1× GPU
Starting from · excl. VAT 29,000 Kč $1,335/GPU/month (~$1.85/GPU/hr)
- GPU
- NVIDIA RTX PRO 6000 Blackwell Max-Q
- GPUs
- 1
- Total VRAM
- 96 GB
- Interconnect
- Single GPU
NVIDIA RTX PRO 6000 Blackwell Max-Q — 96 GB dedicated GDDR7 per GPU, bare-metal SSH, in an ISO 27001 / ISO 9001 data center. One GPU or a connected 2, 4, or 8 GPU node.
Dedicated GDDR7, 1.8 TB/s-class memory bandwidth, and a standard x86 CUDA stack (vLLM, PyTorch). Longer rentals can include a discount.
01 Pricing
Every configuration uses NVIDIA RTX PRO 6000 Blackwell Max-Q (96 GB VRAM each). Multi-GPU options ship as one connected node over PCIe 4.0. Prices are starting from 29,000 Kč per GPU per month — $1,335/GPU/month (~$1.85/GPU/hr) — excl. VAT.
Starting from · excl. VAT 29,000 Kč $1,335/GPU/month (~$1.85/GPU/hr)
Starting from · excl. VAT 58,000 Kč $2,670/month ($1,335/GPU · ~$1.85/GPU/hr)
Starting from · excl. VAT 116,000 Kč $5,340/month ($1,335/GPU · ~$1.85/GPU/hr)
Starting from · excl. VAT 232,000 Kč $10,680/month ($1,335/GPU · ~$1.85/GPU/hr)
Longer rentals – Discounts are available the longer you rent. Email info@koscompute.com with duration and GPU count for an individual offer.
02 Included
Datacenter power and network connectivity are included in the monthly rate.
Root-level SSH access to the machine so you can install drivers, runtimes, and your own stack.
We help with access, connectivity, and hardware issues during your rental.
Hardware runs in Prague under ISO 27001 and ISO 9001 certified data-center operations.
03 FAQ
How the rental works, what the starting price covers, and what you can run. For a quote, email info@koscompute.com with the GPU count and how long you need the machine.
The monthly rate includes datacenter power, network connectivity, and basic support. You get bare-metal SSH with root access on a machine in our Prague data center (ISO 27001 and ISO 9001). You install the drivers, runtimes, and software you want to run.
Listed prices are starting rates for a monthly rental, excluding VAT: 29,000 Kč per GPU, which is $1,335/GPU/month (about $1.85/GPU/hr). The minimum term is 5 days — we do not rent by the hour. A longer rental can include a discount, so the figure on the card is where the price starts. Email us the GPU count and duration for an individual offer.
NVIDIA RTX PRO 6000 Blackwell Max-Q, with 96 GB of dedicated GDDR7 on each GPU. Choose one GPU, or a single node with 2, 4, or 8 GPUs (192 GB, 384 GB, or 768 GB). Multi-GPU nodes are connected inside one machine over PCIe 4.0.
These are standard x86 CUDA machines. You can run vLLM, PyTorch, TensorRT-LLM, Docker, or your own image. A 70B-class model at high quality fits on one 96 GB GPU. Bigger models, long context, or more people using the model at once belong on a 2–8 GPU node. Use the rental for inference, fine-tuning, training, or any other workload you operate yourself.
You do. Access is root SSH, so the machine is under your control for the rental. We help when the problem is access, the network, or the hardware. We do not log in to run your jobs. If you want us to host the model and you only send API requests, that is the managed API on the home page, not this rental.
In a certified data center in Prague. Email info@koscompute.com, say whether you want 1, 2, 4, or 8 GPUs, and say how long you need them. We reply with availability and a tailored offer. Prices on this page exclude VAT.
04 Choose the right path
Call open-source models over our managed API – no hardware to operate. Start on the home page.
Explore the APIRent bare-metal GPUs for custom training, fine-tuning, inference, or any workload you run yourself.
You are hereBuy a Kosmik AI server configured for local inference, with warranty and units in stock.
Buy AI ServersTell us how many GPUs you need and for how long. We reply with availability and a tailored offer.