Research compute that scales with the question
Protein models, molecular screening, weather and climate simulation — run campaigns on GPUs when you need them and keep steady pipelines on dedicated bare metal.
Recommended start
A100 80GB SXM $1.69/hr
768 GB
RAM on the largest bare-metal tier
4 regions
for data placement
80 GB
HBM on A100 and H100
What makes this workload expensive
The bill is rarely about the GPU price alone. These are the problems we design around.
Campaign-shaped demand
A screening campaign needs hundreds of GPUs for weeks, then almost none.
Mixed CPU and GPU pipelines
Pre- and post-processing often run on CPUs; model steps on GPUs.
Data that must stay put
Research data can come with residency and access requirements.
Built for the shape of the job
A100 to B200
FP64-capable A100 and H100 for simulation, plus high-memory H200 and B200 for large models.
Dedicated bare metal
Single-tenant EPYC servers with up to 768 GB RAM for CPU-heavy stages.
Regional placement
Keep datasets in US, EU or Asia-Pacific regions close to where they are governed.
Private networking
VPCs and encrypted volumes keep research data off the public internet.
Where to start, where to scale
A100 80GB SXM
Start here80 GB VRAM · 1× NVIDIA A100
$1.69/hr
H100 SXM
80 GB VRAM · 1× NVIDIA H100
$2.69/hr
Metal E64
128 vCPU · 768 GB RAM · 4× 3.84 TB NVMe
$1,899/mo
From first run to production
-
1
Split the pipeline into CPU and GPU stages.
-
2
Run steady CPU stages on bare metal; burst GPU stages on demand.
-
3
Keep datasets on volumes in the region where they are governed.
-
4
Scale down between campaigns and pay nothing for idle GPUs.
Guides for this workload
September 25, 2026
Running EU AI workloads in Frankfurt: a data residency checklist
September 8, 2026
Checkpointing at scale: stop losing hours to failed nodes
August 20, 2026
Choosing a GPU cloud region: latency, data residency, cost
A100 and H100 provide strong FP64 throughput for simulation codes; consumer RTX cards do not.
Yes. Bare-metal servers and GPU instances can share a private network in the same region.
Yes — choose the EU Central (Frankfurt) region for instances, volumes and object storage.
Other workloads
Start training in minutes
Create an account, add a card, and launch your first GPU instance. Per-second billing means you only pay for what you use.
No minimum commitment · Cancel anytime · $10 free credit for new accounts