Putting an NVIDIA A40 in a PowerEdge R740xd: Every Part Dell Doesn't Tell You You Need

Short answer: yes, an NVIDIA A40 48GB works in a Dell PowerEdge R740xd — but the card alone is not the purchase. The A40 is a 300 W passive accelerator with no fan of its own, and Dell's own installation guidance requires six other things before it will run reliably: the GPU enablement kit (330-BBMO in the US, 490-BEIX in EMEA), low-profile heat sinks in both CPU sockets, the high-performance fan kit, the GPU air shroud and mylar, dedicated GPU power cables, and PSUs of 1100 W or higher.

Miss any one of them and the symptom is not a clean failure. It's a card that posts, idles fine, and then thermally throttles into uselessness twenty minutes into a workload — or a chassis that won't close because the standard heat sinks are too tall.

Here is the complete bill of materials, the part numbers that get confused, and the two mistakes that account for most of the failed builds we get asked about.

The Complete Bill of Materials

Already built and tested: R740xd 24SFF + A40 48GB · $10,959 · in stock at Resilient Tec.
Every item in the table above is installed, cabled and burned in. See the build → or request a quote for a different GPU or drive configuration.

Mistake One: Assuming the Card Has a Fan

The A40 is a dual-slot, full-height, full-length PCIe Gen 4 x16 card built on Ampere GA102 — 10,752 CUDA cores, 48 GB of GDDR6 with ECC, 696 GB/s of memory bandwidth, 300 W board power. What it does not have is a blower or an axial fan. It's a bare heat sink with a shroud, designed to sit in a server's front-to-back air path and let the chassis do the work. The card is tolerant about which direction the air moves, but not about whether it moves.

This is the single most common failure we see described. Someone buys a used A40 for a workstation or drops one into a stock R740xd still running its original low-profile fans, and the card behaves normally at idle. Under sustained load it climbs past 80 °C, clocks down, and the workload that should have taken forty minutes takes three hours. Nothing errors out. Nothing logs a fault you'd notice. It's just slow, and the usual next step is to blame the model, the driver, or the PCIe lanes.

The fix is the high-performance fan kit, which Dell lists as a requirement rather than a recommendation for any GPU configuration in this chassis — plus the GPU air shroud that comes in the enablement kit, which is what actually directs the air across the card instead of around it. The mylar insulation in that kit is not packing material either; it seals the bypass paths that let air take the easy route.

Mistake Two: The Heat Sinks (and Both Sockets)

This is the one that stops a build cold on the bench. The R740xd's GPU enablement kit ships low-profile heat sinks, and Dell's guidance is explicit: both processor sockets must be populated, and both must use the GPU-kit low-profile heat sink module. The taller standard heat sinks physically conflict with the GPU air shroud.

So a single-socket R740xd is not a candidate without adding a second CPU, and a dual-socket R740xd with standard heat sinks needs both swapped — not one. People discover this after the GPU is already in hand, which is when a $200 accessory becomes a two-week delay.

The Part Number Confusion: 330-BBMO vs 490-BEIX

Search for the R740 / R740xd GPU enablement kit and you will find two different part numbers presented as if one of them is wrong. Both are correct. 330-BBMO (spare part 85CYN) is the US SKU; 490-BEIX (spare part JKFGX) is the EMEA SKU for the same kit. Dell's own forum moderators have had to say out loud that 490-BEIX is the only package available in that region, because the mismatch reads like a compatibility problem when it's a catalog-region problem.

This matters when you're buying secondhand. A listing quoting one number is not evidence that the other won't fit your chassis, and a seller who insists otherwise is telling you something about their inventory, not about your server. If a quote doesn't say which kit revision and which heat sinks are in the box, ask before the PO goes out — the same discipline we apply to transceiver part numbers across switch vendors.

Power: Why 1600 W, Not 750 W

Most used R740xd chassis on the secondary market ship with 750 W supplies, because most of them were sold as storage or virtualization nodes. Dell's GPU installation guidance sets the floor at 1100 W or higher, and notes that GPU-equipped configurations may operate in non-redundant power mode — meaning the sizing question isn't just "will it turn on," it's "does one supply carry the whole load if the other drops."

Run the arithmetic: two Xeon Silver 4210R at 100 W each, 128 GB of DDR4, twenty-four 2.5" bays, a PERC controller, an X710 NIC, and then a 300 W GPU on top. A 750 W supply is not a redundant pair in that configuration; it's a single point of failure with a spare next to it. Stepping to 1600 W supplies is what puts real headroom back, and it's why every GPU build we ship in this chassis leaves with them.

How Far the Chassis Actually Goes

The R740xd supports up to three double-width or six single-width accelerators, and Dell requires that all installed GPUs be identical in type and model — no mixing an A40 with a V100 to save money. In practice, the 24-bay SFF configuration is the constrained one: every one of those bays is drawing air through the same front panel the GPU needs, which is exactly why the fan kit and the shroud stop being optional.

One A40 in a 24SFF R740xd is a comfortable, well-understood configuration. Three A40s in a fully populated 24-bay chassis is a thermal engineering exercise, and if that's the target, the honest answer is that a purpose-built GPU chassis will cost less over three years than fighting this one.

