Rent the L40S if…
you serve 7B–30B models at FP8, run image, video or 3D pipelines, or want an always-on endpoint at the lowest monthly cost.
Deploy an L40S · $429/mo
NVIDIA L40S48 GB GDDR6 · Ada Lovelace
vs
NVIDIA H100 SXM80 GB HBM3 · Hopper
NVIDIA L40S vs H100: do you need the H100?
The inference workhorse against the training flagship: 48 GB of GDDR6 for $429 a month, or 80 GB of HBM3 with 3.9× the bandwidth for $1,249.
- L40S, dedicated$429per month · $0.59/h effective
- H100, dedicated$1,249per month · $1.71/h effective
- Monthly saving−66%the L40S costs $820 less a month
The short answer
Which one should you rent?
Most inference that fits in 48 GB does not need an H100. The L40S serves 7B–30B models at FP8, runs image and video generation and handles rendering for about a third of the H100's monthly price. The H100 SXM is the right tool when memory bandwidth or multi-GPU scaling is the bottleneck: training and fine-tuning, high-concurrency serving of large models, anything that needs NVLink across an 8-GPU node.
- Bandwidth. 3.35 TB/s of HBM3 against 864 GB/s of GDDR6: per-token latency and throughput at large batch sizes favour the H100 by a wide margin.
- Compute. About 989 TFLOPS of dense FP16 on the H100 SXM against 362 on the L40S; both support FP8.
- Memory. 80 GB against 48 GB: a 30B-class model at FP8 fits on either; a 70B-class model at FP8 needs H100-class memory or more.
- Price. The H100 costs 2.9× the L40S per month. Unless your job uses its bandwidth, compute and NVLink, that premium buys idle silicon.
Rent the H100 if…
you train or fine-tune, serve large models at high concurrency, or need NVLink between the GPUs of an 8-GPU node.
Deploy an H100 · $1,249/moSpecifications
L40S vs H100: the numbers side by side
NVIDIA's published figures for each part, tensor throughput without sparsity. The last column is how the L40S compares, as a multiple of the H100.
| Specification | NVIDIA L40S | NVIDIA H100 SXM | L40S vs H100 |
|---|---|---|---|
| Architecture | Ada Lovelace · 2023 | Hopper · 2022 | — |
| GPU memory | 48 GB GDDR6 | 80 GB HBM3 | 0.6× |
| Memory bandwidth | 864 GB/s | 3.35 TB/s | 0.26× |
| FP16 / BF16 tensor (dense) | 362 TFLOPS | ~989 TFLOPS | 0.37× |
| FP8 tensor (dense) | 733 TFLOPS | ~1,979 TFLOPS | 0.37× |
| FP4 tensor (dense) | Not supported | Not supported | — |
| Maximum power | 350 W | Up to 700 W | 0.5× |
| Form factor | PCIe Gen4 x16 | SXM5 | — |
| GPU-to-GPU link | PCIe only | NVLink 4 · 900 GB/s | — |
| Multi-Instance GPU (MIG) | No | Up to 7 instances | — |
| ECC memory | Yes | Yes | — |
| Video engines | 3 NVENC · 3 NVDEC (AV1) | 7 NVDEC · 7 NVJPG | — |
| CUDA compute capability | 8.9 | 9.0 | — |
| Dedicated host here | 12 vCPU · 96 GB RAM · 1 TB NVMe | 24 vCPU · 256 GB RAM · 3 TB NVMe | — |
| Price per month here | $429 | $1,249 | 0.34× |
Sources: NVIDIA datasheets and product pages for each GPU. Tensor figures are dense (NVIDIA often prints them doubled “with sparsity”). Prices are this site's flat monthly rates for a dedicated server with the host listed, in USD excluding VAT.
Value
What each dollar buys
Price per month divided into the two resources most jobs run out of first — memory and memory bandwidth — and what the same GPU costs kept on for a month at the market-median hourly rate.
Per GB of GPU memoryL40S $8.94 · H100 $15.61per month, dedicated host included
Per TB/s of bandwidthL40S $497 · H100 $373per month — what decode-heavy inference pays for
Effective hourly rateL40S $0.59/h · H100 $1.71/hflat monthly price over 730 hours
Same GPU on demand, 24/7L40S $1,124 · H100 $2,446market-median hourly rate × 730 h
Model sizes
What fits on one L40S and one H100
The largest model whose weights fit on a single GPU, keeping 10% of the memory for the runtime, activations and a small KV cache. Long contexts and large batches need more — size them with the GPU memory calculator.
| Weights precision | NVIDIA L40S | NVIDIA H100 SXM | Typical use |
|---|---|---|---|
| 16-bit (BF16 / FP16) | up to ~22B parameters | up to ~36B parameters | Full-precision inference; the base weights of a LoRA fine-tune |
| 8-bit (FP8 / INT8) | up to ~43B parameters | up to ~72B parameters | FP8 serving with vLLM or TensorRT-LLM on GPUs that support it; INT8 weights elsewhere |
| 4-bit (GPTQ, AWQ, Q4) | up to ~79B parameters | up to ~130B parameters | Quantised inference on a budget; QLoRA fine-tuning |
Parameters in billions, weights only: 2 bytes per parameter at 16-bit, 1 at FP8 or INT8, about 0.55 at 4-bit (GPTQ, AWQ, GGUF Q4). Mixture-of-experts models need memory for all their experts, not just the active ones.
Prices
L40S and H100 price per month, by term
Per dedicated GPU server, in USD excluding VAT — CPU cores, RAM, NVMe, 20 TB of transfer and 24/7 support included.
| Term | NVIDIA L40S | NVIDIA H100 SXM | Difference per month |
|---|---|---|---|
| Monthly | $429 | $1,249 | $820 |
| 3 months · −5% | $408 | $1,187 | $779 |
| 6 months · −10% | $386 | $1,124 | $738 |
| 12 months · −15% | $365 | $1,062 | $697 |
| Dedicated host | 12 vCPU · 96 GB RAM · 1 TB NVMe · 10 Gbps | 24 vCPU · 256 GB RAM · 3 TB NVMe · 25 Gbps | included |
L40S vs H100: questions answered
Is the L40S good enough for LLM inference?
Yes, for models up to about 30B parameters at FP8, or 70B at 4-bit, on one card. It is our recommended GPU for always-on inference: its lower bandwidth means higher per-token latency on large batches, but for a steady endpoint the monthly price difference is hard to ignore.
How much cheaper is the L40S than the H100?
Here, $429 against $1,249 a month for a dedicated server: the H100 costs 2.9× the L40S. Kept on for a month at market-median hourly rates, the same two GPUs cost $1,124 and $2,446.
Can I train on the L40S?
Fine-tuning small and mid-size models, yes — LoRA on 13B-class models, QLoRA on larger ones. For pre-training or full fine-tunes, its PCIe interconnect and GDDR6 bandwidth make the H100 or the A100 the better tool.
Are the L40S and the H100 in stock?
17 of 40 L40S units were free at the last stock update; 14 of 64 H100 units were free at the last stock update (18 September 2026). Both are provisioned automatically, in under 10 minutes after payment.
Can I rent the L40S or the H100 without KYC?
Yes. Both are rented with an email address and paid from a prepaid balance in Bitcoin, Ether or USDT — $429 and $1,249 a month, dedicated, with no identity check and no card.
Other GPU comparisons
Both are in stock, by the month
Dedicated, single-tenant and live in under 10 minutes: the L40S at $429 or the H100 at $1,249 a month, paid in BTC, ETH or USDT with no KYC.