PerGB

NVIDIA L40

Live eBay prices, cheapest first. The best one right now is $6,200 (used).

Specs

VRAM
48 GB GDDR6
Bandwidth
864 GB/s
FP32
90.5 TFLOPS
FP16
181 TFLOPS
Power
300 W
Connector
PCIe 4.0
Released
2022
Launch price
$7,500 (est.)
CUDA compute
8.9
Llama 3.1 8B at Q4 (est.)
89–134 tokens/s

Running it at home

Cooling
Passive: it has no fan. In a desktop it needs air pushed through it, front to back: a fan shroud on its end (3D-printed ones with a 40 mm blower or a 92 mm fan are sold for most of these cards) or a server chassis. Without one it overheats within minutes under load.
Power
One 16-pin (12VHPWR / 12V-2x6) power input. Rated 300 W: allow that for each card on top of the rest of the PC.
Size
Full height, two slots.
Motherboard
Turn on "Above 4G decoding" in the BIOS: without it many boards won't start with a card with this much memory.

A server card: made for a rack server's airflow and power. With these sorted it works in a normal PC.

Price right now

Live on eBay
Lowest now$6,200
Median ask$11,500
Per GB VRAM$129.17
vs launch price17% under
Live listings32

From the live listings below; shipping included.

Listings

Live on eBay
ListingConditionPriceChecked
Used$6,200or best offerfree shipping
Used$6,799or best offer−41%free shipping
Used$6,799or best offer−41%free shipping
Used$6,799or best offer−41%free shipping
Open box$7,450−35%free shipping
Open box$7,499or best offer−35%free shipping
New$7,698or best offer−33%free shipping
New$8,087−30%+$378.42 shipping
Used$8,500or best offer−26%free shipping
Open box$8,644or best offer−25%free shipping
Open box$8,647or best offer−25%free shipping
New$8,719or best offer−24%free shipping
Open box$9,000or best offer−22%free shipping
New$9,010−22%+$15.00 shipping
Open box$11,034or best offerfree shipping
New$11,200or best offerfree shipping
New$11,800or best offerfree shipping
New$11,800or best offerfree shipping
New$11,800or best offerfree shipping
New$11,800or best offerfree shipping
New$11,800or best offerfree shipping
New$11,800or best offerfree shipping
New$11,800or best offerfree shipping
New$11,800or best offerfree shipping
New$12,100or best offerfree shipping
New$12,100or best offerfree shipping
New$12,100or best offerfree shipping
New$12,100or best offerfree shipping
New$12,345or best offerfree shipping
New$12,395or best offerfree shipping
New$12,449or best offer+$50.00 shipping
New$13,205or best offerfree shipping

Auctions ending soon

No auctions ending soon right now.

Prices include shipping and were read from eBay at the time shown. They can change: check the listing before you buy.

Owner reviews

No owner reviews yet. Own one? Tell others how it runs.

From signed-in owners, checked before they're published. PerGB doesn't pay for or edit reviews.

How it compares

Live on eBay

As much VRAM, about as fast, for less

Faster, for the least

More VRAM, for not much more

From the live listings of every GPU, shipping included; performance against the GeForce RTX 3090 (FP16 or FP32 compute and memory bandwidth). Updated with every build.

Local AI: what it runs

Model (Q4, 8K context)L40 neededCosts nowSpeed (est.)
Llama 3.1 8B1 card$6,20089–134 tokens/s
Qwen3 14B1 card$6,20048–72 tokens/s
gpt-oss 20B1 card$6,200119–178 tokens/s
Gemma 3 27B1 card$6,20026–39 tokens/s
Qwen3 32B1 card$6,20022–33 tokens/s
Llama 3.3 70B1 card$6,20010–15 tokens/s
gpt-oss 120B2 cards$12,99984–126 tokens/s
Qwen3 235B-A22B4 cards$26,59719–29 tokens/s
DeepSeek R1 / V3 671Bmore than 8——

Cost: that many of the cheapest live listings. Speed: an estimate from the memory bandwidth for one person chatting. Try other models, sizes and contexts.

Decision models

Model (Q4, 8K context)L40 neededCosts nowPer question (est.)
Julia-10.144B1 card$6,2005.0–11 ms
Laya0.421B1 card$6,2008.4–18 ms
Kev-4B4B1 card$6,20020–43 ms
lev4B1 card$6,20060–129 ms
OpenJev27B1 card$6,20072–154 ms
Clef27B1 card$6,20072–154 ms
Perplexity Decider v1.227B1 card$6,20072–154 ms
mammoth-127B1 card$6,20072–154 ms

Decision models score the options you give them instead of writing text, so they're timed per question. The time is scaled from each model's reported speed on an RTX PRO 6000 by this card's compute and memory bandwidth, within about half either way. About the models.

Compare elsewhere

Search other shops for NVIDIA L40. These open the shop's own search; we don't show their prices here.