Deployable GPUaaS, from
Endpoint to Edge to Cloud

Deployable GPUaaS, from
Endpoint to Edge to Cloud

How is Juice different?

Juice is a fundamentally different approach to GPU virtualization, distinct from traditional "serverless" GPU services.

Key Characteristics
Just software - deploys to existing GPU systems and networks on prem or in the cloud
Applications stay in place, just the GPU is remote
Split and share GPU on the fly, no resets like vGPU/MIG
No code changes - just run
No k8s needed
Windows, Linux, or both together
Deploy anywhere

Juice runs anywhere - on-prem, cloud, hybrid, or bridging between. Juice can run on bare metal, in containers, on VMs - all it needs is Windows or Linux systems that have network connectivity.

No code, no restrictions
Dynamic allocation
High performance
Scalability

Deploying Juice's GPU-over-IP software gives you versatile remote virtualization that drops right in to your existing infrastructure.

Compatibility
Windows
Linux
Cross-OS
ML frameworks (CUDA API)
  • PyTorch
  • TensorFlow
  • ONNX
Graphics APIs
  • DirectX 10-12
  • OpenGL
  • Vulkan
Compatibility notes
  • CUDA is supported on all systems.
  • Graphics is supported on all systems except Linux clients.
  • Our desktop app (coming soon!) will be for Windows.

Full list of Features

Orchestrate without k8s on the server side
  • Group GPUs into pools
  • Control which users and systems access which pools
  • Use tags to match workloads and requests to GPUs with ultimate flexibility
  • Run bare metal, VMs, containers, whatever you like - Juice just needs an IP address
  • Monitor your whole GPU deployment with dashboards (coming soon!)
  • GPU utilization
  • Session management
Access GPUs through client interfaces
  • Command Line Interface (CLI)
  • "pip install juice"
  • Add GPU-over-IP access natively to your applications
  • Desktop app (coming soon!)
  • SDK (coming soon!)
Juice logo
System Status
info@juicelabs.co