NVIDIA L40
Compare NVIDIA L40 against another machine in the Arena
Reference: Ada Lovelace (microarchitecture) on Wikipedia
- Class
- GPU
- Memory
- 48 GB GDDR6
- Runs (Q4_K_M)
- —
- Bandwidth
- 864 GB/s
- TDP
- 300 W
- Released
- 2022-10-01
- Price (US)
- $6499 used · $7990 new as of 2026-08
- Price (Canada)
- CA$10766 new sourced, as of 2026-08
Sourcing and disambiguation notes
Ada Lovelace generation datacenter card, adjacent to (but distinct from) the workstation RTX 6000 Ada — same 48GB/GDDR6 class but no ECC and typically passive/blower server cooling rather than a workstation shroud. Included here because it is the best-sourced used Ada-class datacenter card found this session; a genuine RTX 6000 Ada used-price/benchmark pair could not be sourced (see gaps file). New-old-stock 'new' figure of $7,990 (Newegg) is now available via the same gpudojo.com tracker cited above (used $6,499 via eBay). Canadian price (2026-08, pre-tax): CA$10,766.00 for the PNY NVIDIA L40 (48GB GDDR6 ECC, passive, dual-slot -- the plain L40, not the L40S). Ratio to the US $7,990 is 1.35x, a normal band.
Specification sources
- https://gpudojo.com/l40 — press, 2026-08-22
- https://cputronic.com/en/gpu/nvidia-l40 — press, 2026-08-22
Measurements
Decode is token generation — the speed you feel while an answer streams. Prefill is prompt processing — the wait before it starts. Why bandwidth predicts decode speed.
No records yet for this hardware. Know of a published benchmark on it? Submit the link.
What verified, single-source and estimated mean, and the same rows with every filter and sort in the benchmarks explorer.
Speed over time
What it can run
Fit is arithmetic, not a measurement — how it is computed. Against 48 GB; models that fit are listed largest first.
No modeled quant fits in 48 GB.