L40S specs
NVIDIA · Shipping now · first available Oct 2023 (estimated). Dense compute, memory, bandwidth and power from the vendor's documents; each precision stays separate.
All accelerators · Compare with L4
Specifications
| Dense BF16 · FP32 accumulate | 362.05 TFLOPS |
|---|---|
| Dense FP8 · FP32 accumulate | 733 TFLOPS |
| Dense FP4 · FP32 accumulate | Unsupported |
| Memory | 48 GB · GDDR |
| Memory bandwidth | 0.864 TB/s |
| Maximum chip power | 350 W |
| Interconnect | PCIe 4.0 |
| First availability | Oct 2023est — estimated / secondary-source figure |
| Availability | Shipping |
Ada Lovelace data-center card with 48 GB of GDDR6, not HBM, and no NVLink. Launch month is an estimate: NVIDIA said in August 2023 that it would be available from that fall.
Dense tensor rates with FP32 accumulation, never sparse or FP16-accumulate headline figures. Memory is in decimal GB. The est marker flags a value its vendor-document review did not confirm.
Rent it
$1.55/GPU-hrCheapest on-demand · Nebius · observed
4 providers: AWS, Nebius, Oracle OCI, Verda. Quotes checked through .
L40S rental prices from every provider →Value ratios
| $ per dense BF16 PFLOP-hour | $4.28 |
|---|---|
| Dense BF16 TFLOPS per watt | 1.03 |
| Bandwidth bytes per 1,000 BF16 FLOP | 2.39 |
| $ per dense FP8 PFLOP-hour | $2.11 |
| Dense FP8 TFLOPS per watt | 2.09 |
| Bandwidth bytes per 1,000 FP8 FLOP | 1.18 |
| $ per GB of memory per hour | $0.032 |
Prices divide the cheapest on-demand list price; theoretical ratios do not measure application speed.
Sources
Specifications reviewed . Dense tensor rates are the unstarred figures (BF16 362.05, FP8 733 TFLOPS); the starred sparse ones are excluded, and no FP4 format is listed. 48 GB GDDR6 at 864 GB/s, 350 W, PCIe Gen4 x16 without NVLink. The page states no process node.
Rental price: Nebius published pricing ↗ (L40S with Intel CPU; 8–40 vCPUs and 32–160GB RAM range).