● Covered by 1 source · 1 reportMedium impact1 neutral
Virtio-nvgpu Enables Near-Native Nvidia GPU Performance in KVM Guests
Key points
- KVM guests achieve 2% performance difference from bare metal.
- Uses unmodified Nvidia user-mode drivers in the guest.
- Forwards Nvidia kernel driver ioctls directly.
- Supports headless streaming and multiple guests on one GPU.
Near-Native GPU Performance
Virtio-nvgpu enables KVM guests to utilize Nvidia GPUs with performance levels comparable to bare-metal systems. Benchmarks show that a guest operating with virtio-nvgpu performs within 2% of a host machine when running an identical headless Vulkan load on an RTX 3060.
This performance parity is maintained for frames above approximately 2 ms, which covers typical game rendering scenarios. The CPU cost for a guest using virtio-nvgpu is also similar to that of the host, making shared GPU resources efficient.
Direct Driver ABI Forwarding
The technology functions by forwarding Nvidia kernel driver ioctls directly between the Linux guest and the host at the driver ABI level. This method bypasses API-level translation, allowing the guest to run Nvidia's own unmodified user-mode drivers, including libraries, Vulkan, and NVENC.
The guest's user-mode driver submits commands through memory mapped by the host, minimizing communication overhead. Over 813,691 frames, the backend processed only 13,792 messages, indicating low cross-boundary traffic primarily for device setup.
Headless Streaming and Multi-Guest Support
The primary application target for virtio-nvgpu is headless streaming, where a compositor within the VM renders, composites, and encodes frames on the GPU before sending compressed video. The VM operates without a monitor, and the host retains control of the GPU.
Testing demonstrated that four guests could run concurrently on a single RTX 3060, each maintaining consistent frame rates and encoding H.264 without reaching NVENC session limits. The total frame rate across four guests was 103.7 fps, compared to 102.9 fps for a single guest, showing efficient resource sharing.
Driver Compatibility
The system was measured with Nvidia driver 595.99.02. ABI profiles supported include 535.129.03, 580.178.04, and 595.71.05, with older versions being refused. A guest enumerates the card, reporting real power, memory, and the host's device UUID.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Primary sources
GitHub cmc_internal/api GitHub github/collect GitHub nestrilabs/virtio-nvgpu GitHub _private/browser GitHub get-started/accessibility GitHub open-source/sponsorsReporting from
Virtio-nvgpu allows KVM guests to access Nvidia GPUs with near-native performance, achieving within 2% of bare-metal speeds and similar CPU costs. This is accomplished by forwarding Nvidia kernel driver ioctls directly, enabling guests to use unmodified Nvidia user-mode drivers for headless streaming applications.