Virtual machine flavors and billing unit rates

This article lists the types (flavors) of virtual machines and their cost in billing units.

The cPouta and ePouta services consume the same billing units as Puhti and Mahti. You can find more information in the CSC computing environment articles.

Users can create virtual machines with larger or smaller compute resources based on their needs. The virtual machine flavors available in cPouta and ePouta are listed below in separate tables. Please note that the values for the memory of each flavor (in GiB) are approximated.

About redundancy

In addition to the amount of cores, memory, and disk, the flavors we offer vary also on the level of redundancy of the nodes hosting the virtual machines. The details on the levels of redundancy follow.

Power redundancy

For the power provisioning of the node hosting the virtual machine, there are two possible values of redundancy.

  • NONE - The node is not protected from sudden power losses. A fault in the power provisioning of the node might make the virtual machine temporarily unreachable.
  • FULL - The node is protected from sudden power losses (UPS).

Data redundancy

Within each virtual machine, the customer data is stored in a root disk (R) and possibly in an ephemeral disk (E). For customer data, there are three possible values of redundancy.

  • NONE - The disk is stored only in the node running the virtual machine and it is not backed up (RAID-0 or LVM striping). A fault in one of the disks of the node might corrupt the data of the virtual machine. Moreover, a fault in the node hosting the virtual machine might make the virtual machine not usable until the fault is fixed.
  • BASIC - The disk is stored only in the node running the virtual machine and it is mirrored within the same node (RAID-1). A fault in a single disk of the node does not compromise the data of the virtual machine. Simultaneous faults in multiple disks of the node might corrupt the data of the virtual machine. Moreover, a fault in the node hosting the virtual machine might make the virtual machine not usable until the fault is fixed.
  • FULL - The disk is stored using multiple nodes in a fault-tolerant fashion (Ceph), so the customer data is not tied to any specific node. In case of a fault in a node used by the customer, it is possible to re-spawn the virtual machine of the customer using an alternative node.

Network redundancy

For the network reachability of the virtual machine, there are two possible values of redundancy.

  • NONE - The node hosting the virtual machine is connected to the cloud platform without a failover link. A fault in the link of the node might make the virtual machine temporarily unreachable.
  • FULL - The node hosting the virtual machine is connected to the cloud platform with an additional failover link.

cPouta flavors

The following tables list the available virtual machine flavors in cPouta and their billing unit coefficients. Note that the default cPouta user account allows users to launch only a subset of the available virtual machine flavors.

Standard flavors

Flavor Cores Memory
(GiB)
Root
disk
(GB)
Ephemeral
disk
(GB)
Total
disk
(GB)
Memory/
core
(GiB)
Redundancy Billing
Units
/h
standard.tiny 1 1 80 0 80 1 0.25
standard.small 2 2 80 0 80 1 0.5
standard.medium 3 4 80 0 80 1.3 1
standard.large 4 7 80 0 80 1.8 2
standard.xlarge 6 15 80 0 80 2.5 4
standard.xxlarge 8 30 80 0 80 3.8 8
standard.3xlarge 8 60 80 0 80 7.5 16

HPC flavors

Flavor Cores Memory
(GiB)
Root
disk
(GB)
Ephemeral
disk
(GB)
Total
disk
(GB)
Memory/
core
(GiB)
Redundancy Billing
Units
/h
hpc.4.5core 5 20 80 0 80 4 6
hpc.4.10core 10 40 80 0 80 4 12
hpc.4.20core 20 82 80 0 80 4.1 25
hpc.4.40core 40 165 80 0 80 4.1 50
hpc.4.80core 80 328 80 0 80 4.1 100
hpc-gen2.24core 24 112 80 0 80 4.7 30
hpc-gen2.48core 48 225 80 0 80 4.7 60

Note that the root disks of the hpc-gen2.24core and the hpc-gen2.48core flavors are hosted on hard disk drives (HDDs).

