When you create a virtual machine (VM) on Hivenet’s Compute, you’ll pick a Linux OS. The options are familiar: Ubuntu, Debian, and Fedora.
Most people can pick Ubuntu and move on. That’s not a cop-out. It’s the honest default when you want broad compatibility and you don’t want to think about OS details on day one.
If you’re planning to use a GPU-enabled virtual machine on Ubuntu, you can run demanding graphical or computational tasks in an isolated environment, achieving performance that nearly matches bare-metal. NVIDIA Virtual GPU (vGPU) technology enables sharing a single physical GPU card across multiple VMs on Ubuntu, reducing costs and maximizing hardware usage. Note: NVIDIA vGPU software often requires separate, costly licenses for advanced GPU sharing configurations. Technologies like NVIDIA vGPU allow multiple VMs to share a single physical GPU, maximizing hardware utilization and reducing the need for multiple expensive cards. Creating and managing vGPUs involves adding one or more vGPUs to a virtual machine and setting the necessary plugin parameters. To manage GPU resources in virtual machines, users should select the appropriate vGPU profile based on application requirements. You must prepare the GPU for vGPU use by following specific configuration steps. Monitoring GPU performance can be done using the nvidia-smi tool from a hypervisor, allowing you to view GPU information and performance metrics.
GPU Passthrough allows a guest VM to achieve near-bare-metal GPU performance, ideal for 3D rendering, video encoding, and gaming. To achieve native performance, users can utilize PCI passthrough of additional GPUs (cards) in the system. Configuring GPU passthrough requires advanced knowledge, including IOMMU support in BIOS/UEFI and precise PCI device binding. Enabling the input-output memory management unit (IOMMU) is a prerequisite for PCI passthrough of GPUs in virtual machines, and setup complexity includes requiring specific hardware support for IOMMU and often involves manual kernel modifications and BIOS tuning. Setting up a GPU virtual machine on Ubuntu involves enabling IOMMU in the BIOS, isolating the GPU with VFIO drivers, and passing the PCI device to the VM. It’s important to identify the correct PCI domain and card when configuring passthrough. You can check your boot-up kernel messages for IOMMU/DMAR messages, and use system commands to filter and list PCI devices—look for lines where your GPU is found. You may need to create a modprobe blocklist to prevent certain drivers (like Nouveau) from loading, as the default open-source Nouveau driver conflicts with NVIDIA drivers in Ubuntu and must be disabled for proper installation. During system setup or updates, you may need to copy configuration files or commands as part of the process. The level of integration or control required for advanced hardware configurations like PCI passthrough and mediated devices can be significant.
A mediated device is essentially the partitioning of a hardware device using firmware and host driver features. Virtualization also offers better security by isolating workloads, ensuring that a crash or security incident in one VM does not affect the host system or other VMs. To configure a license client of the NVIDIA license system, start by generating a client configuration token.
Running TensorFlow or PyTorch inside an Ubuntu container or VM supports AI/ML training and takes advantage of GPUs in modern computing for parallel workloads. The CUDA Toolkit can be downloaded from NVIDIA to enable parallel computing for AI/ML tasks, especially in conjunction with Docker. These capabilities align with key AI trends SMBs can leverage with cloud GPU computing to adopt AI efficiently and affordably. To set up a GPU-accelerated Ubuntu virtual machine, you must select a compatible host, attach the hardware, and install specific drivers.
If you’re still deciding between a VM and a container instance, start here: [[Interlink: VM or container: how to choose in 60 seconds]]. If you want the quick product update on VMs, read: Compute now supports virtual machines (VMs).
Overview of Operating Systems
Your operating system controls everything in your computing environment. It manages hardware resources and provides the platform where applications run. In cloud virtual machines, your OS choice affects installation steps, configuration, and how well your environment performs.
Ubuntu works well for cloud VM setups. This Linux distribution has a straightforward installation process—download the official image, create your VM, and follow the guided steps. Ubuntu's community support and documentation help you find instructions for most use cases. You can set up development environments or deploy production workloads without much trouble.
GPU support matters if you need accelerated computing. Ubuntu and other Linux distributions support GPUs well, so you can install and configure NVIDIA vGPU software for high-performance tasks. If you need vGPU features, check the official documentation for installation and configuration instructions, and consider how cloud GPUs in modern computing can accelerate AI and scientific workloads. This ensures your VM handles demanding applications.
You need to configure both host and guest operating systems correctly when creating a virtual machine. QEMU lets you run a guest OS—like another Linux distribution or Windows—on a host system. QEMU gives you options to set the kernel version, allocate CPU and memory, and configure network settings. You can tailor your VM to your specific needs, especially if you’ve thought through key questions before choosing a distributed compute provider.
