- GPU memory
- 48 GB GDDR6 ECC
- CUDA cores
- 10752
- Memory bandwidth
- 696 GB/s
Loading products and specifications…
Loading products and specifications…
GPU Accelerator
Match the accelerator to the host
NVIDIA A40. Ecc: Supported; Host interface: PCIe 4.0 x16. Ask for current pricing and help choosing the right configuration.
Tell us the configuration, quantity and destination. Your quotation will show availability, condition, warranty and delivery terms.
Start an order requestNVIDIA / PRODUCT DETAILS
A matching connector is only one part of qualification. Check the physical package, power, cooling and software together.
Match memory capacity and supported compute functions to the software workload.
View all specifications →NVIDIA / ORDER & INSTALLATION DETAILS
Distinguish a GPU board, accelerator module and complete NVIDIA system.
Check the qualified host, original power arrangement and cooling.
Review the driver or compute stack and required interconnect components.
Technical overview
Use the full part number when comparing options or asking about system fit.
| Ecc | Supported |
|---|---|
| Host interface | PCIe 4.0 x16 |
| Form factor | PCIe dual-slot passive |
| Manufacturer | NVIDIA |
| Model | A40 |
| Architecture | Ampere |
| Release generation | 2020 |
| CUDA cores | 10752 |
| FP32 peak | 37.4 TFLOPS |
| Published AI performance | Not published as one cross-product metric |
| GPU memory | 48 GB GDDR6 ECC |
| Memory bandwidth | 696 GB/s |
| NVLink / NVSwitch | No guaranteed card-to-card NVLink on the reference product |
| Board / module power | 300 W |
| Cooling | Server-directed passive airflow unless the exact model is explicitly active |
| Slot / platform envelope | Passive server accelerator; bracket height and slot width must match the qualified host |
| Power delivery | PCIe slot plus model-specific auxiliary power; confirm the supplied board assembly |
| MIG | Not supported |
| Virtualization / AI Enterprise | NVIDIA vGPU / AI Enterprise capability depends on license, hypervisor and certified server |
| Software stack | NVIDIA data-center driver, CUDA, NCCL, container toolkit and optional AI Enterprise |
| Display outputs | No display outputs; headless accelerator |
| Media engines | Workload- and generation-specific NVENC/NVDEC; verify codec matrix in the official specification |
| Category | GPU Accelerator |
HOW THIS PRODUCT FITS YOUR SYSTEM
Choose how you plan to use this product.
Match memory capacity and supported compute functions to the software workload.
Confirm the qualified chassis, carrier, riser and interconnect requirements.
Use this checklist to describe the change. The host system and full configuration determine which parts will fit.
YOUR QUOTE REQUEST
Add the quantity, options and installation details you want us to quote.
Confirm the exact manufacturer part number, host compatibility and included components. Optional software is quoted separately only when requested.
SUPPORT & REPLACEMENT
Check support dates, replacement options and whether continued use still makes sense.
We do not have confirmed sales and support dates for this part number.
Send us the full A40 order code, software release and support details before planning long-term use.
Ask us to check the current sales and support statusProducts with similar names or connectors may still be incompatible. Compare the full specifications, installed system and software requirements.
Compare related products These products may be related, but we have not confirmed them as replacements.USING HARDWARE RESPONSIBLY
Consider repairability, remaining support and actual operating demand. Extending service life is useful only when the equipment still meets your security and reliability requirements.
Use measured average input power for the same workload. PSU ratings and processor TDP are different from total system power.
Energy difference = (current W − proposed W) × hours × units ÷ 1,000. Excludes cooling, embodied emissions and purchase costs.CONNECTION DETAILS
This component works inside a compatible system. Verify the host generation and supported option list before choosing memory, processors or accelerators.
Browse the categories above for related equipment, then send us the full configuration so we can check the fit.
PRODUCT DETAILS
Accelerator configuration
Board manufacturer, full P/N, model and form factor
Distinguishes separately orderable boards, modules and system kits.
Host model, available clearance, cooling and power connections
Identifies the installation requirements to confirm for the offered accelerator.
Application, framework, driver and any virtualization requirement
Defines the software-support and entitlement checks for the intended use.
