CUDA/graphics in QEMU-KVM VMs without passing the Nvidia card to them
Details
- External ID
- 49609389
- Source
- HN
- Company
- —
- Product
- CUDA/graphics in QEMU-KVM VMs without passing the Nvidia card to them
- Website domain
- github.com
- Launched
- Sept. 8, 2026
- Cohort
- —
- Upvotes
- 5
- Upvotes percentile
- 0.12998405103668262
- Tags
- —
- Fetched at
- Sept. 12, 2026, 5:46 p.m.
- Updated at
- Sept. 12, 2026, 5:46 p.m.
Description
Hi HN. I built this because I wanted a VM with a real GeForce GPU without losing the card on my desktop.It allows you to run CUDA and graphics in VMs without VFIO/vGPU by forwarding the NVIDIA driver's own ioctl surface from the guest to the host.That means stock libcuda and real Vulkan ICD run in the guest - no CUDA/Vulkan API remoting or translation. Display, LLM, CUDA, headless supported. Your desktop keeps using the same card, multiple VMs supported.Happy to go into the details of the architecture.
Enrichment
- Theme
- gpu compute and acceleration tools
- Vertical
- Horizontal
- Function
- Dev tools
- Audience
- Developer
- AI stance
- Not AI
- Project type
- Hobby / open-source project
- Normalized one-liner
- cuda and graphics in virtual machines
- Manually corrected
- False
Could you build this?
No Interception and forwarding of NVIDIA's proprietary ioctl syscall interface between guest and host VMs requires deep low-level kernel driver reverse-engineering and virtualization internals expertise.
What it would actually take: Requires developing custom guest and host kernel modules or userspace hypervisor extensions (QEMU device emulation) that proxy the undocumented NVIDIA UVM/RM ioctl surface and handle complex shared DMA/virtual memory mappings. The hard part is reverse engineering the closed NVIDIA driver protocol across driver versions to safely map GPU page tables and handle interrupts without crashing the host kernel. Demands senior systems engineers with virtualization (KVM/QEMU), Linux kernel module development, and GPU driver expertise.
Discussion
2 comments analyzed.
Competitors mentioned: API remoting approaches, Translation-based solutions
Competitors
Other products that read as similar to this one — 139 launches clear the similarity bar, closest 8 shown.
Attention rank: #125 of 140 (itself plus its competitors, highest first — normalized so YC and Product Hunt are compared fairly).
Launched 314 days after the earliest competitor.
- xVRAM Developer Preview · ph · 2026-09-18 · 1 upvotes · similarity 0.56
- CUDA-for-AMD-Windows · github · 2026-09-13 · 230 upvotes · similarity 0.56
- macuda · github · 2026-09-16 · 30 upvotes · similarity 0.54
- Holos · hn · 2026-04-20 · 56 upvotes · similarity 0.52
- VRAMGlass · ph · 2026-09-19 · 1 upvotes · similarity 0.49
- VM-curator · hn · 2026-01-25 · 42 upvotes · similarity 0.49
- mvcc · github · 2026-09-23 · 51 upvotes · similarity 0.48
- Llama 3.1 70B on a single RTX 3090 via NVMe-to-GPU bypassing the CPU · hn · 2026-02-21 · 395 upvotes · similarity 0.48
Other launches for this product
- No other launches for this product.
Same idea, different domain
Nobody's really built a dev tools tool for Sales yet.