GPU cloud for
European AI teams.
Rent 24 GB RTX 3090 compute from $0.15 per GPU hour. Launch through the API, connect over SSH, and request European or Frankfurt-oriented placement for the workloads that need it.
- 24 GBGDDR6X VRAM
- $0.15per GPU hour
- Europeplacement by request
- 10496CUDA cores
Built for these workloads
European capacity
One GPU product. The right location for the job.
BHK Cloud offers European and Frankfurt-oriented RTX 3090 capacity by request. Share the workload and preferred location, then receive written confirmation of the placement and currently available capacity before ordering.
- European and Frankfurt-oriented placement, subject to availability
- 24 GB RTX 3090 nodes for inference, fine-tuning, rendering, and batch work
- SSH and API access for the same launch workflow in every confirmed placement
- Clear confirmation before production deployment, including capacity and location
# send model, runtime, storage, and location
$ bhk capacity request --region frankfurt
→ deployment details confirmed in writing
Architecture
24 GB VRAM. Serious compute. Hourly pricing.
RTX 3090 with Ampere architecture, NVLink support, and GDDR6X memory bandwidth tuned for the throughput patterns of large-model inference and fine-tuning at 8-bit.
- 24 GB GDDR6X · 936 GB/s memory bandwidth
- 10,496 CUDA cores · 328 tensor cores (3rd gen)
- NVIDIA NVLink for multi-GPU jobs
- PCIe 4.0 · NVMe-backed persistent volumes
- Pre-baked images: PyTorch 2.x, TF 2.x, CUDA 12.x
# stream logs from running job
$ bhk gpu logs gpu-node-01 --follow
→ Epoch 3/50 · loss=2.089 · 87% GPU
How It Works
From key to running after confirmation.
Set an API key, launch a node, and push your first job after capacity confirmation.
Get an API Key
Log in at ai.bhkcloud.com/dashboard and generate a BHK_API_KEY. One key controls GPU, storage, and billing.
Launch a Node
Request an RTX 3090 with your preferred image. We confirm capacity before arranging access.
Run Your Job
SSH in, point the node at your own dataset, and run. Billing starts on launch and stops on terminate to the second.
Terminate & Pay
bhk gpu terminate gpu-node-01. Billing stops. No idle charges. RTX 3090 has no minimum and no annual commitment. Confirm invoicing terms.
Cluster Profiles
Right-size for your workload.
From single-card experiments to multi-GPU workloads. Pricing starts at $0.15 per GPU hour; topology and availability must be confirmed for parallel jobs.
| Profile | GPUs | VRAM | CUDA Cores | Ideal For | Price / hr |
|---|---|---|---|---|---|
| Single | 1× RTX 3090 | 24 GB | 10,496 | Inference, fine-tuning, experiments | $0.15 |
| Dual | 2× RTX 3090 | 48 GB total* | 20,992 | Larger models, parallel inference | from $0.30 |
| Quad Popular | 4× RTX 3090 | 96 GB total* | 41,984 | Software-parallel training and inference | from $0.60 |
| Octa | 8× RTX 3090 | 192 GB total* | 83,968 | Distributed workloads with validated topology | from $1.20 |
| Custom | 16–256× | 384 GB+ | Scalable | Dedicated clusters, reserved capacity | Custom |
*Total physical VRAM across cards is not automatically a single shared memory pool. Effective model capacity depends on framework support, interconnect topology, partitioning, and communication overhead.
Data that belongs next to the node: encrypted S3 object storage at $2.50 per TB per month, encrypted at rest. A secondary product for datasets, checkpoints and model weights. GPU capacity stays the main offer. Storage details →
Platform
Infrastructure without the overhead.
Everything a machine learning engineer needs, nothing a procurement team invented.
Pre-baked ML images
PyTorch 2.x, TensorFlow 2.x, CUDA 12.x, and cuDNN 9 ready to pull. Custom Docker images via --image flag.
Persistent NVMe volumes
Attach SSD-backed volumes that survive node restarts. Snapshot and clone between regions in one API call.
Bring your own data path
Attach persistent NVMe volumes, or read datasets from the object store you already run. Ask us about the network path before onboarding a large dataset.
SSH & REST access
Shell in directly or drive everything through the REST API. Both are first-class citizens, not afterthoughts.
Orchestration-ready
Terraform provider, Pulumi SDK, and Kubernetes CSI driver. Bring your existing infra-as-code workflow.
Encrypted at rest & in transit
AES-256 at rest, TLS 1.3 in transit. VRAM is wiped on node termination before the hardware returns to pool.
FAQ
GPU cloud, answered.
Questions we get asked before the first bhk gpu launch.
What workloads run well on the RTX 3090?
The RTX 3090's 24 GB VRAM is useful for LLM inference when the complete model, KV cache, concurrency, and runtime overhead fit in memory. Some roughly 30B-class models can fit with aggressive 4-bit quantization, depending on architecture and context. It is also useful for image generation, LoRA or QLoRA fine-tuning, rendering, and video processing.
How does hourly GPU billing work?
You are billed for every hour your GPU node is running, pro-rated to the second. RTX 3090 has no minimum and no annual commitment. Run a single experiment on RTX 3090 and terminate when done. Billing stops the moment you call bhk gpu terminate or terminate via the dashboard.
How fast is GPU provisioning?
Capacity and placement are confirmed before fulfillment. Ask our team about image options and setup for your workload.
Can I request European or Frankfurt GPU placement?
Yes. European and Frankfurt-oriented RTX 3090 capacity is available by request. Send the preferred location, model, expected runtime, and storage needs. We confirm the exact placement and currently available capacity in writing before you order, especially for production workloads.
Can I use my own Docker image?
Yes. BHK GPU nodes support custom Docker images via bhk gpu launch --image docker.io/yourrepo/yourimage:tag. The image is pulled and cached at the region edge. Private registries are supported with credential injection through the API.
Do you support multi-GPU jobs?
Yes. Nodes are available in 1×, 2×, 4×, and 8× RTX 3090 configurations. For distributed training across multiple nodes, use the Quad or Octa profiles with NVLink-bridged inter-GPU bandwidth. Larger clusters are available via the enterprise plan.
What if my model does not fit in 24 GB?
Usually that is a tuning problem before it is a hardware problem. We quantise the model, tune the serving stack and batch the workload so more of it fits in 24 GB, and we say plainly when it genuinely needs more VRAM. The model and the data stay inside your perimeter. Persistent NVMe volumes are available for checkpoints that must survive node restarts.
Ready to launch your first node?
Tell us about your workload and preferred cluster profile. We confirm capacity and access terms before fulfillment.