On-demand workspace
Run your own image and commands. Connect with SSH, JupyterLab or a browser terminal. Release the GPU when the job is done.
GPU runtime + persistent storage
Explore GPU workspaces →GPU Cloud Rental
Rent RTX Pro 6000 GPUs from $1.89/GPU-hour, 96 GB Blackwell
Rent an RTX Pro 6000 on demand — NVIDIA Blackwell, 96 GB of VRAM, billed per second without a long-term commitment. Choose a workspace for inference, fine-tuning and rendering, a dedicated model endpoint, or whole nodes reserved by contract.
$1.89/ GPU-hourbilled per second$1 free credit on signupno minimum commitmentAvailable now
Eligible accounts can try a single GPU with less than one hour of credit. Minimum launch balance, capacity and account limits apply. Persistent storage is charged separately.
At a glance
RTX Pro 6000 Server Edition
- GPU
- RTX Pro 6000 (Blackwell)
- VRAM
- 96 GB
- Configurations
- 1 / 2 / 4 / 8 GPUs
- On-demand rate
- $1.89 / GPU-hour
Current on-demand rental rate: $1.89/GPU-hour, billed per second with no minimum commitment. See pricing for the live rate.
Configurations
RTX Pro 6000 rental pricing & live availability
One rental rate per GPU-hour, whatever the size. Every configuration runs on a single machine, so all GPUs share one host. Billed per second, no minimum commitment.
1× RTX Pro 6000$1.89/hr
Available now
- vCPU
- 16
- RAM
- 96 GB
- VRAM
- 96 GB
2× RTX Pro 6000$3.78/hr
Available now
- vCPU
- 32
- RAM
- 192 GB
- VRAM
- 192 GB
4× RTX Pro 6000$7.56/hr
Available now
- vCPU
- 64
- RAM
- 384 GB
- VRAM
- 384 GB
8× RTX Pro 6000$15.12/hr
Limited
- vCPU
- 200
- RAM
- 1354 GB
- VRAM
- 768 GB
Availability is live and moves with demand. Last updated . Need a size that shows as on request? Talk to us — we can usually free capacity.
Plan your rental
How RTX Pro 6000 rental billing works
Estimate GPU runtime
GPU charge = GPU count × current GPU-hour rate × runtime in seconds ÷ 3,600. A 30-minute run on one GPU uses half a GPU-hour; four GPUs for 30 minutes use two GPU-hours. Storage is billed separately.
Check the current rental rate →Review region and capacity
Choose the region and configuration in the launch form and review the current quote. Capacity can change between viewing this page and launching. GPU count, model precision and context length all affect which configuration suits your workload.
Compare model deployment recipes →Keep the files you need
Save weights and outputs to a Cloud Drive or Shared Filesystem before terminating the workspace. Temporary container data is deleted at termination. Persistent storage continues billing until you delete it.
Choose persistent storage →What it's good for
Workloads that fit 96 GB
RTX Pro 6000 is EcoHash's hero GPU for cost-effective open-model inference and GPU workspaces.
20B–35B open models
96 GB of VRAM fits popular 20B–35B models with room for larger batch sizes and KV-cache headroom.
LoRA & QLoRA fine-tuning
Adapter fine-tuning and experimentation on a single Blackwell GPU, with root access and a browser terminal.
Quantized larger models
Validate selected quantized larger models where 96 GB is enough for single-GPU inference.
Batch & data jobs
Batch inference, data preprocessing, and offline evaluation on workstation-class GPUs.
Best-fit models
Popular models on RTX Pro 6000
Start with these open models through the API, then move to a dedicated endpoint or GPU workspace as usage grows.
Qwen3 Coder 30BCoding & agentsQwen3.8-27BVision & documentsGemma 31BGeneral & visiongpt-oss 20BLow-cost general
Voice and lightweight models (such as Kokoro TTS and Whisper STT) are available through the same API and can scale onto EcoHash GPU capacity for high-throughput workloads — they don't require a dedicated RTX Pro 6000.
Choose your setup
A workspace, an endpoint, or whole nodes?
Dedicated inference
Deploy a model as an endpoint on dedicated GPUs. Call it through the unified API instead of managing a workspace yourself.
GPU runtime + persistent storage
Explore model endpoints →Reserved GPU nodes
Keep whole nodes exclusively for your account over a contracted term. Run your workspaces, model instances, clusters and training there.
Contracted capacity + storage quotas
Explore reserved nodes →Start from a model
Use an existing recipe. Keep control of your GPU.
Choose a published template or available model weights instead of assembling every part of a deployment from scratch.
- 01
Choose how to run it
A GPU-instance template gives you a workspace with shell access. A model-instance template creates a serving endpoint. The template type determines what you launch.
- 02
Review the configuration
Check the image, startup command, GPU count, region and current quote. Where published weights are available, mount them without downloading another copy.
- 03
Launch when you are ready
Opening a template does not create a paid instance. Launch from the console, or use the template with the Compute API or GPU MCP.
Published community weights are read-only and tied to their available region. Save outputs to your own storage. Model terms, capacity and template availability still apply.
GPU instances
A GPU workspace with root access
Provision exactly what a job needs and tear it down when you're done — root access, your choice of image, and per-second billing at the GPU-hour rate.
1, 2, 4 or 8 GPUs per instance
Each configuration includes CPU, RAM and local scratch; the table above shows its CPU and RAM allocation. Bring any container image, or start from the CUDA, JupyterLab or vLLM images we maintain.
SSH, browser terminal, JupyterLab
Native SSH for scp, rsync and VS Code Remote; a shell in any browser with no key required; JupyterLab wired up automatically for notebook images.
File upload & public service URL
Upload files straight into the pod from the console, and expose any HTTP port inside the container at a public URL.
