Many VMs. One GPU.
No passthrough.
VMAT lets every virtual machine on a computer use the same GPU at once, on the GPU you already have. Each VM gets its fair share, and in our tests every answer checked out.
Five VMs (FreeBSD, Fedora, Debian, Ubuntu, FreeBSD) sharing one Intel HD Graphics 630 on Fedora 43. 2:19.
What we measured
How it's different
No passthrough
Passthrough gives the whole GPU to one VM. With VMAT, all your VMs use it together, and none of them owns it.
The GPU you already have
Tested on built-in Intel graphics, Apple silicon and NVIDIA cards. No data-centre GPU or special hardware needed.
Fair, and it adapts
Busy VMs share the GPU's time. When one goes idle or shuts down, the others pick up its share right away.
Checked answers
Our test program checks every GPU result against the CPU and counts the wrong ones. In the runs shown here, there were none.
Where it runs
| Host | GPU | Result |
|---|---|---|
| Linux (Fedora 43) | Intel HD Graphics 630 | 5 VMs, 100% busy, 0 wrong |
| macOS (Apple M2 Pro) | Apple GPU | 4 VMs together at 75–85% of native, 0 wrong |
| Windows 11 | Intel HD Graphics 630 | 4 VMs, 0 wrong, 92% of native |
| Linux (cloud server) | NVIDIA RTX 6000 Ada | 4 VMs at 95% of native, 0 wrong |
Guests tested:
See it on your hardware
VMAT is in early access. Tell us what GPU and host you have and what you'd run in your VMs, and we'll set up a demo.
Request a demo: vmat.dev@gmail.com