NVIDIA L40S 48GB
NVIDIA · Ada Lovelace · 48GB GDDR6 · PCIe · PCIe 4.0
Current asking prices
Median asking price
$7,500
Observed range
$1,559–$8,000
24 listings· 9 sellers· Updated Aug 25, 2026
Market signals
Derived from observed listings across all SKUs in this class.
▶What do OEM / Original / Compatible mean?
OEMCertified by a server vendor (Dell, HPE, Lenovo). May carry firmware locks and qualifications. Priced higher for vendor support and warranty alignment.
OriginalThe underlying component manufacturer's part, without OEM server-vendor branding or firmware lock. Typically the market pricing reference — platform compatibility should still be verified.
CompatibleThird-party alternatives meeting the same spec. Often 15–40% cheaper. May trigger vendor warnings on OEM-locked platforms — verify before substituting.
OEM Premium
+3%
OEM parts cost ~3% more than original manufacturer parts.
Market Spread
135%
Very wide spread. High volatility — verify condition and seller carefully.
Liquidity
Good
24 samples. Solid basis. Ranges reflect real market activity.
▶What do the signal colors mean?
Favorable — Tight spread, meaningful discount, or strong liquidity.
Neutral — Normal market conditions. No action flag.
Investigate — Wide spread, inverted pricing, or low sample count.
High alert — Very wide spread or significant price anomaly.
Insufficient — Too few samples to produce a reliable signal.
Part numbers in this market
6 part numbers tracked. Exact platform, firmware, power, cooling, and OEM compatibility should be verified before purchase.
| Manufacturer | MPN | Certification |
|---|---|---|
| NVIDIA | 490-BJEX | oem |
| NVIDIA | L40S-48GB | original |
| NVIDIA | 900-2G133-0020-000 | original |
| NVIDIA | S0K90C | oem |
| NVIDIA | 4X67A84829 | oem |
| NVIDIA | GPU-NVL40S | oem |
Technical specifications
NVIDIA L40S PCIe datacenter GPU with 48GB GDDR6; ai inference workloads
Manufacturer
NVIDIA
Model family
L40S
Architecture
Ada Lovelace
Form factor
PCIe
Memory
48GB
Memory type
GDDR6
Memory bandwidth
0.864 TB/s
TDP
350W
Interconnect
PCIe
PCIe generation
4
FP16 tensor
362 TFLOPS
FP8 tensor
733 TFLOPS
Workload class
ai_inference
Launch year
2023
Signals are observations from available data at a point in time — not advice. Verify independently before transacting. Disclaimer