CARDINAL
EXACT VARIANT REFERENCE

L40S PCIE 48GB

A 48 GB PCIe accelerator with graphics and video capabilities

SPECIFICATIONS

Memory per accelerator48 GB GDDR6 ECC
Memory bandwidth864 GB/s
Form factorDual-slot PCIe card
Maximum accelerator power350 W

Manufacturer specifications describe the product, not the condition of an offered unit. Checked 2026-09-08

COMPUTE SPECIFICATIONS

Manufacturer peak arithmetic per accelerator. These values do not predict application throughput or latency.

PrecisionExecutionPeak rateSparsityReference
FP32Vector91.6 TFLOPSDenseSource ↗
Conditions

Manufacturer peak per GPU; not sustained workload performance

TF32Tensor183 TFLOPSDenseSource ↗
Conditions

Manufacturer peak per GPU; not sustained workload performance

TF32Tensor366 TFLOPSStructured sparseSource ↗
Conditions

Manufacturer peak per GPU; not sustained workload performance

FP8Tensor733 TFLOPSDenseSource ↗
Conditions

Manufacturer peak per GPU; not sustained workload performance

FP8Tensor1466 TFLOPSStructured sparseSource ↗
Conditions

Manufacturer peak per GPU; not sustained workload performance

INT8Tensor733 TOPSDenseSource ↗
Conditions

Manufacturer peak per GPU; not sustained workload performance

INT8Tensor1466 TOPSStructured sparseSource ↗
Conditions

Manufacturer peak per GPU; not sustained workload performance

INT4Tensor733 TOPSDenseSource ↗
Conditions

Manufacturer peak per GPU; not sustained workload performance

INT4Tensor1466 TOPSStructured sparseSource ↗
Conditions

Manufacturer peak per GPU; not sustained workload performance

BF16Tensor733 TFLOPSStructured sparseSource ↗
Conditions

Manufacturer peak per GPU; not sustained workload performance

FP16Tensor733 TFLOPSStructured sparseSource ↗
Conditions

Manufacturer peak per GPU; not sustained workload performance

  • Peak arithmetic rates are not token throughput, latency or a guarantee of application performance
  • Structured-sparse rates require eligible sparsity and supported kernels; do not compare them directly with dense rates
  • Dense BF16/FP16 omitted: the checked webpage lists 362.05 alongside 733 sparse, an apparent inconsistency requiring clarification; no replacement value calculated
  • RT-core performance is excluded because it is neither vector nor tensor execution in this schema
  • Use the detailed specification table for precision and sparsity, not a generic AI performance headline

PLATFORM AND SOFTWARE

  • No NVLink support
  • No Multi-Instance GPU support
  • Passive cooling requires server airflow
  • Confirm application support for CUDA or the intended graphics stack
  • Confirm separate vGPU licensing when virtualization is required

Manufacturer maximum accelerator rating; actual draw and supported limits depend on system

INTERCONNECT

PCIe Gen4 x16

64 GB/s. Theoretical aggregate bidirectional interface bandwidth

MANUFACTURER SOURCES

NVIDIA L40S specifications ↗NVIDIA · Checked 2026-09-08

HOW CARDINAL HELPS

Share your quantity, destination and server model, if known. We help turn those requirements into a clearly specified request.

  • Help confirm the exact configuration and platform requirements
  • Request supplier records for condition, unit identifiers and available testing evidence
  • Clarify included accessories, delivery and proposed warranty terms before you decide

Any additional testing or guarantees must be agreed in the written offer.