Juice Labs

Juice Labs provides GPU-over-IP software that decouples application processes from physical GPUs, enabling remote GPU sharing and pooling over standard networks. The tool intercepts CUDA, Vulkan, and DirectX 12 API calls, allowing applications to run on separate machines as if they had a local GPU. It supports dynamic and granular resource allocation, fractional GPU sharing, and works over LAN and WAN without application code changes.

Key Features

• GPU Partitioning: Dynamic VRAM and compute allocation, oversubscription, and simultaneous graphics and compute workloads.
• Cross-Platform Compatibility: Run Windows applications on Linux GPUs, and x86 applications on ARM hosts.
• SLURM Integration: Decouple workloads from physical hardware; CPU nodes can access GPU power on-demand.
• Desktop Client: GUI for managing connections to remote GPUs with profile management and performance monitoring.
• Security: TLS 1.3 encryption for data in transit.

Use Cases

• AI inference and training
• Graphics workloads (Blender, Unreal Engine)
• Engineering software (CATIA, Siemens NX)
• HPC clusters with SLURM
• Multi-cloud GPU arbitrage

Who It’s For

DevOps engineers, AI/ML practitioners, creative professionals, and cloud infrastructure teams.

Key Benefits

• No application code changes required
• Dynamic and granular GPU resource pooling
• Works with any NVIDIA GPU, including consumer cards
• Supports both compute (CUDA) and graphics (DirectX/Vulkan) workloads