Auto-expiry protection
Set a duration when you launch and the instance stops on its own — extend it any time. Or launch with no end time and stop it yourself (or via the Compute API). GPU runtime is billed per second. Persistent storage remains separately billed until deleted.
3D rendering & ray tracing
Render in Blender, Octane and more on Blackwell RT cores with 96 GB — workstation-class GPUs that clouds renting H100/A100 don't offer. Need a 4× or 8× render node? Pick the size at launch.
Persistent storage
Attach a Cloud Drive or Shared Filesystem in the launch form; it mounts under /workspace and outlives the instance.
GPU clusters
The same instance, replicated
A cluster is N identical replicas of one instance configuration in one region, sharing a filesystem and a single endpoint — for data-parallel training, multi-replica serving or any worker pool that wants horizontal scale.
2–8 identical replicas
One image, one startup command, one GPU configuration — repeated across every replica and managed from a single panel.
One load-balanced endpoint
Every replica sits behind the same HTTPS URL, so a multi-replica model server or worker pool looks like one service to callers.
Shared Filesystem on every replica
Attach a Shared Filesystem at the cluster level and it mounts at the same path on all replicas — one copy of the weights, one copy of the data.
Scale replicas in place
Change the replica count between 1 and 8 while the cluster runs. Per-replica terminals and uploads are available from the console.
Storage
Persistent storage that travels with your work
Keep datasets and model weights close to the GPUs. Storage outlives any single instance and reattaches on demand.
Cloud Drive
Block storage (read-write-once) for a single instance — a durable workspace that survives restarts.
Shared Filesystem
CephFS-backed shared storage (read-write-many) mounted across instances and cluster replicas — ideal for datasets and model weights.
Sizes, mount paths, export and the live rate table are on the Storage page.
GPU MCP & Compute API
Let your AI agent launch the GPU
Use Claude Code, Codex or another compatible MCP client to find capacity, quote a workspace and manage it on your EcoHash account.
Check capacity and cost
Ask your agent to find available GPUs and quote a launch before it spends credit.
Launch, inspect and terminate
Create a GPU instance, follow its status and get the SSH command. Work over SSH, save your results, then terminate the instance to release its GPUs.
Use your existing setup
Launch saved templates, manage GPU clusters and persistent storage, or launch and follow a dedicated model instance from a template.
Choose the agent's access
Compute Read-only allows viewing and quoting GPU resources. Full Compute also allows launching, changing and terminating them. Model-call permissions are separate.
Try asking your agent
“Find an RTX Pro 6000, quote a GPU workspace with my SSH key, and ask me before launching it.”
MCP server: https://mcp.ecohash.com/gpu
Connect with EcoHash sign-in (OAuth) or an API key with Compute access. Viewing and quoting are free; GPU usage and storage are billed to your account at the usual rates. Terminating deletes temporary instance data. Persistent storage remains and continues billing until deleted.
Boundaries
What it's not for
Choose a configuration from the model’s actual memory requirements, quantization, context length and tested deployment recipe. Published templates include multi-GPU configurations; parameter count alone does not determine fit. Each GPU has its own 96 GB of VRAM — multiple GPUs do not form one pooled memory space. This is PCIe capacity, not an NVLink training cluster. InfiniBand for eligible multi-node workloads requires separate configuration; it is not included with every workspace or cluster.
Guides and next steps
- Best models to run on RTX Pro 6000 96GB →
Choose a model for coding, chat, retrieval or speech before deciding how to run it.
- Qwen-Image-2.1 on RTX PRO 6000: quality and performance test →
Inspect 15 fixed-seed image samples, generation settings and measured runtimes.
- Deploy and tune Qwen3.8-27B on one RTX Pro 6000 →
Follow the serving configuration and tuning decisions used in the deployment guide.
- Run DeepSeek with an EcoHash GPU template →
Review the quote, launch a template and connect to the resulting endpoint.
- Launch an 8-GPU RTX PRO 6000 instance →
Configure a whole-node workspace, SSH, an HTTP endpoint and persistent storage.
FAQ
Common questions
- How much does it cost to rent an RTX Pro 6000 per hour?
- RTX Pro 6000 rents for $1.89 per GPU-hour on EcoHash, billed per second with no minimum commitment. See the pricing page for the live rate.
- What models fit well on a single RTX Pro 6000 (96GB)?
- Start with Qwen3 Coder 30B, Qwen3.8-27B, Gemma 31B or gpt-oss 20B. Check the full checkpoint, precision, context length and KV-cache requirements before choosing a configuration. The deployment guides and measured model pages show specific RTX Pro 6000 setups; parameter count alone does not guarantee fit.
- Can I rent RTX Pro 6000 by the hour?
- Yes. Launch an hourly RTX Pro 6000 GPU workspace with root access and a browser terminal, deploy a dedicated inference endpoint, or call models through the OpenAI-compatible API.
- Can I start a GPU with my welcome credit?
- Eligible accounts can launch a single on-demand GPU even when the balance covers less than an hour. The available runtime depends on the current rate, usable credit and any storage charges. Minimum launch balance, account eligibility and capacity limits apply; the launch form shows the applicable quote or restriction. Save your results before the credit runs out.
- Does stopping the GPU stop storage charges?
- Terminating the instance releases its GPU and deletes temporary container data. Cloud drives and shared filesystems remain and continue billing until deleted. Save your work to persistent storage or copy it out before terminating, and delete storage you no longer need.
- Can my team reserve whole GPU nodes?
- Yes. Whole-node reservations are arranged by contract for an agreed region, capacity, term and storage quota. They are separate from on-demand workspaces and dedicated model endpoints. Contact sales to plan a reservation; it is not an instant self-service purchase.