Arcfra Releases Neutree 1.1 with GPU Virtualization and Model Governance

Arcfra has released Neutree 1.1, adding GPU virtualization and extended model governance capabilities to its Model as a Service platform for enterprise AI inference, announced in a press release.

Neutree 1.1 introduces native virtual GPU support that lets users split GPU resources by memory and compute capacity. This allows one GPU to run several model instances, improving utilization for smaller workloads such as OCR, speech recognition, and embedding inference. For performance intensive large language models, full GPU passthrough remains available.

The update also provides a global view of node level GPU usage, enabling administrators to allocate and schedule resources based on real demand. With hard isolated resource partitions, multiple model instances can operate on a single GPU without interference.

Arcfra has expanded the Neutree model gateway with unified governance for internal and external models. This includes centralized monitoring of usage, API key quotas, rate limits, access scopes, logs, and security audits, all without requiring changes to existing application calls.

Foxconn was among the first enterprises to adopt Neutree for managing AI inference across its infrastructure. Neutree 1.1 is now open source on GitHub, while the enterprise edition is available through Arcfra.