Every configuration we offer, with the exact hardware in it. Filter by what your workload needs, compare up to three, and request the one that fits.
Sample dataHost specifications, regions, availability and prices are illustrative until the capacity service is connected. Accelerator specifications are vendor-published.
B300 · Blackwell Ultra
cl.b300.x8
8× 288 GB
224 vCPU · 4 TB
US-E
Reserved only
$6.20$49.60 per node-hour
- Accelerator
- NVIDIA B300 SXM
- GPU memory
- 8 × 288 GB HBM3e · 2.3 TB
- Memory bandwidth
- 8 TB/s per GPU
- Interconnect
- NVLink 5 · 1.8 TB/s
- Host CPU
- 2× Xeon 6 6960P · 224 threads
- System memory
- 4 TB
- Local NVMe
- 61.4 TB
- Network
- 8× 800 Gb/s
- Confidential computing
- Supported
Largest memory per GPU in the catalog. For long-context inference and frontier-scale training.
Published figures vary: 270–288 GB usable memory and 1,100 W+ board power. Confirm against the NVIDIA HGX B300 datasheet.
cloister nodes create --plan cl.b300.x8B200 · Blackwell
cl.b200.x8
8× 180 GB
224 vCPU · 2 TB
EU-NUS-E
Limited
$4.90$39.20 per node-hour
- Accelerator
- NVIDIA B200 SXM
- GPU memory
- 8 × 180 GB HBM3e · 1.4 TB
- Memory bandwidth
- 8 TB/s per GPU
- Interconnect
- NVLink 5 · 1.8 TB/s
- Host CPU
- 2× Xeon Platinum 8570 · 224 threads
- System memory
- 2 TB
- Local NVMe
- 30.7 TB
- Network
- 8× 400 Gb/s
- Confidential computing
- Supported
Two-die package with fifth-generation NVLink. Large-scale training and high-throughput serving.
cloister nodes create --plan cl.b200.x8H200 · Hopper
cl.h200.x8
8× 141 GB
224 vCPU · 2 TB
EU-NEU-CUS-E
Available
$3.10$24.80 per node-hour
- Accelerator
- NVIDIA H200 SXM
- GPU memory
- 8 × 141 GB HBM3e · 1.1 TB
- Memory bandwidth
- 4.8 TB/s per GPU
- Interconnect
- NVLink 4 · 900 GB/s
- Host CPU
- 2× Xeon Platinum 8480+ · 224 threads
- System memory
- 2 TB
- Local NVMe
- 30.7 TB
- Network
- 8× 400 Gb/s
- Confidential computing
- Supported
Hopper with 76% more memory than H100. Fits 70B-class models with room for KV cache.
cloister nodes create --plan cl.h200.x8H100 · Hopper
cl.h100.x8
8× 80 GB
224 vCPU · 2 TB
EU-NEU-CUS-E
Available
$2.40$19.20 per node-hour
- Accelerator
- NVIDIA H100 SXM
- GPU memory
- 8 × 80 GB HBM3 · 640 GB
- Memory bandwidth
- 3.35 TB/s per GPU
- Interconnect
- NVLink 4 · 900 GB/s
- Host CPU
- 2× Xeon Platinum 8480+ · 224 threads
- System memory
- 2 TB
- Local NVMe
- 30.7 TB
- Network
- 8× 400 Gb/s
- Confidential computing
- Supported
The established training and inference workhorse, with the deepest software ecosystem.
cloister nodes create --plan cl.h100.x8MI300X · CDNA 3
cl.mi300x.x8
8× 192 GB
192 vCPU · 3 TB
US-E
Waitlist
$2.10$16.80 per node-hour
- Accelerator
- AMD MI300X OAM
- GPU memory
- 8 × 192 GB HBM3 · 1.5 TB
- Memory bandwidth
- 5.3 TB/s per GPU
- Interconnect
- Infinity Fabric · 896 GB/s
- Host CPU
- 2× EPYC 9654 · 192 threads
- System memory
- 3 TB
- Local NVMe
- 30.7 TB
- Network
- 8× 400 Gb/s
- Confidential computing
- Not available
192 GB per accelerator on ROCm. Memory-bound inference without model sharding.
cloister nodes create --plan cl.mi300x.x8RTX PRO 6000 · Blackwell
cl.rtx6000.x8
8× 96 GB
128 vCPU · 1.5 TB
EU-CUS-E
Available
$1.45$11.60 per node-hour
- Accelerator
- NVIDIA RTX PRO 6000 PCIe
- GPU memory
- 8 × 96 GB GDDR7 · 768 GB
- Memory bandwidth
- 1.6 TB/s per GPU
- Interconnect
- PCIe Gen5
- Host CPU
- 2× EPYC 9555 · 128 threads
- System memory
- 1.5 TB
- Local NVMe
- 15.4 TB
- Network
- 2× 200 Gb/s
- Confidential computing
- Unconfirmed
Server Edition. Inference, fine-tuning and graphics workloads that fit in 96 GB.
cloister nodes create --plan cl.rtx6000.x8RTX PRO 6000 · Blackwell
cl.rtx6000.x2
2× 96 GB
32 vCPU · 384 GB
EU-NEU-CUS-E
Available
$1.55$3.10 per node-hour
- Accelerator
- NVIDIA RTX PRO 6000 PCIe
- GPU memory
- 2 × 96 GB GDDR7 · 192 GB
- Memory bandwidth
- 1.6 TB/s per GPU
- Interconnect
- PCIe Gen5
- Host CPU
- EPYC 9355P · 32 threads
- System memory
- 384 GB
- Local NVMe
- 3.8 TB
- Network
- 2× 100 Gb/s
- Confidential computing
- Unconfirmed
Server Edition. Inference, fine-tuning and graphics workloads that fit in 96 GB.
cloister nodes create --plan cl.rtx6000.x2L40S · Ada Lovelace
cl.l40s.x4
4× 48 GB
64 vCPU · 768 GB
EU-NEU-CUS-E
Available
$0.95$3.80 per node-hour
- Accelerator
- NVIDIA L40S PCIe
- GPU memory
- 4 × 48 GB GDDR6 · 192 GB
- Memory bandwidth
- 0.864 TB/s per GPU
- Interconnect
- PCIe Gen4
- Host CPU
- EPYC 9354 · 64 threads
- System memory
- 768 GB
- Local NVMe
- 7.7 TB
- Network
- 2× 100 Gb/s
- Confidential computing
- Not available
Cost-efficient inference, small-model fine-tuning and rendering.
cloister nodes create --plan cl.l40s.x4L40S · Ada Lovelace
cl.l40s.x1
1× 48 GB
16 vCPU · 192 GB
EU-NEU-C
Limited
$1.05$1.05 per node-hour
- Accelerator
- NVIDIA L40S PCIe
- GPU memory
- 1 × 48 GB GDDR6 · 48 GB
- Memory bandwidth
- 0.864 TB/s per GPU
- Interconnect
- PCIe Gen4
- Host CPU
- EPYC 9124 · 16 threads
- System memory
- 192 GB
- Local NVMe
- 1.9 TB
- Network
- 1× 25 Gb/s
- Confidential computing
- Not available
Cost-efficient inference, small-model fine-tuning and rendering.
cloister nodes create --plan cl.l40s.x1
No configuration matches every filter.
Loosen a requirement, or tell us what you need — we build to order for clusters.
Prices exclude storage, network add-ons and tax. Reserved terms apply the discounts on the pricing page.
0 / 3 selected