Rent the L4 if…
your model fits in 24 GB — 7B–8B LLMs at FP8, Whisper, embeddings, rerankers — or you transcode video, and you want the lowest monthly price for an always-on GPU.
Deploy an L4 · $169/mo
NVIDIA L424 GB GDDR6 · Ada Lovelace
vs
NVIDIA L40S48 GB GDDR6 · Ada Lovelace
NVIDIA L4 vs L40S: which inference GPU should you rent?
Two Ada Lovelace data center GPUs built for inference. The L4 is the small, efficient one — 24 GB in 72 W; the L40S doubles the memory and brings nearly 3× the bandwidth and tensor throughput, for 2.5× the monthly price.
- L4, dedicated$169per month · $0.23/h effective
- L40S, dedicated$429per month · $0.59/h effective
- Monthly saving−61%the L4 costs $260 less a month
The short answer
Which one should you rent?
Rent the L4 for services whose model fits in 24 GB and whose traffic is steady: 7B–8B chat models at FP8, speech-to-text, embeddings, rerankers and video transcoding with AV1. It is the cheapest way to keep a GPU online all month. Move to the L40S when the model or the batch outgrows 24 GB — 13B–30B models at FP8, image and video generation — or when per-token latency matters: its memory bandwidth is nearly three times higher.
- Memory. 24 GB on the L4 against 48 GB on the L40S, both GDDR6 with ECC.
- Bandwidth and compute. 300 GB/s and about 121 TFLOPS of dense FP16 on the L4, against 864 GB/s and about 362 TFLOPS on the L40S.
- Power. 72 W against 350 W: the L4 is a low-profile card designed for dense, efficient inference.
- Video. Both carry Ada video engines with AV1 and RT cores: two encoders and four decoders on the L4, three of each on the L40S.
Rent the L40S if…
you serve 13B–30B models, generate images or video, need lower latency per token, or render: 48 GB and 864 GB/s give room the L4 does not have.
Deploy an L40S · $429/moSpecifications
L4 vs L40S: the numbers side by side
NVIDIA's published figures for each part, tensor throughput without sparsity. The last column is how the L4 compares, as a multiple of the L40S.
| Specification | NVIDIA L4 | NVIDIA L40S | L4 vs L40S |
|---|---|---|---|
| Architecture | Ada Lovelace · 2023 | Ada Lovelace · 2023 | — |
| GPU memory | 24 GB GDDR6 | 48 GB GDDR6 | 0.5× |
| Memory bandwidth | 300 GB/s | 864 GB/s | 0.35× |
| FP16 / BF16 tensor (dense) | 121 TFLOPS | 362 TFLOPS | 0.33× |
| FP8 tensor (dense) | 242 TFLOPS | 733 TFLOPS | 0.33× |
| FP4 tensor (dense) | Not supported | Not supported | — |
| Maximum power | 72 W | 350 W | 0.21× |
| Form factor | PCIe Gen4 x16, low profile | PCIe Gen4 x16 | — |
| GPU-to-GPU link | PCIe only | PCIe only | — |
| Multi-Instance GPU (MIG) | No | No | — |
| ECC memory | Yes | Yes | — |
| Video engines | 2 NVENC · 4 NVDEC (AV1) | 3 NVENC · 3 NVDEC (AV1) | — |
| CUDA compute capability | 8.9 | 8.9 | — |
| Dedicated host here | 6 vCPU · 48 GB RAM · 500 GB NVMe | 12 vCPU · 96 GB RAM · 1 TB NVMe | — |
| Price per month here | $169 | $429 | 0.39× |
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 memoryL4 $7.04 · L40S $8.94per month, dedicated host included
Per TB/s of bandwidthL4 $563 · L40S $497per month — what decode-heavy inference pays for
Effective hourly rateL4 $0.23/h · L40S $0.59/hflat monthly price over 730 hours
Same GPU on demand, 24/7L4 $657 · L40S $1,124market-median hourly rate × 730 h
Model sizes
What fits on one L4 and one L40S
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 L4 | NVIDIA L40S | Typical use |
|---|---|---|---|
| 16-bit (BF16 / FP16) | up to ~11B parameters | up to ~22B parameters | Full-precision inference; the base weights of a LoRA fine-tune |
| 8-bit (FP8 / INT8) | up to ~22B parameters | up to ~43B parameters | FP8 serving with vLLM or TensorRT-LLM on GPUs that support it; INT8 weights elsewhere |
| 4-bit (GPTQ, AWQ, Q4) | up to ~39B parameters | up to ~79B 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
L4 and L40S 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 L4 | NVIDIA L40S | Difference per month |
|---|---|---|---|
| Monthly | $169 | $429 | $260 |
| 3 months · −5% | $161 | $408 | $247 |
| 6 months · −10% | $152 | $386 | $234 |
| 12 months · −15% | $144 | $365 | $221 |
| Dedicated host | 6 vCPU · 48 GB RAM · 500 GB NVMe · 10 Gbps | 12 vCPU · 96 GB RAM · 1 TB NVMe · 10 Gbps | included |
L4 vs L40S: questions answered
Can the L4 run Llama 3 8B?
Yes: at FP8 the weights take about 8 GB and at 16-bit about 16 GB, leaving room for the KV cache in 24 GB. It is the smallest GPU we recommend for a production 8B endpoint.
Is the L40S worth 2.5× the price of the L4?
If you use its memory and bandwidth, yes: it holds models the L4 cannot and generates tokens up to about three times faster. If your model fits in 24 GB and traffic is moderate, the L4 does the job for $169 a month — add L4s before moving up.
Which is better for video transcoding?
The L4 is designed for it: AV1, H.265 and H.264 encode and decode in 72 W, at the lowest price in the catalogue. The L40S has more encoders but is overqualified unless the same card also runs AI or rendering.
Are the L4 and the L40S in stock?
26 of 60 L4 units were free at the last stock update; 17 of 40 L40S 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 L4 or the L40S without KYC?
Yes. Both are rented with an email address and paid from a prepaid balance in Bitcoin, Ether or USDT — $169 and $429 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 L4 at $169 or the L40S at $429 a month, paid in BTC, ETH or USDT with no KYC.