Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →The NVIDIA A40 is a 48 GB passive-cooled Ampere accelerator built for data-center visualization, virtual workstations and compute—not a drop-in gaming GPU. Its strongest case is combining substantial ECC memory with NVIDIA vGPU support in a server that can deliver the required airflow and power. Its 300 W card limit, lack of MIG and configuration-dependent virtualization support matter as much as its headline specifications.
What the NVIDIA A40 is designed for
NVIDIA positions the A40 for professional visualization and data-center workloads. In practice, its 48 GB of ECC memory and support for NVIDIA vGPU software make it relevant to virtual desktop infrastructure (VDI), virtual workstations and mixed workstation-and-compute deployments. It is a dual-slot PCIe card with passive cooling, so it belongs in a compatible server or workstation chassis with directed airflow—not an ordinary desktop case chosen only for card clearance.
As an Amazon Associate I earn from qualifying purchases.
ServeTheHome published its two-page mini-review on March 18, 2022. The review’s main argument is that the A40’s appeal is not necessarily maximum training speed: the card can serve virtual desktop users and potentially run compute work when those users are offline. That is a workload and deployment rationale, not a current ranking of all available GPUs. Read ServeTheHome’s A40 hardware overview.
NVIDIA A40 specifications
| Specification | NVIDIA-listed value |
|---|---|
| GPU architecture | Ampere |
| Memory | 48 GB GDDR6 with ECC |
| Memory bandwidth | 696 GB/s |
| CUDA cores | 10,752 |
| RT cores | 84 second-generation |
| Tensor cores | 336 third-generation |
| Peak FP32 performance | 37.4 TFLOPS, non-Tensor |
| Peak FP16 Tensor performance | 149.7 TFLOPS; 299.4 TFLOPS with structural sparsity enabled |
| NVLink bandwidth | 112.5 GB/s bidirectional |
| PCIe bandwidth | 64 GB/s, PCIe Gen4 |
| Form factor and dimensions | Dual-slot; 4.4 in (H) × 10.5 in (L) |
| Cooling | Passive thermal solution |
| Maximum card power | 300 W |
| Power connector | 8-pin CPU-style connector |
| Display outputs | Three DisplayPort 1.4 outputs; disabled by default in the virtualization configuration |
| Video engines | 1× NVENC / 2× NVDEC; AV1 decode listed |
| MIG | Not supported |
These are manufacturer specifications from NVIDIA’s March 2022 data sheet; peak compute rates are not promises of application performance. See NVIDIA’s A40 data sheet.
#1 Best Overall
- Discrete graphics card memory 40 GB
- Memory bandwidth (max) 1555 GB/s
- Graphics processor family NVIDIA
- Graphics processor A100
Installation: airflow, power and clearance
The A40’s passive heatsink depends on chassis fans to move air through it. A card can fit physically and still be unsuitable if the host does not provide the airflow path the card needs. Before buying or installing one, check:
- Physical fit: dual-slot clearance and space for a 10.5-inch-long, 4.4-inch-high card, including any chassis restrictions around full-length cards.
- Airflow: forced airflow through the heatsink in a chassis designed for passive data-center accelerators.
- Power: a suitable 8-pin CPU-style GPU power lead and adequate capacity for the card and the rest of the system.
- Intended mode: whether the deployment needs physical display outputs, NVIDIA vGPU, or both; display behavior depends on configuration.
- Platform compatibility: confirmation from the system vendor that the chassis, power delivery and software stack support the specific card and workload.
NVIDIA rates the card at up to 300 W. ServeTheHome’s 2022 review reports that its cards at full utilization drew approximately 296–300 W within the configured power cap. Across sixteen observed A40s, the reviewer measured 25–31 W idle per card. The review also estimates that an A40 added roughly 360–400 W of system draw at the PDU in its reviewed systems, depending on PSU efficiency and cooling; server fans can materially contribute to total draw. Those are observations and estimates from the review’s systems, not guaranteed figures for another host. See ServeTheHome’s power and performance discussion.
vGPU support, profiles and limits
The A40 supports NVIDIA vGPU software for use cases including virtual PCs and apps, RTX Virtual Workstation and Virtual Compute Server. NVIDIA’s rolling RTX vWS sizing guide, accessed September 29, 2026, lists A40 as supported hardware and shows profiles ranging from 1 GB to 48 GB. Profile availability, licensing, hypervisor compatibility and supported configurations depend on the relevant vGPU software release and deployment. Check the current release documentation and licensing terms before designing a system around a particular profile.
