GPU servers · Ada Lovelace
NVIDIA L40S GPU server
An NVIDIA L40S with 48 GB of GDDR6 ECC on a 32-core EPYC 9355P host with 256 GB of ECC memory: production inference and rendering, billed monthly, built to order in 72 hours.
NVIDIA L40S
$549/mo $909 −40%
- GPU
- L40S · 48 GB
- Host
- AMD EPYC 9355P · 32C/64T
- Memory
- 256 GB DDR5 ECC
- Storage
- 2 × 3.84 TB NVMe
- Network
- 10 Gbps unmetered
- Delivery
- Built to order · 72 h
AMSFRA
Deploy NVIDIA L40S · $549/mo- 10 Gbps unmetered port
- Always-on DDoS protection
- IPMI / KVM access
- IPv4 + routed /64 IPv6
- $0 setup, cancel anytime
Specifications
NVIDIA L40S specifications.
| GPU | NVIDIA L40S · 48 GB · Ada Lovelace |
|---|---|
| GPU memory | 48 GB GDDR6 ECC |
| Tensor formats | FP8 · BF16 |
| Processor | AMD EPYC 9355P · Zen 5 |
| Cores / threads | 32 cores / 64 threads |
| Base / boost clock | 3.55 / 4.4 GHz |
| Memory | 256 GB DDR5 ECC |
| Storage | 2 × 3.84 TB NVMe |
| Disk layouts | RAID 1 (3.84 TB) · RAID 0 (7.68 TB) · No RAID (2 × 3.84 TB) |
| Network port | 10 Gbps, dedicated to this server |
| Traffic | Unmetered: no quota, no overage, no 95th percentile |
| IP addresses | 1 IPv4 + routed /64 IPv6 · extra IPv4 $2.50/mo |
| DDoS protection | Always on, included |
| Remote access | IPMI / KVM console in your browser |
| Delivery | Built to order — ready within 72 hours |
Processor and GPU figures are the manufacturers’ public specifications. One-click installs: Debian, Ubuntu, AlmaLinux, Rocky Linux, Proxmox VE and FreeBSD; any other OS from an ISO over IPMI.
Why this server
What the NVIDIA L40S is good at.
48 GB of ECC memoryQuantized 30B to 70B models and batched serving on one data center GPU.
EPYC 9355P host32 Zen 5 cores and 256 GB of ECC DDR5, so data loading never starves the GPU.
Serve without egress billsPush inference responses and pull model weights at 10 Gbps without counting gigabytes.
Recommended for
Price
The same hardware, 40% below the market.
Median public list price of the same processor, memory and storage at major European and US hosts, with their standard 1 Gbps port. Their 10 Gbps unmetered option comes on top; here it is included.
NVIDIA L40S256 GB · 2×3.84 TB NVMe
−40%−64% vs. 10G offers
Elsewhere $909 for the hardware plus about $630 for 10 Gbps unmetered, $1,539 in total; PowerDedi $549 with 10 Gbps unmetered included.
Monthly USD list prices checked September 2026, setup fees and VAT excluded. No setup fee here, no contract, and the price is locked in USD: you pay in crypto at the rate of the day. How we price below the market
Locations
Deploy the NVIDIA L40S in 2 cities.
Same hardware, same price and the same 10 Gbps unmetered port in every city. Pick the one closest to your users.
Compare
Other GPU server configurations.
| Server | GPU | Memory | Storage | Delivery | Price |
|---|---|---|---|---|---|
| RTX PRO 4000 Blackwell | RTX PRO 4000 · 24 GB | 64 GB DDR5 | 2 × 1.92 TB NVMe | 24 h | $169/mo |
| NVIDIA L40S This server | L40S · 48 GB | 256 GB DDR5 ECC | 2 × 3.84 TB NVMe | 72 h | $549/mo |
| RTX PRO 6000 | RTX PRO 6000 · 96 GB | 384 GB DDR5 ECC | 2 × 3.84 TB NVMe Gen5 | 72 h | $899/mo |
FAQ
NVIDIA L40S: questions, answered.
Still have a question? Customers with an active server get answers from engineers in a support ticket.
Support ticketsL40S or RTX PRO 6000 Blackwell?
The L40S has 48 GB and is a proven data center card for FP8 inference. The RTX PRO 6000 doubles the memory to 96 GB of GDDR7 and adds FP4: it fits 70B models at 8-bit and LoRA fine-tuning, for $350 more a month.
What fits in 48 GB?
About 70B parameters at 4-bit, or 30B-class models at 8-bit with room for the KV cache.
How fast is the NVIDIA L40S delivered?
It is built to order and ready within 72 h of your payment. You follow each step from your client area.
Which cities can I deploy it in?
Amsterdam, Frankfurt. Every location has the same 10 Gbps unmetered port, DDoS protection and IPMI.
Which cryptocurrencies do you accept?
Five: Bitcoin (BTC), Ethereum (ETH), Monero (XMR), Tether (USDT) on the TRON network (TRC-20) and Solana (SOL). Crypto payments
Can I upgrade or cancel at any time?
Yes. Billing is monthly with no contract. Extra IPv4 addresses can be added at any time; memory and disk upgrades are quoted per configuration. Cancel whenever you want: your server stays active until the end of the period you paid for.
Guides
Related guides
- gpt-oss-120b and gpt-oss-20b hardware requirements: VRAM and GPUsWhat OpenAI’s gpt-oss-120b and gpt-oss-20b need to run: weight sizes from the model card, the memory each token of context takes, single-GPU serving with vLLM and Ollama, documented GPU support, and which 24, 48 and 96 GB cards fit which model.
- RTX PRO 6000 rental in 2026: hourly cloud prices vs a monthly serverWhat an NVIDIA RTX PRO 6000 costs by the hour at the main GPU clouds and marketplaces in October 2026, the break-even point between hourly billing and a monthly server, the costs the hourly rate leaves out, and when hourly still wins.
- The 2026 DRAM squeeze: what it means for dedicated server buyersServer DRAM contract prices rose at record quarterly rates from late 2025 into 2026. The data, why it happened, what it does to dedicated-server pricing and how to rent only the memory you need.
Deploy the NVIDIA L40S for $549 a month.
Pay in Bitcoin, Ethereum, Monero, USDT or Solana. 10 Gbps unmetered, no setup fee, no contract.
Ready in 72 hCancel anytimeBTC, ETH, XMR, USDT or SOL