I/O flavors

Flavor Cores Memory
(GiB)
Root
disk
(GB)
Ephemeral
disk
(GB)
Total
disk
(GB)
Memory/
core
(GiB)
Redundancy Billing
Units
/h
io.70GB 2 10 20 70 90 5 3
io.160GB 4 20 20 160 180 5 6
io.340GB 8 37 20 340 360 4.6 12
io.700GB 16 75 20 700 720 4.7 24

Note that both the root and the ephemeral disks of all I/O flavors are hosted on solid-state drives (SSDs).

GPU flavors

Flavor Cores GPUs Memory
(GiB)
Root
disk
(GB)
Ephemeral
disk
(GB)
Total
disk
(GB)
Memory/
core
(GiB)
Redundancy Billing
Units
/h
gpu.1.1gpu 14 1 112 80 0 80 8 60
gpu.1.2gpu 28 2 224 80 0 80 8 120
gpu.1.4gpu 56 4 447 80 0 80 8 240

Note that the root disks of all GPU flavors are hosted on solid-state drives (SSDs).

ePouta flavors

The following tables list the available virtual machine flavors in ePouta and their billing unit coefficients.

Standard flavors

Flavor Cores Memory
(GiB)
Root
disk
(GB)
Ephemeral
disk
(GB)
Total
disk
(GB)
Memory/
core
(GiB)
Redundancy Billing
Units
/h
standard.tiny 1 1 80 0 80 1 0.25
standard.small 2 2 80 0 80 1 0.5
standard.medium 3 4 80 0 80 1.3 1
standard.large 4 7 80 0 80 1.8 2
standard.xlarge 6 15 80 0 80 2.5 4
standard.xxlarge 8 30 80 0 80 3.8 8
standard.3xlarge 8 60 80 0 80 7.5 16

HPC flavors

Flavor Cores Memory
(GiB)
Root
disk
(GB)
Ephemeral
disk
(GB)
Total
disk
(GB)
Memory/
core
(GiB)
Redundancy Billing
Units
/h
hpc.fullnode.haswell 46 230 80 0 80 5 72
hpc.3.28core 28 115 80 0 80 4.1 48
hpc.3.56core 56 240 80 0 80 4.3 96
hpc.4.5core 5 20 80 0 80 4 8
hpc.4.10core 10 42 80 0 80 4.2 15
hpc.4.20core 20 85 80 0 80 4.3 30
hpc.4.40core 40 168 80 0 80 4.2 60
hpc.4.80core 80 335 80 0 80 4.2 120

I/O flavors

Flavor Cores Memory
(GiB)
Root
disk
(GB)
Ephemeral
disk
(GB)
Total
disk
(GB)
Memory/
core
(GiB)
Redundancy Billing
Units
/h
io.haswell.2core 2 10 20 70 90 5 4.5
io.haswell.4core 4 20 20 160 180 5 9
io.haswell.8core 8 37 20 350 370 4.6 18
io.haswell.16core 16 75 20 700 720 4.7 36
io.haswell.32core 32 150 20 1400 1420 4.7 72
io.haswell.46core 46 230 20 2100 2120 5 108

High memory flavors

Flavor Cores Memory
(GiB)
Root
disk
(GB)
Ephemeral
disk
(GB)
Total
disk
(GB)
Memory/
core
(GiB)
Redundancy Billing
Units
/h
tb.3.480RAM 56 458 20 1650 1730 8.2 110
tb.3.1470RAM 80 1402 80 2500 2580 17.5 320

Note that the root disks of all high memory flavors are hosted on solid-state drives (SSDs), while the ephemeral disks are hosted using NVM Express (NVMe).

