How One RTX 3060 Served 4 Virtual Machines at Once
#nestri #virtionvgpu #nvidia
Nestri’s experimental virtio-nvgpu let four Linux virtual machines share one GeForce RTX 3060 without handing the entire GPU to any single guest. This breakdown explains how forwarded NVIDIA driver requests, translated resources, and shared memory delivered 103.69 FPS across four VMs.
The benchmark did not turn one GPU into four. Each guest received roughly one quarter of the card’s output, physical display support remains out of scope, and strong multi-tenant isolation is unfinished. We also examine streaming performance, NVIDIA vGPU licensing, driver requirements, and the limits behind Nestri’s results.
🔗 The project:
virtio-nvgpu (Nestri's guest driver and Rust host forwarder)
Nestri (the developers who published it)
nesprobe (the checked-in NVIDIA structure layout tables)
⚡ Also mentioned:
• NVIDIA vGPU licensing guide (per-user pricing, supported boards)
• NVIDIA RTX vWS (the $250/user/year workstation tier)
• VFIO (the passthrough approach this replaces)
• NVENC (the encoder the guests stream with)
• Slime Rancher 2 (the one game run through it)
#nestri #virtionvgpu #nvidia #gpupassthrough #vgpu #kvm #qemu #linux #homelab
👇 SUBSCRIBE & WATCH NEXT
Subscribe for a new deep-dive every week: / @cloudaicode
User Queries :
share one gpu between virtual machines
gpu passthrough without losing host display
nvidia vgpu license cost per user
can geforce cards do vgpu
nestri virtio nvgpu explained
run multiple vms with gpu acceleration
rtx 3060 multiple virtual machines graphics
free alternative to nvidia vgpu
gpu paravirtualization vs passthrough performance
cloud gaming server on one gpu