ThinkSystem NVIDIA L40 48GB PCIe Gen4 Passive GPU
Overview
GPU Architecture: NVIDIA Ada Lovelace
NVIDIA CUDA Parallel Processing Cores: 18,176
NVIDIA Tensor Cores (4th gen): 568
NVIDIA RT Cores (3rd Gen): 142
Peak FP32 performance (non-Tensor): 90.5 TFLOPS
Peak FP16 Tensor performance with FP16 Accumulate: 181.05 TFLOPS, 362.1 TFLOPS*
Peak Tensor Float 32 (TF32) performance: 90.5 TFLOPS, 181 TFLOPS*
Peak Bfloat16 (BF16) performance with FP32 Accumulate: 181.05 TFLOPS, 362.1 TFLOPS*
Peak FP8 Tensor performance: 362 TFLOPS, 724 TFLOPS*
Peak Integer Performance: INT8: 362 TOPS, 724 TOPS*, INT4: 724 TOPS, 1448 TOPS*
RT Core performance: 209 TFLOPS
GPU Memory: 48 GB GDDR6
Memory Bandwidth: 864 GB/s
ECC: Yes
NVIDIA NVLink: No support
System Interface: PCIe Gen 4, x16 lanes
Form Factor: PCIe full height/length, double width (10.5" x 4.4")
Multi-Instance GPU (MIG): No support
Max Power Consumption: 300 W
Thermal Solution: Passive
vGPU Software Support: NVIDIA vPC/vApps, NVIDIA RTX Virtual Workstation (vWS)
Display connectors: 4x DisplayPort 1.4a (disabled by default**)
Max Simultaneous Displays: Up to four 5K Monitors at 60Hz per card or dual 8K displays @ 60Hz (requires displayPort 1.4 DSC); Each display port can support 4K at 120 Hz with 30-bit color
Graphics APIs: DirectX 12 Ultimate, Shader Model 6.6, OpenGL 4.6, Vulkan 1.3
Compute APIs: CUDA 12.0, Direct Compute, OpenCL 3.0
