# Iommu bindings inside an instance or a container
**URL:** https://forum.verda.com/t/iommu-bindings-inside-an-instance-or-a-container/58
**Category:** Product Q&A
**Created:** [September 25, 2025, 12:21pm UTC](https://forum.verda.com/t/iommu-bindings-inside-an-instance-or-a-container/58 "2025-09-25T12:21:14Z")
**Posts on this page:** 4
**Page:** 1
### Author:  [@pixelrune](https://forum.verda.com/u/pixelrune)
#### Post date: [September 25, 2025, 12:21pm UTC](https://forum.verda.com/t/iommu-bindings-inside-an-instance-or-a-container/58/1 "2025-09-25T12:21:14Z")
Hey is it possible for iommu bindings to be accessible inside an instance or a container? Or is it possible to upload a custom os to be used by an instance?
---
### Author:  [@TeamVerda](https://forum.verda.com/u/TeamVerda)
#### Post date: [September 25, 2025, 12:27pm UTC](https://forum.verda.com/t/iommu-bindings-inside-an-instance-or-a-container/58/2 "2025-09-25T12:27:20Z")
Hi,
Could you please elaborate further on your use case with instance if you need direct GPU access? These questions usually lead to more customized solutions where we set up dedicated bare metal servers.
You can find more information here: [GPU Clusters — DataCrunch](https://datacrunch.io/clusters)
---
### Author:  [@pixelrune](https://forum.verda.com/u/pixelrune)
#### Post date: [September 25, 2025, 1:30pm UTC](https://forum.verda.com/t/iommu-bindings-inside-an-instance-or-a-container/58/3 "2025-09-25T13:30:05Z")
Im trying to run some gpu heavy e2e ci test which would only work on windows. And ideally the windows environment would also be virtualized which then (for the passthrough of the pci device) would require access to the iommu mappings
---
### Author:  [@TeamVerda](https://forum.verda.com/u/TeamVerda)
#### Post date: [September 25, 2025, 2:12pm UTC](https://forum.verda.com/t/iommu-bindings-inside-an-instance-or-a-container/58/4 "2025-09-25T14:12:20Z")
We currently don’t support Windows on our instances. However, as mentioned above we could set up a customized solution for you on our bare-metal servers. This is, of course, dependent on the size of your workloads, as it requires some manual work.