01The problem
Partners waited on someone else's queue.
The group runs structure prediction and docking in large batches, often thousands of candidates per partner request. They shared a university cluster with other departments, and a busy week meant a job sat in the queue for days.
- Predictable capacity for a steady, year-round workload
- A fixed monthly cost that finance could plan around
- Data kept in the EU, as partner contracts require
02The solution
Four reserved servers and one queue.
The lab reserved four H100 Cloud Servers in EU-FRA-1 on a 12-month term at the reserved rate of €2.00 per GPU-hour. A small scheduler spreads jobs across all 32 GPUs and keeps them busy around the clock.
NODEASUS ESC N8A-E12
8/8 online
GPU 0
H100
GPU 1
H100
GPU 2
H100
GPU 3
H100
GPU 4
H100
GPU 5
H100
GPU 6
H100
GPU 7
H100
NVLink
all-to-all
When a partner deadline needs more, the team adds on-demand servers for a few days and stops them when the batch is done.
Placeholder
Placeholder quote. Two plain sentences from the customer's engineering lead about what changed, approved before publishing.