After arm deal fell through, NVIDIA has new plans
For the first time since abandoning his pursuit of chip designer Arm Ltd., Nvidia Corp. Chief Executive Jensen Huang talked at length to the public about his company’s intentions on Tuesday, including the debut of the next generation of its primary product.
Huang revealed Hopper, a new series of graphics processing units that would ultimately replace Ampere, in his keynote talk at Nvidia’s annual developer conference, the GPU Technology Conference, or GTC.
“The engine of the world’s AI computing infrastructure makes a giant leap,” Huang says in a presentation streamed online for the virtual event.
The new architecture is based on four-nanometer technology created by Taiwan Semiconductor Manufacturing Co. Ltd., as has been the company’s norm in recent years.
“Twenty H100 GPUs can sustain the equivalent of the entire world’s internet traffic, making it possible for customers to deliver advanced recommender systems and large language models running inference on data in real time,” Nvidia said.
First NVIDIA event
It was Nvidia’s first major event after the firm withdrew its $40 billion bid to purchase chip designer Arm from SoftBank Group Corp. after authorities around the world questioned the transaction. Nvidia’s plan, as put out in Tuesday’s presentation, has not changed, with Nvidia continuing its drive to produce its first central-processing unit for servers utilising Arm designs, which will compete with Intel Corp.’s standard x86 server technology.
The GPU will also have second-generation multi-instance technology and will be capable of supporting seven tents on a single GPU. The firm also claims that it will be able to do so safely because of its secret computing assistance.
“Hopper brings confidential computing to accelerated computing by using a combination of hardware and software. When confidential computing is enabled or trusted execution environment is created through a confidential virtual machine that encompasses both the CPU and GPU together, the data transfers between the CPU to GPU, and GPU to GPU in a node are encrypted and decrypted at the full PCIe line rate,” Nvidia senior director of data center computing Paresh Kharya says.
“H100 also has a hardware firewall that secures the entire workload on the H100 GPU, and also isolates it between the memory and the computer engines, so that no one other than the owner of the trusted execution environment with the key can touch any of the data encoded inside it.
“This design guarantees total VM isolation and prevents access or modification by any unauthorised entity including the hypervisor, the host operating system, or even anyone with physical access.”