Impartial tool to calculate TCO and compare HPC option costs, including cloud vs on-premises
This TCO calculator is a useful tool for calculating the costs of an HPC system. It does not capture the benefits, and thus the total value of ownership. Cloud HPC can bring benefits that can significantly increase the value of an HPC solution, even when the TCO for Cloud may be higher than on-prem. NAG’s Cloud HPC Migration Service can assist your organization in realizing the potential of Cloud HPC’s benefits while minimizing costs, thus maximizing your total HPC value.
The initial data in the calculator are not necessarily intended to be representative and were mostly sourced from the audience at a tutorial on TCO models for HPC. You can edit the values in the white boxes, and the values in the yellow boxes should be updated automatically. Any data you enter, and any results, stay on your computer and are not sent to NAG or anyone else. This calculator is provided as is. No responsibility is accepted by NAG for your use of this model.
BASELINE TCO MODEL FOR ON-PREMISES HPC SERVICE
PARAMETERS | |
---|---|
Cost of electricity $/kWhr |
0.10 |
PUE (ratio of total power used by the datacenter vs power delivered to the HPC systems) |
1.60 |
Capital (i.e., total amount spent on the hardware) |
$10,000,000 |
Number of racks [calculated] | |
Power consumption kW/rack |
50 |
Utilization (i.e., percentage of the available cycles that are used) |
85% |
Service time (i.e., percentage of time that the system is available to users) |
95% |
Total number of core-hrs user per year [calculated] | |
Cost of a person per year |
$250,000 |
Price per node |
$7,000 |
Nodes per rack |
80 |
Price per rack [calculated] | |
Cores per node |
48 |
Cost of hosting per rack per year |
$30,000 |
Effort spent on procurement & commissioning (person years) |
2.5 |
Effort spent on decommissioning (person years) |
1.5 |
System management team (number of persons or 'FTE') |
3 |
User support team (number of persons or 'FTE') |
3 |
CALCULATED COST PROFILE FOR ON-PREMISES HPC | |||||||
---|---|---|---|---|---|---|---|
Year | 0 (Commission) | 1 (User Service) | 2 (User Service) | 3 (User Service) | 4 (Decommission) | TOTAL | Fraction |
Occupancy (fraction of year the system is in place) |
0.25 |
1 |
1 |
1 |
0.25 |
||
Hardware | |||||||
Electricity | |||||||
Cooling | |||||||
Support | |||||||
Datacenter | |||||||
Procure & Commission | |||||||
Decommissioning | |||||||
Sysadmin | |||||||
User support | |||||||
TOTAL |
KEY METRICS | |
---|---|
TCO | |
Actual cost per core-hr used |
TCO MODEL FOR CLOUD COMPUTING TO DELIVER SAME AMOUNT OF ACTUAL USED CORE-HRS
PARAMETERS | |
---|---|
Cost of cloud cycles - $ per core-hr |
$0.040 |
Total number of core-hrs user per year [calculated] | |
Cost of a person per year |
$250,000 |
Effort spent on procurement & commissioning (person years) |
1.5 |
Effort spent on decommissioning (person years) |
0.5 |
Management & support team for cloud service / virtual cluster (number of persons or 'FTE') |
3 |
User support team (number of persons or 'FTE') |
3 |
CALCULATED COST PROFILE FOR CLOUD HPC | |||||||
---|---|---|---|---|---|---|---|
Year | 0 (Commission) | 1 (User Service) | 2 (User Service) | 3 (User Service) | 4 (Decommission) | TOTAL | Fraction |
Occupancy (fraction of year the service is in place) |
0.25 |
1 |
1 |
1 |
0.25 |
||
Procure & Commission | |||||||
Decommissioning | |||||||
Sysadmin | |||||||
User support | |||||||
Cost of cloud cycles | |||||||
TOTAL |
KEY METRICS | |
---|---|
TCO to use this same computing capacity from on-premises system | |
TCO to use this same computing capacity from the cloud | |
Actual TCO per core-hr used in the cloud | |
Ratio of cloud TCO to on-premise TCO (i.e., cloud is more expensive than on-premises if factor is >1) |
This is a simple model - although it is good enough to be useful quickly for many situations. We could add further detail to the model if needed. For example:
HPC system, Storage, Network, People (system management), People (user support), People (service management), Electricity, Cooling, Facility costs, Financing effects, Security (physical & cyber), Security (IP etc.), Resilience (systems), Resilience (service), Equipment monitoring, Service measurement & reporting, Insurance & Business Continuity, Procurement project, Supplier management, Application porting, Software licenses, User training & support, Documentation, Data Migration, Commission, Warranty/support/maintenance, Decommission, Service fees, and more.
Do contact us for advise on using the tool, help with building your own model appropriate to your situation, to get an Excel copy of the tool, or for anything else we might be able to help with.
© Copyright 2020 The Numerical Algorithms Group Ltd