NVIDIA · Data Center

NVIDIA A40

The NVIDIA A40 is a graphics card released on October 5, 2020, built on the Ampere architecture. It has a boost clock of 1.74 GHz. It scores 741 in the PassMark G3D benchmark. It supports CUDA compute capability 8.6.

Ampere

5/10 key specs recorded · 30 facts

Depth: 19 recorded fields, 10 features, 5 PassMark workloads

Specification data last verified October 1, 2026 against PassMark Videocard Benchmark, NVIDIA Developer: CUDA GPUs, NVIDIA Developer: Video Encode and Decode GPU Support Matrix and Wikipedia GPU list tables (wikigrid-2026-09). Coverage: 5 of 10 tracked specifications recorded.

PassMark Videocard Benchmark NVIDIA Developer: CUDA GPUs NVIDIA Developer: Video Encode and Decode GPU Support Matrix Wikipedia GPU list tables (wikigrid-2026-09)PassMark Videocard Benchmark

Shading units

n/a

84 SMs

Memory

n/a

GDDR6

Boost clock

1.74 GHz

Board power

n/a

Performance

The NVIDIA A40 scores 741 in the PassMark G3D benchmark, higher than 47.1% of the 2,658 GPUs benchmarked (PassMark rank #550), measured across 6 submitted results. Its PassMark G2D score is 23.

PassMark G3D Mark

741

Rank #550
Faster than 47.1% of 2,658 GPUs
6 samples

0relative to the fastest GPU on record (41,928)

PassMark workload tests

DirectX 9 58 fps

reference: fastest recorded 526 fps · faster than 61.6% of 1,675 GPUs

DirectX 10 70 fps

reference: fastest recorded 231 fps · faster than 84% of 1,602 GPUs

DirectX 11 79 fps

reference: fastest recorded 362 fps · faster than 80.5% of 1,430 GPUs

DirectX 12 37 fps

reference: fastest recorded 213 fps · faster than 72.2% of 1,274 GPUs

GPU Compute 2,353 ops/s

reference: fastest recorded 26,591 ops/s · faster than 67.6% of 1,447 GPUs

Averages of PassMark PerformanceTest submissions; frames/sec are PassMark test-scene rates, not game FPS. Tests PassMark does not report for this card are omitted. PassMark listing

In the PassMark workload tests the NVIDIA A40 averages DirectX 9 58 fps, DirectX 10 70 fps, DirectX 11 79 fps, DirectX 12 37 fps and GPU Compute 2,353 ops/s, across 6 submitted results. Frames per second are PassMark test-scene rates, not game frame rates.

As stated by PassMark

Category Desktop

First benchmarked 2025-11-12

Samples 6

PassMark listing (collected 2026-10-01)

Not to be confused with

This page covers the NVIDIA A40 specifically. It is a different product from NVIDIA A40-12Q, NVIDIA A40-24Q, NVIDIA A40-2Q, NVIDIA A40-3Q, NVIDIA A40-48Q, NVIDIA A40-4Q, which have their own specifications. Check the exact model name before comparing figures.

Graphics processor

Architecture Ampere

Codename Not recorded

Chip Not recorded

Process node Not recorded

Transistors Not recorded

Die size Not recorded

The NVIDIA A40 is based on the Ampere architecture.

Shader configuration

Shading units Not recorded

Compute units Not recorded

Streaming multiprocessors 84

Execution units Not recorded

Texture units (TMUs) Not recorded

Render output units (ROPs) Not recorded

Ray tracing cores Not recorded

Tensor / AI cores Not recorded

XMX engines Not recorded

Infinity cache Not recorded

The NVIDIA A40 84 streaming multiprocessors.

The NVIDIA A40 does not use XMX engines; those are specific to Intel Arc graphics. Infinity Cache is an AMD architectural feature and does not apply to the NVIDIA A40.

Clock speeds

Base clock Not recorded

Boost clock 1.74 GHz

Game clock Not recorded

Memory clock Not recorded

Effective memory clock Not recorded

The NVIDIA A40 boosts to 1.74 GHz.

Memory

Size Not recorded

Type GDDR6

Bus width Not recorded

Bandwidth Not recorded

Market

Launched October 5, 2020

Launch price Not recorded

Status Not recorded

Form factor Not recorded

Segment Desktop

Series Data Center

The NVIDIA A40 was released on October 5, 2020.

Benchmarks

PassMark G3D 741

PassMark G2D 23

Rank #550

Samples 6

Graphics and compute APIs

The NVIDIA A40 supports CUDA compute capability 8.6.

CUDA compute capability 8.6

Feex has no recorded Vulkan version for the NVIDIA A40. Feex has no recorded OpenCL version for the NVIDIA A40.

Features

The NVIDIA A40 has 1 supported feature recorded outside the media engine. A feature that no source mentions is shown as not listed, never as unsupported.

Compute and AI

Supported: CUDA Compute capability 8.6

Support for NVIDIA's CUDA computing platform, used by many creative, scientific and AI applications.

Not listed (2)

No source Feex holds states whether the NVIDIA A40 supports Tensor cores and NVLink / SLI. That is silence, not a statement that they are unsupported.

The NVIDIA A40 supports CUDA (Compute capability 8.6) under Compute and AI.

Media engine

The NVIDIA A40 provides hardware encode for H.264 (AVC) and HEVC (H.265) and hardware decode for H.264 (AVC), HEVC (H.265), AV1 and VP9. It does not support hardware encode for AV1.

NVIDIA NVENC encoder 7th Generation Encoder engines: 1NVIDIA's dedicated video-encoding engine, used by streaming and capture software; newer generations encode more efficiently.

NVIDIA NVDEC decoder 5th Generation Decoder engines: 2NVIDIA's dedicated video-decoding engine; newer generations support more codecs and formats.

Hardware video codec support of the NVIDIA A40
Codec	Encode	Decode
H.264 (AVC)	Supported4:4:4: Yes	Supported
HEVC (H.265)	Supported4:2:2: No; 4:4:4: Yes; 10-bit: Yes; B-frames: Yes; 8K: Yes	Supported4:4:4: Yes
AV1	Not supported	Supported
VP9	Not stated by the vendor	Supported

supported explicitly not supported not stated by the vendor

Media engines recorded for the NVIDIA A40: NVIDIA NVENC encoder (7th Generation, encoder engines: 1) and NVIDIA NVDEC decoder (5th Generation, decoder engines: 2).

Frequently asked questions

How fast is the NVIDIA A40?

The NVIDIA A40 scores 741 in the PassMark G3D benchmark, higher than 47.1% of the 2,658 GPUs benchmarked. Its PassMark rank is #550. In the PassMark DirectX 12 test it averages 37 fps.