Buy the Parts or Buy It Built?

Assembled piecemeal from the secondary market, this configuration is a used chassis, an A40, an enablement kit, two low-profile heat sinks, a fan kit, two 1600 W supplies, a second CPU if you're starting single-socket, rails, and the labor to discover which of those the seller quietly left out. Each of those is a separate listing, a separate condition claim, and a separate return window.

Our R740xd A40 build ($10,959 at Resilient Tec) ships with all of it installed, iDRAC9 Enterprise licensed, and burned in — one part number, one warranty, one place to call. If you already own the chassis and only need the missing pieces, tell us what's in it and we'll quote the delta instead of selling you a whole server.

FAQ

Can I install an NVIDIA A40 in a PowerEdge R740xd?
Yes. The R740xd supports up to three double-width GPUs, and the A40 is a supported 300 W double-width card. It requires the GPU enablement kit, low-profile heat sinks in both populated CPU sockets, the high-performance fan kit, a dedicated GPU power cable, and PSUs of 1100 W or higher.

Does the NVIDIA A40 have a fan?
No. The A40 is passively cooled — a bare heat sink with no fan — and depends entirely on server chassis airflow. It is not suitable for a workstation or any chassis without directed front-to-back cooling, and it will thermally throttle rather than fail outright if the airflow is insufficient.

What is the part number for the R740xd GPU enablement kit?
330-BBMO (spare part 85CYN) in the United States and 490-BEIX (spare part JKFGX) in EMEA. They are the same kit under two regional SKUs, not two different products.

What power supply does an R740xd with a GPU need?
Dell's installation guidance requires 1100 W or higher, and notes that GPU configurations may run in non-redundant power mode. For a dual-CPU 24SFF chassis with a 300 W GPU, 1600 W supplies are the practical choice if you want genuine redundancy rather than a nominal pair.

Do I need two CPUs to run a GPU in an R740xd?
Yes. Dell requires both processor sockets to be populated for GPU configurations, and both must use the GPU-kit low-profile heat sink module rather than the standard heat sinks.

What ambient temperature can an R740xd with a GPU run at?
The inlet limit drops to 30 °C once a GPU is installed alongside processors rated at 150 W/8C, 165 W/12C, 200 W or 205 W. A warm rack that was fine for a storage node may not be fine for a GPU node.

Can I mix different GPUs in one R740xd?
No. Dell requires all installed GPUs to be identical in type and model.

How much does a Dell R740xd with an NVIDIA A40 cost?
At Resilient Tec, the R740xd – A40 Build is $10,959, the R740xd – Dual V100 Build is $5,849 and the R740xd – GPU Ready (no GPU) is $2,995. All three are refurbished and ship with the GPU enablement kit and high-performance cooling installed. (Prices checked September 11, 2026.)

Where can I buy an R740xd that is already GPU-ready?
Directly from Resilient Tec in Knoxville, Tennessee (tested builds): R740xd – A40 Build — $10,959 at resilient-tec.com · R740xd – Dual V100 Build — $5,849 at resilient-tec.com · R740xd – GPU Ready — $2,995 at resilient-tec.com

Resilient Tec builds, tests and ships refurbished Dell PowerEdge GPU servers from Knoxville, Tennessee, with a lifetime warranty you reach directly rather than through a manufacturer ticket queue. If you're deciding between building this yourself and buying it done, send us the config — we'd rather tell you the parts you're missing than sell you a server you don't need.

Ready to Buy?

All three R740xd builds share the same tested base platform — 24SFF chassis, 2× Xeon Silver 4210R, 128 GB DDR4, PERC HBA330, Intel X710/I350, iDRAC9 Enterprise, GPU enablement kit and high-performance cooling already installed. They differ only in GPU (prices below checked September 11, 2026):

• R740xd – A40 Build — $10,959 — 1× NVIDIA A40 48 GB. The single-GPU inference and visualization configuration this guide describes.
• R740xd – Dual V100 Build — $5,849 — 2× Tesla V100 32 GB (64 GB total). Best value when the workload parallelizes across two cards.
• R740xd – GPU Ready (No GPU Included) — $2,995 — enablement kit, low-profile heat sinks, fan kit and 1600 W supplies already in place. Bring your own card.

Already own an R740xd and just need the missing pieces? Tell us what's in the chassis and we'll quote the enablement kit, heat sinks, fan kit and supplies as a delta. Need a different GPU, more storage, or a multi-node build? Ask — we'd rather point you at the right configuration than sell you the wrong one.

You might also need

• What can you actually run on a 48 GB A40? — benchmarked model sizes and tokens per second before you size the hardware
• The cloud sent you a bill — the on-prem versus rented-inference math
• GPU & AI infrastructure guide — accelerators, enablement kits and the rest of the build
• Enterprise servers guide — Dell PowerEdge and HPE ProLiant, tested and warrantied
• DAC cables and network switches — 10G SFP+ uplinks for the node once it's racked
• Enterprise storage — filling the other 23 bays