What LLMs can an L40S run?

48 GB VRAM 864 GB/s 350 W 2023 Rented, not bought

L40S 48 GB addresses 48 GB at 864 GB/s. Of the 16 open models tracked on this site, it runs 13 at Q4_K_M with 8k of context. The largest is Qwen3 32B, needing about 22.4 GB and generating an estimated 32 tokens per second — comfortable for chat.

The 48 GB card you meet as a rental line item, not as a purchase.

What an L40S runs, model by model

Every model at Q4_K_M with 8k of context, the setting most people actually use. This card has 45.6 GB to spend once the 5% safety margin comes off its 48 GB, and it reads that memory at 864 GB/s. Those two numbers decide the last two columns: the first says what loads, the second says how fast it answers. Both are estimates from stated formulas, and the VRAM calculator shows the working.

On this card that means anything at or under 45.6 GB counts as fitting, and anything above 40.8 GB is marked tight: it loads, but there is little left for the conversation, and a desktop on the same 48 GB can push it over.

The model to actually run on it

Qwen3 32B is the best use of this card: 22.4 GB of the 48 GB available, an estimated 32 tokens per second, comfortable for chat, and 96,256 tokens of context still available.

It also fits at Q8_0, in about 38.8 GB, which is worth taking whenever the memory allows: Q8 is near-lossless where Q4 costs a little accuracy.

How much context actually fits

Model size is the question people ask; context is the one that bites. The KV cache grows linearly with the conversation, so a model that loads comfortably can still run out of memory halfway through a long document. On 48 GB, the context you get is whatever the weights leave behind — which is why two cards that run the same model can feel completely different in use. These are the largest contexts this card holds, at Q4_K_M weights.

Quantising the KV cache to Q8 roughly doubles what you can hold — Qwen3 32B on this card goes from 96,256 to 131,072 tokens — at a quality cost most people never notice. Capped at 128k tokens here; a model's own architecture may stop lower, and sliding-window models such as Gemma 3 use less KV than the formula assumes.

Where this card stops

The first model out of reach is Llama 3.3 70B: about 45.8 GB at Q4_K_M and 8k context, against 48 GB of usable memory. That is close enough to be worth a caveat: it is inside the raw 48 GB and only misses the 95% line this site uses for a fit. On a card with a display attached, treat it as no. On a headless card, with a shorter context, it runs. Dropping to Q3_K_M would need about 37.7 GB, which fits, though Q3 loses enough quality that a smaller model at Q4 is usually the better trade. Offloading the remainder to system RAM works and is ten to fifty times slower; it is a way to see a model run, not a way to use one.

The honest take

The L40S is the same Ada silicon family as the RTX 6000 Ada, packaged for servers: passively cooled, 350 W, no display outputs. On paper it is a slightly slower 48 GB card at 864 GB/s. In practice it is the card most cloud providers hand you when you ask for 48 GB, so its real relevance is as a price and speed reference for renting rather than a thing you install.

What it is good at

• 48 GB comfortably holds 32B models with very long context, and brings 70B at Q4 to the edge of what fits.
• Widely stocked by providers, so it is usually available without queueing.
• Good for image and video generation, where its Ada compute is put to use.

What it is not

• 864 GB/s is the lowest bandwidth of the 48 GB options here.
• Passive cooling means it only works in a chassis with forced airflow — it cannot go in a desktop.
• No display output, so it is an accelerator only.

The thing people get wrong: It has no fan. The card depends entirely on server chassis airflow, which is why buying one second-hand for a tower build ends badly.

Buy or rent

There is nothing to decide here: you rent it. Use it as the reference when comparing a 48 GB rental against buying an RTX 6000 Ada.

There is no purchase price to reason about, only an hourly one, and it differs by provider and by week. Put the figure you are actually looking at, with your electricity rate and your real monthly hours, into the build vs rent calculator — it already knows this card draws 350 W under load. It returns the month owning becomes cheaper, or tells you it never does. Hours per month decides it far more often than the price does.

Nodegrove is the rented side of that comparison: a workspace that stays saved, with a GPU attached only while you are using it. It is the right answer when your hours are low or your needs change; buying is the right answer when they are high and stable.

Compared with the alternatives

• RTX 6000 Ada — 48 GB · 960 GB/s · best fit Qwen3 32B at ~35 tok/s. Forty-eight gigabytes in a normal computer, at 300 watts, without the noise.
• A100 — 80 GB · 2,039 GB/s · best fit Llama 3.3 70B at ~35 tok/s. The old datacenter workhorse, still fast where it counts, and cheap to rent.
• RTX PRO 6000 Blackwell — 96 GB · 1,792 GB/s · best fit gpt-oss 120B at ~424 tok/s. Ninety-six gigabytes and 5090-class bandwidth on one card.

All 13 cards are compared side by side on the GPU index, and the same numbers from the model's point of view are on each model fit table. To check one specific pairing rather than read a table, use the can-I-run-it checker.

Questions

Can an L40S run a 70B model?

Not at Q4_K_M in one card. Llama 3.3 70B needs about 45.8 GB at 8k context and this card can address 48 GB. Your options are a smaller model, a harsher quantisation with very little context, splitting across two cards, or renting a larger one for the hours you need it.

What is the best model to run on an L40S?

Qwen3 32B. At Q4_K_M and 8k context it needs about 22.4 GB of the 48 GB available and generates an estimated 32 tokens per second, which is comfortable for chat. It also fits at Q8, at about 38.8 GB, which is worth taking when it fits.

How many tokens per second does an L40S generate?

It depends entirely on the size of the model, because generating a token means reading every active weight out of memory. At 864 GB/s and 70% efficiency, this card produces an estimated 130 tokens per second on an 8B model at Q4, and about 32 on the largest model it holds, Qwen3 32B. Reading speed is roughly 5 to 8 words per second, so anything above 25 feels immediate.

Is 48 GB enough for running LLMs locally?

It runs 13 of the 16 models tracked here at Q4_K_M with 8k of context, up to 32.8B parameters. The honest test is not the model list but the context: Qwen3 32B on this card holds about 96,256 tokens before memory runs out.

L40S or RTX 6000 Ada for local models?

Both address about the same memory, so they run the same models. On speed, the RTX 6000 Ada is faster: 960 GB/s against 864 GB/s, and bandwidth is what sets chat speed.

Can I put an L40S in a desktop PC?

No. It is passively cooled — there is no fan on the card at all — and it depends on the forced airflow of a server chassis. Installed in a tower it will overheat. The desktop equivalent of this silicon is the RTX 6000 Ada.

L40S or A100 for inference?

The A100 has more memory and much more bandwidth, 80 GB at 2,039 GB/s against 48 GB at 864 GB/s, so it generates faster and holds larger models. The L40S has newer Ada compute, which helps with image and video generation and with prompt processing. For chatting with a large language model, the A100 is the better rental.

• Memory: weights (parameters × bytes per parameter) + KV cache (2 × layers × KV heads × head dim × context × bytes) + 0.5 GB runtime + 4% of weights. Full derivation in the VRAM calculator.
• Speed: 70% × bandwidth ÷ active weight bytes, single stream, no batching, prompt processing excluded. See the tokens-per-second estimator.
• Card specification: 48 GB, 864 GB/s, 350 W — manufacturer figures.
• Model architecture values come from each model's published config; the self-hosted LLM guide explains what each one changes.