Hackers circumvent Nvidia’s limitation on virtual graphics hardware

Virtual machines are a common tool in the server world, which can be used, for example, to temporarily create an isolated environment with tailored computing power. In this way, it is possible to let different parts of processors and graphics cards handle a wide variety of tasks, something that is used, for example, in Nvidia’s Geforce Now solution, where powerful resources are shared between several users.

For those who virtualize graphics hardware with products from Nvidia, however, the more expensive Tesla and Quadro series are needed. Until a few weeks ago, these series were also required to use an Nvidia graphics card in a virtual machine, so-called passthrough. Nowadays, this barrier has been officially removed, but Tom’s Hardware now reports that hackers have found a much less official way to circumvent the former barrier and thus divide Geforce graphics cards into several virtual parts.

In Nvidia language, the function is called virtual GPU (vGPU) and it is a pure software function, which for cost reasons is only verified and supported for graphics cards intended for data centers and professional use. It is also mainly in situations with very accessible computing power and many users that vGPU is really required, but for niche enthusiast projects, perhaps the new solution may come in handy.

The code snippet that enables virtualization of Geforce-stamped cards is available via Github and works by tricking the system that the graphics card used is a model with official support. The requirement is that Linux is used and that virtual machines use KVM software, but also that the graphics card in question has one of the graphics circuits GP102, GP104, TU102, TU104 or GA102 under the shell. This includes the top cards from “Pascal”, “Turing” and today’s “Ampere”.

Tom’s Hardware emphasizes that the lack of Windows support and untested stability is unlikely to have a major impact on the full solution, even if, for example, miningcompanies have consumer-oriented graphics cards available in larger volumes. Whether Nvidia will take action to remedy what has happened remains to be seen, while a less likely scenario is that the recent passthroughThe support is accompanied by vGPU with the manufacturer’s blessing.

Read about more Nvidia restrictions that have disappeared: