cloud.researchstem.com

Research-grade compute.
From 32 cores to 8 x B300.

Dedicated CPU, RAM, GPU and VRAM for the Research STEM programme — provisioned as single-tenant nodes with NVLink and InfiniBand fabrics, no noisy neighbours, no hidden egress.

Max cores / node
2 x 128
Max system RAM
4 TB DDR5
Max VRAM / node
2.3 TB HBM3e
Interconnect
800G IB

01 / CPU & memory

Dense EPYC nodes, up to 256 cores

Every tier ships dedicated physical cores, ECC DDR5 and local Gen5 NVMe scratch. Scale from a single 32-core node to dual-socket 128-core systems.

Compute 32

32 cores

CPUAMD EPYC 9354P / Intel Xeon 6538N
RAM256 GB DDR5 ECC
NVMe2 x 3.84 TB NVMe Gen5
Network25 GbE

€1.41 / hr

Compute 64

64 cores

CPUAMD EPYC 9554P
RAM768 GB DDR5 ECC
NVMe4 x 3.84 TB NVMe Gen5
Network100 GbE

€2.79 / hr

Most used

Compute 128

128 cores

CPUAMD EPYC 9754
RAM1.5 TB DDR5 ECC
NVMe8 x 3.84 TB NVMe Gen5
Network200 GbE

€5.40 / hr

Compute 2 x 128

256 cores

CPUDual AMD EPYC 9754
RAM3 TB DDR5 ECC
NVMe8 x 7.68 TB NVMe Gen5
Network2 x 200 GbE

€10.65 / hr

02 / Single GPU

One accelerator, full node performance

Single-GPU instances for development, visualisation and focused training runs. Passthrough GPUs, no time-slicing.

GPUVRAMArchitectureHostTypical workloadFrom
GeForce RTX 509032 GB GDDR7Blackwell16 vCPU · 128 GB RAMPrototyping, rendering, small fine-tunes€2.10 / hr
RTX PRO 6000 Blackwell96 GB GDDR7 ECCBlackwell24 vCPU · 256 GB RAMVisualisation, simulation, mid-size models€3.90 / hr
NVIDIA H100 SXM80 GB HBM3Hopper32 vCPU · 384 GB RAMTraining, HPC, CUDA/NCCL workloads€4.43 / hr
NVIDIA H200 SXM141 GB HBM3eHopper32 vCPU · 512 GB RAMLong-context LLM inference & training€5.70 / hr

03 / Multi-GPU

NVLink clusters up to 8 x B300

Multi-GPU nodes wired with NVLink inside the chassis and rail-optimised InfiniBand across the rack, for distributed training and large-scale simulation.

4 x RTX PRO 6000 Blackwell

384 GB VRAM

Host64 cores · 768 GB RAM
FabricPCIe Gen5 · 100 GbE

€14.70 / hr

8 x H100 SXM

640 GB HBM3 VRAM

Host128 cores · 1.5 TB RAM
FabricNVLink · 8 x 400G InfiniBand

€32.40 / hr

8 x H200 SXM

1.1 TB HBM3e VRAM

Host128 cores · 2 TB RAM
FabricNVLink · 8 x 400G InfiniBand

€42.60 / hr

8 x B200 SXM

1.4 TB HBM3e VRAM

Host2 x 128 cores · 3 TB RAM
FabricNVLink 5 · 8 x 800G InfiniBand

€58.35 / hr

8 x B300 (Ultra)

2.3 TB HBM3e VRAM

Host2 x 128 cores · 4 TB RAM
FabricNVLink 5 · 8 x 800G InfiniBand

€74.25 / hr

Multi-node B300 pod

on request VRAM

Host16 – 64 nodes
FabricRail-optimised 800G fabric

Reserved capacity

04 / Platform

Built for reproducible science

Everything a lab needs around the silicon: scheduling, datasets, and audited access.

Slurm or Kubernetes

Bring your own scheduler or use our managed Slurm partitions with fair-share quotas per research group.

Data locality

Parallel filesystem with 400 GB/s aggregate throughput, snapshots and no egress fees between projects.

Compliance

EU regions, single-tenant hardware, SSO with institutional identity, full audit trail per job.

Allocation requests

Reserve capacity for your next run

Tell us your model size, dataset and deadline. We size the node — from a single RTX 5090 to a multi-node B300 pod — and return a quote within one working day.