US-based Modal Plans to Adopt Nvidia Vera Processors

en.Wedoany.com Reported - On October 19, US artificial intelligence cloud computing company Modal announced plans to adopt Nvidia Vera central processing units to provide computing power for its Modal Sandboxes cloud isolated computing service. The company has received its first Vera server and completed preliminary performance testing; in high-concurrency AI task tests, the Vera processor's task throughput reached up to twice that of comparable Intel Granite Rapids systems. Modal plans to introduce it into its existing computing infrastructure to support AI agent operation and reinforcement learning tasks.

Modal previously mainly used x86 processors from Intel and AMD to provide cloud computing resources for AI developers. Its Sandboxes service can simultaneously run hundreds of containers and virtual machines on a single server, executing code compilation, file retrieval, program testing, and automation tasks. The company will continue to retain Intel and AMD processors and select hardware platforms based on different computing loads.

Modal's engineering team developed the Agentless Agent Bench testing tool, using 50 tasks derived from Terminal-Bench to test Vera, Intel Granite Rapids, AMD Zen 5, and Amazon Graviton processors. The tests covered workloads including code compilation, Git operations, file search, and device emulation. Results showed that Vera achieved lower latency in some single-task tests, and after increasing the number of concurrent containers, its task throughput advantage further expanded.

Nvidia's Vera processor uses its self-developed Olympus core architecture, is equipped with 88 CPU cores, supports 176 threads, and has LPDDR5X memory bandwidth of up to 1.2 TB per second. The processor can be used in standalone CPU servers and can also be deployed together with Nvidia GPUs on the Vera Rubin computing platform. Modal's existing infrastructure mainly runs on the x86 architecture; integrating Vera's Arm-compatible architecture requires adjustments to the server operating environment, container image builds, and related software systems.

Modal recently completed $750 million in financing, reaching a valuation of $15.75 billion. The company is working with Nvidia and cloud service providers to advance Vera processor deployment and plans to use it for AI agent and reinforcement learning tasks. Vera has completed Modal's internal testing but has not yet been fully put into its production computing cluster.