What's under the hood?

Scaleway offers Virtual Cloud, BareMetal and dedicated GPU instances.

Virtual Cloud Instances

Virtual cloud instances are offering the best performance/price ratio for most workloads. Different CPU architectures are proposed: The Development and General Purpose ranges are based on AMD EPYC CPUs. The ARM64 range is based on Cavium Thunder X ARM CPUs.

  • The Development instances range provides stable and consistent performance for development needs. Spin up a development or test environment within seconds. Refer to the Development Instance offer details for more information.

  • The General Purpose instances range, the solution for demanding workloads. Powerful AMD EPYC CPUs back those instances and offer up to 48 Cores, 256GB of RAM and 600GB of replicated local NVMe SSD storage. Refer to the General Purpose offer details for more information.

  • The ARM instances range is based on Cavium ThunderX SoCs and provides up to 64 Cores ARM 64bit, 128GB of RAM and 1TB SSD storage. Refer to the ARM offer details for more information.

BareMetal Instances

The BareMetal cloud instances, also known as the C1 or C2 instances, are composed of:

  • C1 instance: 4-cores, ARMv7 CPU with 2GB of RAM, and a 1 Gbit/s network card.
  • The C2-S instance: 4-cores, x86-64 CPU with 8GB of RAM, and a 2.5Gbit/s network card.
  • C2M and C2L instance: 8-cores, x86-64 CPU with up to 32GB of ram, 2*2.5Gbit/s network cards. The C2L also provides a 250GB direct SSD.

All of these instances are designed for the cloud and horizontal scaling. You can add additional SSD volumes (local storage) on our BareMetal offers to get as much as 1TB of SSD storage per instance. For more information, refer to Baremetal Instances.

Dedicated GPU Instances

GPU instances are very powerful compute instances, providing lots of RAM, vCPU, and storage.

They are equipped with Nvidia Tesla P100 GPUs, which are designed for handling rapidly, a massive amount of data. They are useful for heavy data processing, artificial intelligence and machine learning, video encoding, rendering, and so on. The GPU is dedicated to each instance and directly exposed through PCI-e. For more information, refer to GPU Instances.

What is the difference between Virtual and BareMetal Cloud Instances?

Our BareMetal cloud instances are real dedicated servers. Each physical server is allocated to a single customer and one only.

With our Virtual Cloud Instances, several customers share one physical server. Each customer get a share of the server’s resources and is completely isolated from others using full virtualization.

There is some tradeoff to do when choosing your instance type. The C1 server does not offer IPv6 while the Virtual Cloud Instances do not offer flexible volume size.

 IPv6Flexible volume sizeConstant Performances
Compute InstancesYesNo, fixed.No

Is it possible to migrate my server from a Virtual Compute Instance to a Bare Metal Cloud Instance?

Yes! Scaleway is the first provider worldwide to offer the option to migrate from a Virtual Compute Instance to a Bare Metal Cloud Instance in 3 clicks.

Are there any restrictions on limited edition instances?

There are no technical restrictions on limited edition instances, but as the name indicates, these instances are only available in limited quantities. We cannot, therefore, guarantee that you will be able to restart one of these instances if it is powered off, especially in the event of a stock shortage.

However, no impact is expected on switched-on instances, and they offer the same reliability as the other offers.

How does a C1 instance perform?

A C1 instance gives a constant CPUMark of 12K. This is equivalent to an AWS M3 medium instance. The C1 instance provides a continual unit of computing perfect for horizontal scalability.

Which floating point instructions are available on ARM instances?

The ARM architecture provides high-performance and high-efficiency hardware support for floating-point operations in half-, single-, and double-precision arithmetic.

On C1 Instances, we use a Marvell XP 370 CPU, based on ARMv7 architecture, that supports VFPv3.

For our ARM64 Cloud Instances, we use Cavium ThunderX SoCs, based on ARMv8 architecture, that supports Advanced SIMD (NEON) by default.

Do you offer SSD volumes?

Yes. We provide 100% SSD volumes. We call them local storage.

You can add extra local storage volumes on BareMetal instances if you need more space (up to 1TB per compute instance)!

Which hypervisor do you use?

With the C1 and C2, there is no hypervisor. You have a BareMetal SSD server with root access. It is yours. There is no virtualization: forget noisy neighbors! We are doing physicalization.

Our Virtual Cloud Instances are using KVM.

Why did you choose ARM for the C1?

We developed with Marvell an ARM SoC offering excellent performances for web applications. The Linux and Ubuntu community has been working on ARM architecture for a long time as it is already used in the majority of our mobile devices. Our C1 server is smaller than a credit card and consumes only a few watts.

We believe that systems on a chip (SoC) are the future of cloud computing.

Why did you go back to x86 for the C2?

We found that the Intel Avoton CPU offered the best performances in their price range allowing us to create an amazingly efficient server for the cloud, the C2. They are an excellent alternative to ARM CPUs for applications which need more memory.

The Intel Avoton are systems on a chip (SoC) with an excellent efficiency letting us provide more powerful instances at a reasonable price.

Which OS support your hardware?

