All 6 data centres operational

Paid in crypto · No identity check · Root access in under 5 minutes

GPU servers

Every card we rent, and what each one is for.

Twelve NVIDIA GPUs, from a 24 GB L4 at $125 a month to an eight-card B200 node. Every one is dedicated, rented by the month, paid in crypto and opened without an identity check.

  • 12GPU models in the fleet
  • 41configurations, 1 to 10 cards
  • $125the cheapest card, per month
  • 6data centres, all cards in all of them

Pick by memory, or by what you need to run

Sorted by entry price. “Largest model that fits” is the biggest LLM the single card holds at 4-bit and 8k context — computed, not asserted, by the same arithmetic the configurator uses.

Every NVIDIA GPU available, with memory, bandwidth, node sizes and entry price
Card Architecture Memory Bandwidth Largest model that fits Node sizes From
L4 PCIe card Ada Lovelace 24 GBGDDR6 300 GB/s Qwen 3 32B 1×, 2×, 4×, 8×, 10× $125/mo
RTX 4090 PCIe card Ada Lovelace 24 GBGDDR6X 1,008 GB/s Qwen 3 32B 1×, 2×, 4×, 8× $193/mo
RTX A6000 PCIe card Ampere 48 GBGDDR6 ECC 768 GB/s Llama 3.3 70B 1×, 2×, 4×, 8× $286/mo
RTX 5090 PCIe card Blackwell 32 GBGDDR7 1,792 GB/s Qwen 3 32B 1×, 2×, 4×, 8× $335/mo
A100 PCIe PCIe card Ampere 40 GBHBM2 1,555 GB/s Qwen 3 32B 1×, 2×, 4×, 8× $392/mo
L40S PCIe card Ada Lovelace 48 GBGDDR6 ECC 864 GB/s Llama 3.3 70B 1×, 2×, 4×, 8× $714/mo
A100 PCIe PCIe card Ampere 80 GBHBM2e 1,935 GB/s Llama 3.3 70B 1×, 2×, 4×, 8× $1,091/mo
H100 PCIe PCIe card Hopper 80 GBHBM3 2,000 GB/s Llama 3.3 70B 1×, 2×, 4×, 8× $1,469/mo
A100 SXM4 SXM module, NVLink Ampere 80 GBHBM2e 2,039 GB/s Llama 3.3 70B 4×, 8× $4,395/mo
H100 SXM5 SXM module, NVLink Hopper 80 GBHBM3 3,350 GB/s Llama 3.3 70B 4×, 8× $6,061/mo
H200 SXM5 SXM module, NVLink Hopper 141 GBHBM3e 4,800 GB/s Qwen 3 235B-A22B (MoE) 4×, 8× $8,405/mo
B200 SXM6 SXM module, NVLink Blackwell 180 GBHBM3e 8,000 GB/s Qwen 3 235B-A22B (MoE) 4×, 8× $11,790/mo

Every card links to its own page: full specification, price at each node size and term, the models it holds at each precision, and the throughput to expect.

Not sure which card holds your model?

Pick a model, a precision and a context length; the configurator reports every configuration that fits and what each one costs.

Open the configuratorRead the guides

Related pages

Sign in

Console, invoices and out-of-band access.

No account yet?

There is no separate sign-up. Your account is created while you place your first order — you choose the email and the password on the payment step, and the console is open by the time the machine is.

Configure a server

Language