Accelerator, adapter or DPU; exact board code and supported host.
Switch or computing system; its own SKU, firmware and configuration.
Product and edition; deployment rights and support where applicable.
Where NVIDIA lists both a current SKU and legacy Mellanox OPN, retain their exact documented mapping. A family name does not confirm a port layout, GPU board version or software entitlement.
How to read the numbers
| Specification | NVIDIA A100 40GB SXM4 | NVIDIA A40 | NVIDIA A10 | NVIDIA A100 80GB PCIe |
|---|---|---|---|---|
| Architecture | Ampere | Ampere | Ampere | Ampere |
| Year | 2020 | 2020 | 2021 | 2021 |
| Memory | 40 GB HBM2 | 48 GB GDDR6 ECC | 24 GB GDDR6 ECC | 80 GB HBM2e |
| Bandwidth | 1555 GB/s | 696 GB/s | 600 GB/s | 1935 GB/s |
| CUDA cores | 6912 | 10752 | 9216 | 6912 |
| FP32 peak | 19.5 TFLOPS | 37.4 TFLOPS | 31.2 TFLOPS | 19.5 TFLOPS |
| AI metric | Not published | Not published | Not published | Not published |
| Power | 400 W | 300 W | 150 W | 300 W |
| Form factor | SXM4 module | PCIe dual-slot passive | PCIe single-slot passive | PCIe dual-slot |
Use this as a market-position reference, not an automatic drop-in replacement.
48 GB GDDR6 ECC · 300 W · PCIe dual-slot passive
Revalidate memory, power, cooling, interface, software and supply lifecycle.
Workload placement
AI training, inference, HPC and accelerated data processing
Virtual desktops, media services and scale-out inference after license and server qualification
Direct desktop installation without a qualified server/platform
A GPU model alone does not identify the complete product to order. Include the board or OEM P/N, memory capacity, form factor, cooling, bracket, region and warranty.
No. Treat the lineage as a workload-positioning guide. Validate host interface, physical envelope, cooling, power, firmware, driver, software entitlement and application certification.
Compare like-for-like precision, sparsity state, clocks, power limit and form factor. Gaming AI TOPS, Tensor FLOPS and measured application throughput are not interchangeable units.
No guaranteed card-to-card NVLink on the reference product PCIe peer access, NCCL topology, CPU lanes, NUMA placement and rack cooling must also be validated.
01 / Application planning
Match the option, spare or assembly number and the relevant hardware revision.
Review the host, connectors, physical fit and dependencies as one configuration.
Keep quantities, included kits and target systems attached to each BOM line.
02 / Inspection & testing
Agree the inspection and test scope for the offered condition before confirming the order.
03 / Packing & international delivery
Confirm the packing list, delivery address and receiving requirements before shipment.
Protect
Specify packing around the component, its sensitivity and the quantity ordered.
Prepare
Select packing suitable for the hardware, combined order and transport route.
Deliver
Confirm the destination and shipment scope before arranging collection.
Send the country, postal code, requested date and receiving restrictions with your enquiry. Packing, freight and delivery timing are confirmed in the written quotation.
Shipping guide ↗04 / Warranty & after-sales
Keep the warranty terms with the part number and supplied condition. The quotation or order defines the applicable coverage for your item.
Warranty & RMA guide ↗Include the product label, order reference and a description of the fault.
Provide the requested identification and diagnostic details.
Agree the authorized next step before shipping a return.
Repair, replacement or other remedies depend on the agreed warranty terms and the fault assessment.
Prepare the next step
Include quantities, destination, test evidence and warranty requirements.
Model selection & order planning
NVIDIA A40
Explore more products below. Please confirm compatibility before selecting a replacement.
Before you request a quote
Include the full model, manufacturer P/N, revision and quantity.
Tell us the host model, generation, installed configuration and required supporting hardware.
Specify condition, testing requirements, destination and requested delivery date.
GPU Accelerator
ModelA800 40GB Active
GPU Accelerator
ModelA10
GPU Accelerator
ModelA800 80GB PCIe
GPU Accelerator
ModelH100 NVL 94GB