Research Lab & University HPC Planner
Size shared research compute from people and usage: monthly GPU-hours or core-hours demand, peak concurrency and target utilisation give the accelerators or cores needed, then servers, power, energy cost and — if you enter a server price — capital cost.
Results
- GPUs by monthly demand
- 13calculated
- GPUs by peak concurrency
- 20calculated
- GPUs required
- 20calculated
- Servers
- 3calculated
- IT power (max)
- 18 kWcalculated
- Annual energy cost
- ₹12.3 lakhestimated
What it calculates
- Resource units needed (GPUs or cores)
- Servers
- Power
- Annual energy cost
- Capital cost
Worked example
Using the default inputs below; change any input above to recalculate.
| Resource | GPUs |
|---|---|
| Researchers | 40 |
| Hours per researcher per month (GPU-hours or core-hours) | 150 |
| Peak simultaneous users | 25 % |
| Units per active user at peak | 2 |
| Target utilisation | 65 % |
| Units per server | 8 |
| Power per server | 6 kW |
| Server price (optional) | 0 ₹ |
| PUE | 1.5 |
| Tariff | 8 ₹/kWh |
| GPUs by monthly demand | 13 | calculated |
|---|---|---|
| GPUs by peak concurrency | 20 | calculated |
| GPUs required | 20 | calculated |
| Servers | 3 | calculated |
| IT power (max) | 18 kW | calculated |
| Annual energy cost | ₹12.3 lakh | estimated |
Methodology
Demand-based sizing: units = monthly unit-hours ÷ (730 hours × target utilisation). Peak-based sizing: simultaneous users × units each. The larger wins; servers, power and energy follow.
Formulas
- Demand
- U = users × hours ÷ (730 × utilisation)
Assumptions
- 730 hours per month.
Limitations
- Deadline-driven bursts (paper or thesis submissions) may need extra headroom or cloud bursting.