GPU flavors

Flavor Cores GPUs Memory
(GiB)
Root
disk
(GB)
Ephemeral
disk
(GB)
Total
disk
(GB)
Memory/
core
(GiB)
Redundancy Billing
Units
/h
gpu.1.1gpu 14 1 112 80 0 80 8 60
gpu.1.2gpu 28 2 225 80 0 80 8 120
gpu.1.4gpu 56 4 447 80 0   80 8 240
gpu.2.1gpu 20 1 172 80 1000 1080 8.6 100

Note that both the root and the ephemeral disks of the GPU flavors are hosted on solid-state drives (SSDs).

Which type of flavor should I use?

Standard flavors

Typical use cases:

  • Web services (non-HPC)
  • Software development

These are generic flavors that are useful for running regular web services such as a web server with a database backend. They provide better availability compared to the HPC flavors.

Cloud administrators can move these virtual machines from one host machine to another without causing a break in service. This means that you are likely less affected by maintenance.

These flavors are not suitable for computationally intensive workloads. The virtual CPUs used in these instances are overcommitted, which means 32 hyperthreaded CPU cores are used to provide more than 32 virtual cores.

Flavor characteristics:

  • Redundant power
  • CPU: Varies
  • Network: Redundant 25 Gb/s
  • Flavor disk: Stored in the central storage
  • Single-node or disk failures may cause downtime, but instances are recoverable.

HPC flavors

Typical use cases:

  • Scientific applications

If your use case is computationally intensive, you should use one of the HPC flavors. The availability of these instances is not as high as the standard flavors, but you get better performance. The HPC flavors have faster CPUs and no overcommitment of CPU cores.

cPouta HPC flavor characteristics:

hpc.4.*:

  • No redundant power
  • CPU: Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz, hyper-threading
  • Network: Redundant 25 Gb/s
  • Flavor disk: Stored in the central storage
  • Single-node or disk failures may cause downtime, but instances are recoverable.

hpc-gen2.*:

  • No redundant power
  • CPU: Intel(R) Xeon(R) CPU E5-2680 v3, with hyper-threading
  • Network: Single 40 Gb/s
  • Flavor disk: Local SATA disk, RAID-0
  • Instances can be lost due to a single-node or disk failure.

ePouta HPC flavor characteristics:

hpc.4*:

  • Redundant power
  • CPU: Intel(R) Xeon(R) CPU Gold 6148, with hyper-threading
  • Network: Redundant 25 Gb/s
  • Flavor disk: Stored in the central storage
  • Single-node or disk failures may cause downtime, but instances are recoverable.

hpc.3*:

  • Redundant power
  • CPU: Intel(R) Xeon(R) CPU E5-2680 v4, with hyper-threading
  • Network: Redundant 25 Gb/s
  • Flavor disk: Stored in the central storage
  • Single-node or disk failures may cause downtime, but instances are recoverable.

hpc.*.haswell:

  • Redundant power
  • CPU: Intel(R) Xeon(R) CPU E5-2690 v3, with hyper-threading
  • Network: Redundant 10 Gb/s
  • Flavor disk: Stored in the central storage
  • Single-node or disk failures may cause downtime, but instances are recoverable.

I/O flavors

Typical use cases:

  • Hadoop/Spark
  • Non-critical centralized databases
  • Clustered databases

I/O flavors are intended to provide the best I/O performance on the virtual machine root and ephemeral disks. They are backed by local SSDs on the servers they run on. The SSDs are configured in a RAID-0 configuration for maximal performance. This means there is an increased risk of loss of a virtual machine in case of hardware problems. The risk of disk failure is larger than in the other flavors, so it is especially important to be aware of the risks of data loss with these flavors.

As these instances are also tightly tied to the hardware, you may expect downtime of instances during the maintenance of the hardware. The resize/migration functionalities do not work for these instances. The bulk of the storage is available as an ephemeral disk, typically in /dev/vdb.

Often you want to create clusters of servers with the io.* flavors. In these cases, you probably want to have your virtual machines land on different physical servers. This cannot currently be done in the web interface. To do this, please refer to the anti-affinity group commands in our command line instructions.

