NVIDIA · Ada Lovelace
Dedicated NVIDIA L40S server, $356 a month.
The whole card — 48 GB of GDDR6 ECC at 864 GB/s — with 12 dedicated vCPU, 96 GB of RAM and 750 GB of NVMe, on a 25 Gbit/s port with free egress. Root in under 5 minutes. Priced from a cost sheet you can read below, with zero margin on the server.
- 1× to 8× cards, PCIe 4.0
- 4 regions, same price
- No setup fee, no KYC, crypto balance
- 99.9% SLA credited ×10
- 48 GBGDDR6 ECC per card
- 864 GB/smemory bandwidth
- PCIe 4.0interconnect
- 25 Gbit/sport, $0 egress
- 350 Wboard power, in the cost sheet
The cost sheet
What an L40S costs us, per card, per month.
Six lines, added up, rounded up to the next dollar. That is the price. The assumptions are the same for every card and are explained on the pricing page.
- Hardware$7,600 card + $2,500 chassis share, over 36 months$280.56
- Power276 kWh at $0.052 · 470 W, 70% load, PUE 1.15$14.36
- Rack & coolingPCIe chassis, air$25.00
- Network & IPtransit share, one IPv4, port$9.00
- Spares & failures2% of the card price per year$12.67
- Operationsmonitoring, on-call, provisioning$14.00
- Totalrounded up to the next dollar$355.58 → $356
Cheapest published price we found for the same card: $511 per month at Hyperstack L40 (reserved) (reserved), checked 2 September 2026 — 30% more than our cost. Where the money comes from, then: options with a normal margin, power bought at scale, full racks, no card network. Read how.
Configurations
One card to eight. The price is linear.
CPU, RAM and NVMe scale with the cards. Cards share the chassis over PCIe 4.0.
| Server | GPU memory | CPU | RAM | NVMe | Network | Per month | 12-month rate | |
|---|---|---|---|---|---|---|---|---|
| 1× L40S | 48 GB | 12 vCPU | 96 GB | 750 GB | 25 Gbit/s | $356 | $320.40 | Order |
| 2× L40S | 96 GB | 24 vCPU | 192 GB | 1.5 TB | 25 Gbit/s | $712 | $640.80 | Order |
| 4× L40S | 192 GB | 48 vCPU | 384 GB | 3 TB | 25 Gbit/s | $1,424 | $1,281.60 | Order |
| 8× L40S | 384 GB | 96 vCPU | 768 GB | 6 TB | 25 Gbit/s | $2,848 | $2,563.20 | Order |
Per month for a one-month term; the 12-month rate is the same server on a 12-month term (−10 %). Options — extra NVMe, IPv4, 25 Gbit/s uplink, private VLAN — are added on the order form at the prices on the pricing page.
Terms
Prepay longer, pay less per month.
A longer term costs us less to run — one provisioning, no re-image between tenants, a firmer power forecast — and the difference is passed on. The rate is locked for as long as the server renews.
| Term | Discount | Per month | Charged at order | Saved over the term |
|---|---|---|---|---|
| 1 month | cost sheet | $356.00 | $356.00 | — |
| 3 months | −3 % | $345.32 | $1,035.96 | $32.04 |
| 6 months | −6 % | $334.64 | $2,007.84 | $128.16 |
| 12 months | −10 % | $320.40 | $3,844.80 | $427.20 |
For one L40S; multiply by the card count. Renewal takes the same term at the same rate. How renewals, grace and suspension work.
Specifications
Per card, before you multiply.
- GPU
- NVIDIA L40S
- Architecture
- Ada Lovelace
- GPU memory
- 48 GB GDDR6 ECC
- Memory bandwidth
- 864 GB/s
- Interconnect
- PCIe 4.0
- Form factor
- PCIe, dual-slot, passively cooled
- Board power
- 350 W
- CPU per card
- 12 dedicated vCPU
- System RAM per card
- 96 GB ECC
- Local NVMe per card
- 750 GB
- Network
- 25 Gbit/s · one IPv4 · $0 egress
- Cards per server
- 1×, 2×, 4×, 8×
- Out-of-band
- IPMI console, power control, virtual media
What fits
Popular models on 1× L40S.
Memory needed = parameters × bytes per parameter × 1.2 (KV cache and runtime). Change the card count above and the table follows. The full guide.
| Model | BF16 | FP8 | INT8 | INT4 |
|---|---|---|---|---|
| Llama 3.1 8B | 20 GB | 10 GB | 10 GB | 5 GB |
| Mistral Small 3.1 24B | 58 GB | 29 GB | 29 GB | 15 GB |
| Gemma 3 27B | 65 GB | 33 GB | 33 GB | 17 GB |
| Qwen3 32B | 77 GB | 39 GB | 39 GB | 20 GB |
| Llama 3.3 70B | 168 GB | 84 GB | 84 GB | 42 GB |
| Qwen2.5 72B | 173 GB | 87 GB | 87 GB | 44 GB |
| Llama 4 Scout · 109B MoE | 262 GB | 131 GB | 131 GB | 66 GB |
| Mixtral 8x22B · 141B MoE | 339 GB | 170 GB | 170 GB | 85 GB |
| Qwen3 235B-A22B · MoE | 564 GB | 282 GB | 282 GB | 141 GB |
| Llama 3.1 405B | 972 GB | 486 GB | 486 GB | 243 GB |
| DeepSeek-V3 / R1 · 671B MoE | 1611 GB | 806 GB | 806 GB | 403 GB |
Green fits in 48 GB with a working KV cache; red does not. FP8 needs Ada, Hopper, Blackwell or MI300X silicon. MoE models count every expert.
