keynote 2021 cgi full#
The H100 GPU has AES-GCM-256 to securely encrypt/decrypt data transfers between the CPU and GPU at full PCIe line rates. The GPU supports a trusted execution environment (TEE) that secures and isolates the entire workload running on a single H100 GPU, multiple H100 GPUs within a node, or individual instances on the GPU when running in multiple instance (MIG) mode). In an important development for security and workload isolation, the H100 chip has added support for confidential computing, making it the only GPU to support it. Hopefully someday Nvidia will bestow this technology upon the rest of us, so that our next presentation for work won’t even require us to be there at all.The Hopper H100 chip and relative performance over Ampere Nvidia And as Nvidia demonstrated, creating a good replica requires a lot of hardware.īut we like the fake AI presenter. But human faces have so many complexities that subtle details can be off, like the reflections on eyes.
Nvidia keynote 2021 cgi how to#
The technology works by analyzing large amounts of footage and video of a person’s face and then figuring out how to alter it to achieve different expressions. In response to the COVID pandemic last year, the company revealed a new AI-powered suite of tools that improves the quality of video calls by learning your facial expressions so it can render them on a remote server, reducing the amount of data you use on your own computer and improving clarity for other participants even if your connection quality drops.Īrtificial intelligence still isn’t perfect at mimicking complex facial expressions. Even still, it’s quite an impressive feat that one can watch the presentation and still not be sure what parts of actual, recorded video, and which are fake.ĪI magic - Nvidia has shown other ways its powerful graphics processors can be used to influence one’s reality. Nvidia says it also applied some other “AI magic” to make his clone look realistic. Using a truck full of DSLR cameras, a full face and body scan was captured to create a 3D model, and then artificial intelligence was trained to mimic his gestures and expressions.
Granted, creating the rendered version of Huang involved a lot of work. It’s hard to actually identify the fake portion, however, which is the most impressive part. But part of the presentation showed Huang magically disappear and his kitchen explode, which made viewers wonder what exactly was real or rendered. Sleight of hand - The speech happened in April, and only about 14 seconds of the nearly two-hour presentation were animated. Nvidia, the maker of popular graphics cards, revealed yesterday that parts of a keynote speech made by its CEO were actually computer-generated animation - an entire virtual replica of Jensen Huang and his kitchen in the background. Most people hate giving presentations, but thankfully someday you might be able to give one without actually delivering it yourself at all.
0 kommentar(er)