The availability of these instances is not as high as the standard flavors, but the I/O performance is significantly better.

cPouta IO flavor characteristics:

io.*:

  • Redundant power
  • CPU: Intel(R) Xeon(R) CPU E5-2680 v3, with hyper-threading
  • Network: Redundant 10 Gb/s or 40 Gb/s
  • Flavor disk: Local SSD disks, RAID-0
  • Instances can be lost due to a single-node or disk failure.

ePouta IO flavor characteristics:

io.haswell.*:

  • Redundant power
  • CPU: Intel(R) Xeon(R) CPU E5-2680 v3, with hyper-threading
  • Network: Redundant 10 Gb/s
  • Flavor disk: Local SSD disks, RAID-0
  • Instances can be lost due to a single-node or disk failure.

GPU flavors

Typical use cases:

  • High performance compute applications leveraging GPUs
  • Machine and deep learning, e.g. TensorFlow
  • Rendering

The GPU flavors are intended to provide high performance computing using GPGPU (General Purpose computing on Graphical Processing Units). GPGPUs can significantly speed up certain algorithms and applications. The gpu.1. flavors have NVIDIA Tesla P100 GPGPUs. The gpu.2.1gpu in ePouta have a NVIDIA Tesla V100 GPGPU.

The GPGPUs are suitable for deep learning, scientific computing as well as for remote desktops, rendering or visualization. The GPGPU flavors are backed by local SSD on the servers. The SSDs in gpu.1 flavors are configured in RAID-1. This is where the OS root disk is stored. With gpu.2 flavors, the SSDs are bigger and the SSDs are configured in RAID-0 for faster staging of datasets. You can use the volumes for storing larger data sets and persistent data. If you need to read and write a lot of data between the disk and GPGPU, using volumes might affect performance when compared to local SSD disk.

To take advantage of the acceleration which GPGPUs provide, the applications you run must support them. If you write your own applications, the optimization service helps in leveraging the GPGPUs.

GPGPUs can be used for a lot of cool and interesting things, but please remember the resource usage must comply with the Terms of use.

Limitations and caveats: 

  • As we use PCI passthrough to get the whole GPGPU into the instance. The administrators are not able to access the GPGPU and check its health. Please report any errors or problems with the GPGPUs to CSC (and attach the output of the command "nvidia-smi -q").
  • The applications must be able to utilize the GPU to get a speedup.

These instances are also tightly tied to the hardware. You may expect downtime of instances during the maintenance of the hardware.

Users also have the possibility to use NVIDIA Volta V100 GPGPUs in the batch system on Puhti-AI: https://research.csc.fi/techspecs.

cPouta flavor characteristics:

gpu.1.*:

  • No redundant power
  • CPU: Intel(R) Xeon(R) CPU E5-2680 v4, with hyper-threading
  • Network: Redundant 10 Gb/s
  • Flavor disk: Local SSD disks, RAID-1
  • Instance can be lost due to a single-node or multiple simultaneous disk failures.

ePouta flavor characteristics:

gpu.1.*:

  • Redundant power
  • CPU: Intel(R) Xeon(R) CPU E5-2680 v4, with hyper-threading
  • Network: Redundant 10 Gb/s
  • Flavor disk: Local SSD disks, RAID-1
  • Instance can be lost due to a single-node or disk failure.

gpu.2.*:

  • Only available via request to servicedesk@csc.fi
  • Redundant power
  • CPU: Intel(R) Xeon(R) Gold 6148, with hyper-threading
  • NUMA Aware: yes (CPU <> memory, not PCI devices)
  • Network: Redundant 10 Gb/s
  • Flavor disk: Local SSD disks, RAID-0
  • Instance can be lost due to a single-node or disk failure.

Installation and configuration of GPU Flavors

We have specific CUDA images available for the GPU nodes. These images come pre-installed with the freshest CUDA version. Note that the CUDA images are not configured with auto update. You can use any other images with the GPU flavors, but you have to install the required libraries yourself. How CSC customizes the images.