Any Linux distribution with support of the armhf architecture should work out of the box on the C1. The C2 instance can run all Linux distributions compatible with i386 or amd64 architectures. The ARM64 instances can run all Linux distributions compatible with aarch64

Which Linux distributions are available?

We provide a wide range of different Linux distributions and InstantApps for compute instances.

If you’re a daily user or an expert of any that is not available in our current selection, we need your feedback. Please tell us about best practices and gotchas, so we’ll be able to produce a nice default image for the rest of the community.

In any way, you are free to bootstrap your own distribution. See the documentation on creating an image from scratch.

What is the ImageHub?

The Scaleway’s ImageHub is a catalog of system images which allow you to deploy, manage and scale your favorite applications in seconds, from image templates.

What's the price of an image from the ImageHub?

Images in themselves are totally free of charge.

How long does it take to boot an instance?

A new instance starts in less than 44 seconds on SSD!

If you have written a lot of data on your local volumes, your instance may take more time to start and stop.

How can I check that I have a GPU in my instance?

You can check the presence of your GPUs by running: lspci | grep nvidia

If the Nvidia drivers are installed, you can get more information on your GPUs by running: nvidia-smi


Are there special requirements before ordering a GPU instance?

For security reasons, access to the GPU instances is subject to account validation. To launch a GPU instance, you must contact the support by ticket, so they can increase your instance quotas.

Important: Quota allocation depends on your account details. We keep the right to refuse access to GPU instances in case of concerns with the information you provided.

How may I access my resources?

You can access your resources through the control panel or with our API.

Do you offer backups?

Yes. You can backup your data with snapshots or upload your backups to our object storage with any S3 compatible tool, for instance s3cmd.

What is the difference between an image and a snapshot?

An image is a complete backup of your server including all volumes.

A snapshot can be done from a specific volume (for example you have a server with a volume containing the OS and another one containing the application data, and you want to use different snapshot strategies on both volumes).

Creating snapshots of your volumes gives you total freedom which volumes you want to back up while images are convenient for full backups of your instance.


How do I safely power cycle my instance?

There are two steps to stop an instance

  • Power off from your instance
  • Power off from the Control Panel

1 - Power off from your instance

Log in to your instance as root, execute the halt command.
This action will flush any pending changes to the disk, and then terminate power to your instance.

2 - Power off from the Control Panel

To power off an instance from the control panel, click the OFF button next to it:

Power off

IMPORTANT: Using the “OFF” button in the Control Panel to power off you instance is similar to pulling the plug on a running computer.
Powering off an instance from the console without powering it off from your OS first can lead to filesystem corruption.

Which options are available to power off my instance?

When you request to power off your instance, four options are available:

  • Hard Reboot
  • Power off
  • Terminate
  • Standby

When you select the Hard Reboot option, your instance performs an electrical reboot. Your data remains on the local storage. Whenever possible, you should reboot your instance from the OS to avoid data corruption.

When you select the Power off option, all data on the local volume of your instance is transferred to a volume store, and your physical node is released back to the pool of available machines. Depending on the amount of data to be archived, the process can take time.

When you select the Terminate option, your instance and volumes are deleted. You will not be able to recover them. The flexible IP attached to your instance will not be deleted.

When you select the Standby option, your instance is stopped, but all your data remains on the local storage. Your instance is still allocated to your account, and you can restart it anytime, without delay.

Note: With the Standby option, you are still charged the same amount as if the instance is running.

Why powering off my instance takes time?

When you request to power off your instance, the local volumes are transferred to a volume store to be available on the next start of your instance. The stop duration is directly related to the amount of data you wrote on your local volumes.

When you restart the instance, your volumes will be downloaded from the volume store to the instance’s local storage pool.

If you want to make an electrical restart of your instance, run a “Hard Reboot” from the Control Panel. Your local volumes will not be uploaded.

IMPORTANT: Only stop your instance if you want to snapshot it or if you do not need it for an extended period.

Do I have to power off my instance to take a snapshot?

Warning: Hot Snapshots can potentially lead to inconsistent data or data loss. Before taking a Hot Snapshot, you should always sync the data on your disk and stop IO-intensive processes. Whenever possible, you should take a cold (instance powered off) snapshot.

Virtual Cloud Instances support hot snapshots. This means you can perform a snapshot while your instance is running. Hot snapshots allow you to create a point-in-time backup without downtime.

IMPORTANT: At the moment we do not support Hot Snapshots on BareMetal Cloud Instances.

Can I have access to the serial of my instance?

Yes! Just click on the console button from the instances view. In the console, press enter to get the prompt.

How can I reinstall my instance?

To reinstall your instance, simply delete your existing instance. When you delete your instance, your flexible IP address is kept in your account, and you can spawn a new instance with the same IP address.

Why is my instance not starting?

As a cloud provider, our promise is to deliver resources on demand. Behind this promise, there is a reality, and our infrastructure is backed by physical hardware.

We can face stock issues on some instance type during short time-frame periods that can impact the provisioning. In such a situation, we encourage you to provision on a different instance type or check the availability of the resource in a different region.

The number of volumes attached to your instance in case of a low stock situation can also have an impact on the provisioning.

Note that we do our best to void this kind of scenario.