In the guide’s described RTX vWS context, NVIDIA notes a context-switching limit of 32 users per A40 GPU. Treat that as a product-and-software-context limit to validate against the current release and license—not as a blanket promise that 32 concurrent users will receive a particular experience. The data sheet says the card is configured for virtualization by default with physical display connectors disabled; NVIDIA says management software can enable them. Consult NVIDIA’s current RTX vWS GPU sizing guide.
Rank #3
- Powered by the NVIDIA GeForce RTX 4080 (16GB) graphics processing unit (GPU) with a 2.51 GHz boost clock speed
- PCI Express 4.0 and earlier PCI Express 3.0. Offers compatibility with a range of systems
- 9,728 NVIDIA CUDA Cores, 2.51 GHz Boost Clock, Dedicated Ray Tracing Cores
- Microsoft DirectX 12 Ultimate, Vulkan RT APIs
A40 partitioning is not MIG
The A40 does not support Multi-Instance GPU (MIG), so it cannot divide one physical GPU into isolated MIG instances. NVIDIA does support connecting two A40s with NVLink; its product documentation describes scaling accessible GPU memory from 48 GB to 96 GB for a pair. That is a two-card NVLink configuration, not a way to partition one A40, and software must support the intended multi-GPU workload. Verify the bridge spacing and system compatibility for any planned pair. See NVIDIA’s A40 product information.
A40 vs. A100 for training: what the review does and does not show
ServeTheHome offers workload-specific estimates rather than a universal benchmark result. In its 2022 discussion, PCIe A100 training is characterized as roughly twice A40 performance. For smaller ResNet-50 training, the review estimates an 80 GB, 500 W SXM4 A100 at about 2.4–2.5 times A40 performance. The reviewer cautions that the gap may grow with larger models and greater use of memory and NVLink, and notes variation among cards and between larger eight- and ten-GPU systems.
Rank #4
- 16,384 NVIDIA CUDA Cores
- Supports 4K 120Hz HDR, 8K 60Hz HDR and variable refresh rate as indicated in HDMI 2.1A
- New streaming multiprocessors: up to 2x power and power efficiency
- Fourth generation tensor cores: up to 2x AI power
- Third-generation RT cores: up to 2x ray tracing performance
Use those figures only as rough context for the workloads described in that review. They are not guaranteed ratios for another model, software stack, batch size or multi-GPU configuration. For a purchase decision, compare results for the actual workload and system, then weigh memory capacity, bandwidth, interconnect, cooling, software support and current system cost. The review’s observation that A40 systems tended to cost much less than an A100 SXM4 solution reflects its 2022 market context, not a verified current price comparison. Read the review’s performance context.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Who should consider the A40?
- Potential fit: organizations building VDI or virtual-workstation systems that can use NVIDIA vGPU software and provide suitable chassis airflow and power.
- Potential fit: compute deployments that benefit from 48 GB ECC memory and can make use of the card’s performance, while accepting workload-specific results rather than assuming peak rates.
- Look elsewhere or validate carefully: deployments that require MIG partitioning, a conventional fan-cooled desktop card, or a specific performance target unsupported by workload-matched benchmarks.
- For an A40 pair: consider NVLink only if the software and workload can use the connection, and verify bridge and chassis compatibility before purchase.
The A40 can be compelling when its vGPU capabilities and large memory solve a concrete deployment need. Its passive cooling and server-oriented power and integration requirements make host compatibility central to the decision.
Quick Recap
Best Value
- NVIDIA Ada Lovelace Streaming Multiprocessors: Up to 2x performance and power efficiency
- 4th Generation Tensor Cores: Up to 2X AI performance
- 3rd Generation RT Cores: Up to 2X ray tracing performance
- Axial-tech fans scaled up for 23% more airflow
- New patented vapor chamber with milled heatspreader for lower GPU temps
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