High memory flavors (only in ePouta)

Typical use cases:

  • Scientific applications requiring large amounts of memory

These flavors have large amounts of memory and are meant for use cases which require and can utilize such amounts of memory. Typical use cases of these flavors include genome sequencing and analysis applications.

The resize/migration functionalities do not work for these instances.

If you need to move a workload from another type of VM to an instance with a high memory flavor, i.e., a TB instance, either move all data and install all applications manually on the new TB instance or create a snapshot of the source VM. Then convert that snapshot to a volume and use the volume to create the new TB-flavor VM.

If you need to move a workload from a TB instance to another instance, either move all data and install all applications manually on a new VM or create a snapshot of the source VM. Please note that all ephemeral disk data will be lost in the process and will not be stored in the snapshot because only the TB VM root disk is stored in the snapshot.

Flavor characteristics:

tb.3.*:

  • Redundant power
  • CPU: Intel(R) Xeon(R) CPU E5-2680 v4, with hyper-threading or Intel(R) Xeon(R) CPU E5-2698 v4, with hyper-threading
  • Network: Redundant 25 Gb/s
  • Flavor disk: Local SSD disks, RAID-0
  • Instances can be lost due to a single-node or disk failure.

Deprecated flavors

This is the set of original flavors that has been available since the launch. You should not launch any new virtual machines using any of these flavors. Existing virtual machines that use these flavors will continue to work. We will maintain these flavors for a period of time, but they will be removed at some point in the near future.

Flavor Cores Memory
(GiB)
Root
disk
(GB)
Ephemeral
disk
(GB)
Total
disk
(GB)
Memory/
core
(GiB)
Billing
Units
/h
hpc-gen1.1core 1 3.7 80 (RAID0) 0 80 3.7 2
hpc-gen1.4core 4 15 80 (RAID0) 0 80 3.7 8
hpc-gen1.8core 8 30 80 (RAID0) 0 80 3.7 16
hpc-gen1.16core 16 60 80 (RAID0) 0 80 3.7 32
hpc-gen2.2core 2 10 80 (RAID0) 0 80 5 4
hpc-gen2.8core 8 40 80 (RAID0) 0 80 5 15
hpc-gen2.16core 16 80 80 (RAID0) 0 80 5 30
tiny 1 1 10 (RAID0) 110 (RAID0) 120 1 2
mini 1 3.5 10 (RAID0) 110 (RAID0) 120 1.7 2
small 4 15 10 (RAID0) 220 (RAID0) 230 3.8 8
medium 8 30 10 (RAID0) 440 (RAID0) 450 3.8 16
large 12 45 10 (RAID0) 660 (RAID0) 670 3.8 24
fullnode 16 60 10 (RAID0) 900 (RAID0) 910 3.8 32
hpc.mini 2 3.5 80 0 80 1.8 5
hpc.small 4 7 80 0 80 1.8 10
hpc.medium.haswell 8 40 80 0 80 5 20
hpc.large.haswell 16 80 80 0 80 5 40
hpc.xlarge.haswell 32 156 80 0 80 5 80
hpc.medium.westmere 8 14 80 0 80 1.8 8
hpc.large.westmere 16 28 80 0 80 1.8 16
hpc.xlarge.westmere 23 41 80 0 80 1.8 24
hpc.largemem.westmere 23 90 80 0 80 4 36
tb.4.735RAM 80 735 80 (SSD/RAID0) 3300 (SSD/RAID0) 3380 9.2 220 (350)
tb.westmere.32core 32 488 80 (RAID6) 3250 (RAID6) 3330 15.2 200
tb.westmere.64core 64 976 80 (RAID6) 6500 (RAID6) 6580 15.2 400

[https://research.csc.fi/pouta-adding-images]:

Last edited Wed Jul 15 2020