STRATA Reserve capacity
Strata/GPUs/RTX PRO 6000

Rent NVIDIA RTX PRO 6000 Blackwell 96GB

Blackwell96 GB GDDR71.8 TB/s

The RTX PRO 6000 Blackwell puts 96 GB of memory on a single PCIe card. It is one of the most cost-effective GPUs for inference of mid-size models and for image and video generation.

$2.03/GPU-hr · On demand$1.97/GPU-hr · 3 months$1.91/GPU-hr · 6 months$1.83/GPU-hr · 12 months$1.75/GPU-hr · 24–36 months
NVIDIA RTX PRO 6000 Blackwell 96GB GPU

Specifications

GPU memory96 GB GDDR7 with ECC
Memory bandwidth1.8 TB/s
Tensor cores5th generation, FP4 and FP8
Form factorPCIe 5.0, server edition
VideoHardware encode and decode engines
Typical node1–8 GPUs per server

What the RTX PRO 6000 is best for

How renting works

On-demand GPUs start in minutes and are billed per minute from a prepaid balance. Dedicated RTX PRO 6000 clusters are allocated to you alone for 3 to 36 months at a lower, locked rate. Pay by card, bank transfer, USDT or USDC.

Compare

RTX PRO 6000 vs H100

Questions

Where can I rent an RTX 4090 or RTX 5090 GPU? Strata GPU vs RunPod: which one should I use? Strata GPU vs Vast.ai: which is better for renting GPUs? What is the cheapest way to rent H100 GPUs? What GPU do I need to run Llama 70B or another 70B model? How do I rent a GPU server for AI? A step-by-step guide What happens if on-demand GPUs are unavailable? Get 1-hour H100 access for benchmarks How to stop paying for idle rented GPUs (and surprise bills) Bare-metal SSH GPU server vs managed GPU pod for LLMs Cloud GPUs for self-hosting ComfyUI (images and video) Which GPU to rent for models needing 80 GB, 130 GB or more VRAM How to stop re-downloading model weights on rented GPUs GPU and network volume in different datacenters: how to avoid it Rent H100 and H200 GPUs in Asia-Pacific: Singapore, Tokyo, Mumbai

FAQ
Is the RTX PRO 6000 good for LLM inference?

Yes. With 96 GB of memory it serves 30B models in 16-bit or 70B models in 4-bit on one card, at a lower hourly price than data-center HBM GPUs.

How does it compare to the H100?

The H100 has much faster HBM memory and NVLink, so it wins for training and large-scale serving. The RTX PRO 6000 has more memory per card and costs less per hour, which suits single-GPU inference and generation.

Can I use it for video generation?

Yes. 96 GB is enough for current open video models at high resolution, and the card has hardware video encoders for the final output.

Other GPUs