Kamiwaza AI has announced the release of Kamiwaza v0.8.0, expanding its AI orchestration platform with new capabilities for NVIDIA DGX Spark systems. The v0.8.0 update introduces the “Two-Node Community” mode, allowing users to combine two DGX Spark units for pooled memory and compute, targeting larger-model inference without scaling to full rack-mounted data center infrastructure.
The upgrade includes automatic detection of paired DGX Spark systems connected via high-speed interconnects, so the platform treats a pair as a single logical resource. With automatic model parallelism, workloads are distributed across the pair without requiring users to manually partition neural network layers. The unified memory awareness feature enables use of the combined memory of two DGX Spark units, supporting over 256 GB unified memory for running frontier models exceeding single-node capacities.
Kamiwaza notes that v0.8.0 optimizes orchestration, scheduling, and deployment specifically for DGX Spark and NVIDIA Blackwell-class hardware. Both single-node and two-node configurations are supported, with DGX Spark-aware container management that detects system capabilities and selects a service container stack (with support for the sm121 instruction set) for model deployment.
The Kamiwaza Community Edition now integrates these two-node capabilities, aimed at democratizing access to larger model inference for developers, labs, and smaller teams. For hyperscale or colocation environments, Kamiwaza Enterprise enables cluster management across many DGX Spark units, supporting distributed edge cloud topologies with “One API” for consistent deployment and operation. Users can prototype workloads locally and migrate them unchanged to larger distributed environments for operational scale.
Installation details for both editions are available via Kamiwaza’s documentation portal. For more information, see Kamiwaza AI.
Source: Kamiwaza AI