Software
Pre-installed before you log in.
Pick the system and the template on the order form; serving templates pull the model you name before first boot. Every template, with its ports.
Operating systems
- Ubuntu 24.04 LTSKernel 6.8 · the default for every template
- Ubuntu 22.04 LTSKernel 5.15 HWE · for stacks pinned to CUDA 11.8 / 12.1
- Debian 12Bookworm · minimal, driver from the vendor repository
- Rocky Linux 9RHEL-compatible · Slurm, enterprise and HPC stacks
- Windows Server 2022Standard · RDP on :3389 · licence billed at our rate · $28/mo licence
Templates
- Bare OS + driver
- PyTorch 2.7
- TensorFlow 2.18 + JAX 0.5
- Jupyter Lab
- vLLM 0.9 + model
- SGLang 0.4 + model
- Text Generation Inference 3 + model
- Ollama + Open WebUI + model
- ComfyUI
- Stable Diffusion WebUI Forge
- Render node
- Docker + GPU toolkit
- Kubernetes node (k3s)
- Slurm worker
Root over SSH, IPMI console and an API token come with every server. Access docs.
What it is for
The jobs this card is right for.
-
Fine-tuning & inference
LoRA, QLoRA and full fine-tunes on 7B to 70B. vLLM and SGLang templates serve the result from the same server.
Also a fit. Our first pick here is the H100 SXM at $1,134/mo.
-
Image & video generation
Flux, SDXL, Wan and HunyuanVideo on GDDR7 cards that were built for exactly this. ComfyUI is one click away.
Also a fit. Our first pick here is the RTX 5090 at $163/mo.
Regions
Deployable in four regions, at the same price.
Reykjavík18 ms London · 40 ms New York100% geothermal + hydro
Helsinki27 ms Frankfurt · 36 ms London100% wind + hydro
Amsterdam7 ms Frankfurt · 12 ms London100% wind
Singapore38 ms Tokyo · 45 ms SydneyGrid + certified RECs
Guides
Read before you rent.
Questions
About the L40S server.
How much does a dedicated L40S server cost per month?
$356 per card per month for a one-month term, which is the sum of the six lines of its cost sheet rounded up to the dollar: hardware over 36 months, power at our contract rate, rack and cooling, network and IP, spares, operations. A 3, 6 or 12-month term is 3, 6 or 10 % cheaper per month ($320.40 on 12 months). No setup fee, no egress charge, no contract beyond the term.
Is the L40S dedicated or shared?
Dedicated. You rent a physical server with the whole card: no vGPU profile, no time-slicing, no other tenant on the machine. nvidia-smi shows the full 48 GB because you have the full 48 GB. The CPU cores, the RAM and the NVMe are yours too, and so is the IPMI console.
How many L40S cards can I have in one server?
1, 2, 4 or 8, at $356 per card: the chassis share is already in each card's cost, so the price is linear. Cards connect over PCIe 4.0. For NVLink all-to-all between cards, look at the SXM models.
Which LLMs fit on an L40S?
With the fit rule (parameters × bytes per parameter × 1.2), one card holds up to Qwen3 32B at FP8 and up to Qwen2.5 72B at INT4. An eight-card server (384 GB) holds up to Qwen3 235B-A22B · MoE at FP8. The order form checks the model you name against the server you pick before you pay.
Which operating systems and templates run on the L40S?
Ubuntu 24.04 LTS, Ubuntu 22.04 LTS, Debian 12, Rocky Linux 9, Windows Server 2022. Templates: Bare OS + driver, PyTorch 2.7, TensorFlow 2.18 + JAX 0.5, Jupyter Lab, vLLM 0.9, SGLang 0.4, Text Generation Inference 3, Ollama + Open WebUI, ComfyUI, Stable Diffusion WebUI Forge, Render node, Docker + GPU toolkit, Kubernetes node (k3s), Slurm worker. Serving templates pull the model you name before first boot.
How long does provisioning take, and can I cancel?
A single-card server is online in under 5 minutes, an eight-card node in under fifteen minutes, in any of the four regions. The term renews automatically from your balance at the locked price; turn renewal off and the server stops at the end of the paid term, or cancel at any time. No notice period.
What is the cheapest L40S rental elsewhere?
The lowest published monthly price we found for the same card on 2 September 2026 was $511 at Hyperstack L40 (reserved) (reserved; hourly rates converted at 730 hours). That is 30 % more than our cost sheet. We re-check every month and print the date.
Order an L40S now.
$356 a month at cost, online in under 5 minutes. Create an account, top up in crypto, pick a region.