Heterogeneous GPU Sharing on Kubernetes
KVM Backend for VirtualBox
A transparent, in-container GPU resource controller that enforces memory and compute limits by intercepting CUDA calls without application or driver changes.
Here are my personal paper reading notes (including machine learning systems, AI infrastructure, and other interesting stuffs).
Tensor Fusion is a state-of-the-art GPU virtualization and pooling solution designed to optimize GPU cluster utilization to its fullest potential.
Enable SR-IOV vGPU for consumer Ampere and Ada GeForce cards.
Paravirtual NVIDIA GPU for KVM guests — run unmodified CUDA, PyTorch and Vulkan inside a VM at host parity, on a GPU the host keeps using. No passthrough, no vGPU licence. Experimental.
Research: vGPU unlock on consumer NVIDIA RTX 5090 (Blackwell/GB202). 19 binary patches, full CPU-side pipeline working, GSP firmware blocked by fused-off VF PRIV registers.
The GPU your VM believes in. A stock NVIDIA driver runs against fabricated hardware; real compute is forwarded to a host GPU. Rust.
A GPU Virtualisation benchmarking tool - For LLM like workloads, overhead evaluations and on resource isolation metrics
TensorFusion landing page and product docs
NVIDIA vGPU on a consumer GeForce RTX 3090 (GA102) via legacy mdev. Working config, the exact kernel requirement, ~80 measured dead ends, and a tool to port the fix to other driver builds. Fixes the runlist-disable leak behind 'Immediate pteblit timed out' and silent no-render guests.
Archived - WSL2-style NVIDIA GPU ioctl forwarding for KVM/QEMU guests on Linux hosts - superseded by nvkvm-pv
Vulkan on a remote GPU over plain TCP — Mesa Venus over the internet, with a full research log
Transparent library for CUDA/NVML that virtualizes GPU access, enforces memory limits, and enables container-friendly sharing with minimal performance overhead.
Sovereign private GPU cloud platform. Slices physical GPUs into 4–8 isolated, 60 FPS in-browser AI workstations streaming full graphical Linux desktops directly to any web browser with persistent storage.
🎮 Enable GPU passthrough for KVM on full AMD systems, simplifying setup for Arch Linux and Windows 11 virtual machines.
Bitfusion Io — independent third-party profile of a public API surface, by API Evangelist. Bitfusion (bitfusion.io) was an Austin, Texas startup, backed by DCVC, that built "elastic AI" infrastructure software. Its FlexDirect platform was a transparent GPU/CPU virtualization layer that pooled network-attached accelerators (GPUs, FPGAs, ASICs) acros
Educational repository for HAMi, GPU virtualization and sharing on Kubernetes. Episode 7 of the CNCF Project Focus series.
To associate your repository with the gpu-virtualization topic, visit your repo's landing page and select "manage topics."