Check your VM's configuration with the sysfs virtual file system. It shows detailed information about hardware and kernel settings. You can view sysfs files to check loaded drivers, confirm GPU availability, and verify your system setup. Other tools help too—use the top command to monitor resource usage or sysctl to tune kernel parameters. These tools help you improve performance and reliability.
Configure your operating system to work with the underlying cloud infrastructure. You might need to set a specific kernel version, adjust network options, or allocate the right resources for your workload. Many cloud providers, including Compute with Hivenet, offer additional resources and documentation to help you get the most from your VM, and you can explore why developers choose Compute with Hivenet for GPU workloads when planning your environment, including insights from Hivenet's first certified GPU supplier on moving from mining to AI workloads.
Understanding how to install, configure, and verify your operating system ensures your virtual machines are secure, efficient, and ready for any use case. You can run AI training with GPUs or stable production services. Use the right tools and follow best practices to create a system that meets your needs in the cloud.
How these three differ in practice
Ubuntu is the “most tutorials match this” choice. A lot of AI/ML, CUDA, and Docker guidance on the internet is written with Ubuntu in mind, so you spend less time translating instructions, including step‑by‑step guides for serving Llama 3.1‑8B on an Ubuntu VM.
Debian is the “stable by default” choice. It’s often used for long-running services and conservative setups where you value predictability over the newest packages.
Fedora is the “newer toolchain” choice. It tends to get recent versions of system software earlier, which can be great for development work, but it can also mean you’re closer to the edge of change.
None of these are “better.” They’re different tradeoffs.
If you don’t have a strong preference, pick Ubuntu
Ubuntu is the least surprising option for most Compute users. It’s usually the easiest way to follow third-party guides, install common tooling, and get an AI workload running without extra friction.
This matters more than people admit. A big chunk of “setup time” is simply matching a tutorial’s assumptions. Ubuntu reduces that mismatch.
If your main reason for using a VM is Docker, Ubuntu is also a comfortable default. This blog post explains why. For the exact setup steps, use: Install Docker on a Compute VM.
Pick Debian if your VM is going to behave like a server
Debian is a good choice when your instance will run for a while and you want the OS to stay boring.
That includes cases like:
A long-lived inference API where you want fewer OS surprises.
A production-ish service where you value stability and clear upgrade decisions.
A team environment where you’d rather update on your schedule than chase the newest versions.
Debian can absolutely run modern AI stacks. The difference is that you may occasionally do a bit more manual work when a third-party guide assumes Ubuntu packages or Ubuntu-specific defaults.
Pick Fedora if you want newer system software for dev work
Fedora is a good choice when you care about getting recent versions of tools and libraries without having to fight your OS.
This often fits:
Developer environments where you’re iterating fast.
Workloads where you want newer compilers, runtimes, or system tooling.
Teams that already use Fedora and want the VM to feel like “home.”
Fedora can be great on Compute. It just tends to be a little less “follow the internet” friendly than Ubuntu, because fewer tutorials assume it first.
The question people actually mean: “Which OS should I use for AI?”
If you’re running ML workloads and you don’t have a strong OS preference, Ubuntu is usually the easiest choice because third-party tooling and instructions line up with it more often.
If you already have a stable server baseline and you want to keep things predictable, Debian is a good fit.
If you’re doing dev-heavy work and you like newer system software, Fedora can feel cleaner.
If you want help choosing the runtime before you even pick the OS, this is the quick decision page: Virtual machine vs container for machine learning. If you’re still unsure whether you even need a GPU VM, start here: GPU virtual machine: what it is and who actually needs one.
A practical way to avoid regret
Don’t treat OS choice as permanent identity. Treat it as a starting point.
If you’re experimenting, choose Ubuntu, get a clean run, and write down what you installed and why. If you later decide you want Debian’s stability or Fedora’s newer toolchain, you can rebuild the environment with clearer requirements instead of guessing.
One more thing that trips people: reaching your app from the outside world is a separate decision from the OS. If you’re running a web UI or API on the VM, plan how you’ll access it. This explainer is the quick version: SSH, HTTPS, TCP, UDP: how to expose a service from a Compute VM. The docs tutorial has the exact steps: Expose a service from a Compute VM: SSH, HTTPS, TCP, and UDP.
Try Compute
If you want the lowest-friction path, launch a VM with Ubuntu, connect over SSH, and get your workflow running. Once it’s real, you can decide whether you want the more conservative Debian feel or the more up-to-date Fedora feel.
