GeekRadar / The comparison desk

COMPARE.

GPU

NVIDIA

NVIDIA L40

Graphics processor

Video memory: 48 GB^*
Memory bus: 384 bit^*
Rated memory bandwidth: 864 GB/s^*

Benchmark results

Geekbench GPU (reported compilation) · version unreported · OpenCL329,891^*Platform unreported Compare results
Geekbench GPU (reported compilation) · version unreported · Vulkan246,199^*Platform unreported Compare results

Processing & platform

Architecture: Ada Lovelace^*
Process node: 5 nm^*
GPU cores: 18,176^*
Graphics interface: PCIe 4.0 x16^*
Reported chip die area: 609 mm²^*
Transistors: 76,300 million^*
GPU L2 cache: 96 MB^*
Graphics processor: AD102^*
Texture mapping units: 568^*
Raster output units: 192^*
Streaming multiprocessors: 142^*
Tensor cores: 568^*
Ray tracing cores: 142^*
Chip foundry: TSMC^*
Card or module format: Dual-slot^*
Declared GPU TDP: 300 W^*
Suggested system PSU: 700 W^*
External power connectors: 1x 16-pin^*
DirectX / feature level: 12 Ultimate (12_2)^*
OpenGL: 4.6^*
OpenCL: 3.0^*
Vulkan: 1.4^*
CUDA compute capability: 8.9^*
Shader model: 6.9^*

Memory & storage

Memory options: 48^*
Video memory: 48 GB^*
Video memory type: GDDR6^*
Memory bus: 384 bit^*
Rated memory bandwidth: 864 GB/s^*

Power specifications

Thermal design power: 300 W^*

System configuration

Detailed specifications

Specifications 1 · 17 fields

GPU architecture — NVIDIA L40 — 48 GB GDDR6: Ada Lovelace^*
Theoretical peak: BF16 tensor dense per GPU — NVIDIA L40 — 48 GB GDDR6: 181.05^* TFLOPS
Theoretical peak: FP16 tensor dense per GPU — NVIDIA L40 — 48 GB GDDR6: 181.05^* TFLOPS
Theoretical peak: FP32 non-tensor dense per GPU — NVIDIA L40 — 48 GB GDDR6: 90.5^* TFLOPS
Native hardware support: FP4 tensor — NVIDIA L40 — 48 GB GDDR6: No^*
Native hardware support: FP6 tensor — NVIDIA L40 — 48 GB GDDR6: No^*
Theoretical peak: FP8 tensor dense per GPU — NVIDIA L40 — 48 GB GDDR6: 362^* TFLOPS
Theoretical peak: INT8 tensor dense per GPU — NVIDIA L40 — 48 GB GDDR6: 362^* TOPS
Theoretical peak: TF32 tensor dense per GPU — NVIDIA L40 — 48 GB GDDR6: 90.5^* TFLOPS
Physical configuration — NVIDIA L40 — 48 GB GDDR6: PCIe, dual-slot FHFL^*
Native hardware support for GPU interconnect — NVIDIA L40 — 48 GB GDDR6: No^*
PCIe interface declaration — NVIDIA L40 — 48 GB GDDR6: PCIe Gen4 x16^*
Maximum board power — NVIDIA L40 — 48 GB GDDR6: 300^* W
Peak memory bandwidth per GPU — NVIDIA L40 — 48 GB GDDR6: 864^* GB/s
Memory capacity per GPU — NVIDIA L40 — 48 GB GDDR6: 48^* GB
Memory technology — NVIDIA L40 — 48 GB GDDR6: GDDR6^*
GPU model — NVIDIA L40 — 48 GB GDDR6: NVIDIA L40^*

Specifications 2 · 31 fields

proc transistors: 76,300 million^*
proc die size: 609^* mm²
cores: 18,176^*
tensor cores: 568^*
rt cores: 142^*
sm count: 142^*
base clock: 735^*
boost clock: 2,490^*
memory clock: 2,250^*
ram: 48^*
name: NVIDIA L40^*
ram type: GDDR6^*
bus width: 384^*
bus interface: PCIe 4.0 x16^*
ram bandwidth: 864^* GB/s
fp16 performance: 90.52^*
fp32 performance: 90.52^*
fp64 performance: 1.4143^*
tdp: 300^* W
suggested psu: 700^* W
slot width: Dual-slot^*
power connectors: 1x 16-pin^*
api cuda: 8.9^*
is retail board: No^*
gpi value: 36.27^*
vendor: nvd^*
manufacturer: NVIDIA^*
arch name: Ada Lovelace^*
card release date: 2022-10-13^*
proc foundry: TSMC^*
proc process size: 5^* nm

Specifications 3 · 40 fields

CUDA compute capability

8.9^*

DirectX / feature level

12 Ultimate (12_2)^*

OpenCL

3.0^*

OpenGL

4.6^*

Shader model

6.9^*

Vulkan

1.4^*

Architecture

Ada Lovelace^*

Graphics interface

PCIe 4.0 x16^*

Memory bus

384^* bit

Reported shader processors

18,176^*

Reported theoretical FP16 peak

90.52^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Reported theoretical FP32 peak

90.52^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Reported theoretical FP64 peak

1.4143^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Declared L1 cache configuration

128 KB (per SM)^*

Declared GPU L2 cache

96 MB^*

Reported display outputs

4x DisplayPort 1.4a^*

Theoretical pixel fill rate

478.1 GPixel/s^*

External power connectors

1x 16-pin^*

Reported chip die area

609 mm²^*

Chip foundry

TSMC^*

Process node declaration

5 nm^*

Transistors

76,300 million^*

Graphics processor

AD102^*

Reported video memory

48^* GB

Rated memory bandwidth

864.0 GB/s^*

Video memory type

GDDR6^*

Raster output units

192^*

Ray tracing cores

142^*

Reported shading units

18,176^*

Card or module format

Dual-slot^*

Streaming multiprocessors

142^*

Suggested system PSU

700 W^*

2:4 tensor sparsity

Yes^*

Declared GPU TDP

300 W^*

Tensor cores

568^*

Theoretical texture fill rate

1,414.3 GTexel/s^*

Texture mapping units

568^*

GPU cores

18,176^*

Video memory

48^* GB

Video memory type

GDDR6^*

Specifications 4 · 3 fields

Device

NVIDIA L40^*

Scope and conditions

benchmark_context

suite: Geekbench GPU (reported compilation)^*
backend_columns: CUDA^*
Metal^*
OpenCL^*
Vulkan^*
test_configuration_verified: No^*
comparable: No^*

OpenCL

329,891^*

Scope and conditions

benchmark_context

suite: Geekbench GPU (reported compilation)^*
backend_columns: CUDA^*
Metal^*
OpenCL^*
Vulkan^*
test_configuration_verified: No^*
comparable: No^*

Vulkan

246,199^*

Scope and conditions

benchmark_context

suite: Geekbench GPU (reported compilation)^*
backend_columns: CUDA^*
Metal^*
OpenCL^*
Vulkan^*
test_configuration_verified: No^*
comparable: No^*

* Provisional data