This is just an observation and public service announcement. If you have the Solarflare SFC-9120 10Gb SFP+ card installed in an R740xd it will spin up the fans and be louder than it needs to be. Most likely due to iDRAC not being able to read the temperature of the card. This is common with Solarflare cards.

Here’s a screenshot of the fan speed at idle load without the card:

And here’s a shot with the card installed:

At 48% PWM the server is quite loud even with high performance fans installed.

When you’re buying a used Dell PowerEdge server or planning an upgrade, it’s easy to focus on CPU core counts and clock speeds. However, one often overlooked specification is memory speed. Even if you install high-speed DIMMs, your processor ultimately determines the maximum speed your memory will operate at.

To see how this works in the real world, I analyzed a large inventory of Dell PowerEdge 14th Generation servers. Rather than looking at the rated speed of the installed memory, I focused on the configured memory speed—the speed the operating system reports the memory is actually running.

The results were interesting.

2933 MT/s Was the Fastest Configured Memory Speed

Across the Dell 14th Generation PowerEdge servers in the dataset, 2933 MT/s was the highest configured memory speed observed.

Although several systems were equipped with 3200 MT/s DIMMs, none of the 14G platforms operated above 2933 MT/s. This aligns with Intel’s Cascade Lake Xeon Scalable processors, which officially support DDR4-2933 as their maximum memory frequency.

CPUs That Reached 2933 MT/s

The following Dell 14G systems achieved the highest configured memory speed found in the dataset.

*The final entry reports memory configured at 2933 MT/s despite the inventory listing 2666 MT/s DIMMs. This may reflect mixed DIMM reporting or inventory collection inconsistencies, but it was included because the configured speed is what the operating system reported.

Most Servers Operated at 2666 MT/s

While 2933 MT/s grabbed the headlines, it wasn’t the most common result.

The majority of Dell PowerEdge 14G systems in the dataset were configured to run memory at 2666 MT/s. This included many popular processors such as:

• Xeon Gold 5217
• Xeon Gold 5218
• Xeon Gold 5222
• Xeon Gold 6230
• Xeon Gold 6234
• Xeon Gold 6242R
• Xeon Gold 6130
• Xeon Gold 6134
• Xeon Gold 6136
• Xeon Gold 6138
• Xeon Gold 6140
• Xeon Gold 6144
• Xeon Gold 6146
• Xeon Gold 6150
• Xeon Gold 6152
• Xeon Gold 6154

These processors deliver excellent overall performance, but they don’t always operate memory at the platform’s maximum possible speed.

Faster DIMMs Don’t Always Mean Faster Memory

One trend appeared repeatedly throughout the dataset.

Several systems were populated with 3200 MT/s DIMMs, but the memory still operated at 2933 MT/s.

Examples include:

• PowerEdge R740 with Xeon Gold 6246
• PowerEdge R740xd with Xeon Gold 6248R
• PowerEdge R640 with Xeon Gold 6252
• PowerFlex R640 S with Xeon Gold 6226

This isn’t a hardware problem—it’s exactly what you’d expect. The processor’s integrated memory controller determines the maximum supported speed, so faster DIMMs simply downclock to the CPU’s supported frequency.

Likewise, there were numerous examples where servers containing 2666 MT/s or even 2933 MT/s memory operated at only 2400 MT/s or 2133 MT/s because of the installed CPU or memory population rules.

What This Means for Buyers

If you’re shopping for a Dell PowerEdge 14th Generation server and memory bandwidth is important, choosing the right processor matters just as much as buying faster DIMMs.

Based on the systems analyzed:

• 2933 MT/s is the highest configured memory speed observed on Dell PowerEdge 14G servers.
• Intel Xeon Gold 62xx Refresh processors consistently delivered the fastest memory performance.
• Intel Xeon Platinum 82xx processors also reached 2933 MT/s in multiple systems.
• Most enterprise deployments still operate at 2666 MT/s.
• Purchasing 3200 MT/s DDR4 memory for a 14G server won’t provide additional performance beyond 2933 MT/s because the CPU limits the operating frequency.

Final Thoughts

The data reinforces an important point: memory performance isn’t determined by the DIMMs alone.

On Dell PowerEdge 14th Generation servers, the processor, memory population, and platform capabilities all work together to determine the final operating speed. If your goal is to maximize memory bandwidth, processors such as the Xeon Gold 6246R, 6248R, 6226R, 6238R, 6252, or the Xeon Platinum 8260–8280 family consistently demonstrated the ability to operate memory at the highest configured speed observed in this dataset: 2933 MT/s.

Rather than chasing the fastest memory modules available, choosing the right CPU will have the biggest impact on whether your server can take full advantage of high-speed DDR4 memory.

You may have come across a so called GPU enablement kit for the R740 and R740xd servers.  In this post I’ll touch on what exactly is in the kit and whether or not you need one. Also make note of the part numbers in the title (where applicable)

What is in the kit?

GPU Air Shroud (DP/N 0RFFDP)

The first piece of hardware in the kit that diverges from what you’d call a traditional setup is the air shroud. Dell makes what they call a GPU air shroud designed specifically for larger GPUs. The purpose of this specialty shroud is to direct more air to the GPU. In theory, it channels more air to the GPU compared to the stock shroud.

Low profile 1U heatsinks (DP/N 0C6R9H)

What should be in the kit but often excluded are 1U low profile heatsinks. If you order the GPU kit and have 2U heatsinks installed you’re going to immediately notice something – the shroud doesn’t fit. The original air shroud/baffle has cut outs which fit around the 2U high performance heatsinks. The GPU shroud does not. If you plan on ordering the GPU kit make sure it has the 1U heatsinks or order them separately.

All 3 risers

If you’ve ordered an R740 or R740xd there’s a good chance you may have gotten a stripped down server equipped with riser 1 with no x16 slots, a low profile x8 riser, and a rear filler to fill up the space where riser 2 and 3 would normally occupy. The GPU kit enables your server with three x16 slots for up to 3 dual width GPUs. All 3 risers are crucial especially if you will be installing 3 GPUs in the server.

Please keep the following in mind – if you use a x16 variant of riser 1 you effectively lose the ability to use a Dell Mini Mono RAID controller. This is because all x16 risers will take up the slot required by the mini controller. You must instead use a PCIe raid controller.

High performance fans (DP/N 04VXP3)

There are 2 types of fans for most Dell PowerEdge servers. Standard performance and high performance. High performance fans are recommended for configurations involving GPUs, rear flex bays, NVME drives, and processors with a higher than average TDP. In this case it’s a good idea to use high performance fans with your GPU. The server will run a lot more quietly with these installed.

GPU power cable

Lastly the kit should have the appropriate GPU power cable to go from the GPU to the motherboard.

Is the kit necessary?

I think the kit started out as a good idea but there are a few downsides. The first major downside is being forced to use 1U low profile heatsinks due to the design of the GPU air shroud. High TDP processors will require the 2U heatsinks to remain cool. What are you to do if you have hot processors and also a GPU? I think having to trade the 2U heatsinks for an air shroud that is arguably only slightly more air efficient for GPUs is not a good tradeoff. It makes more sense if your processors are of a lower TDP which may be the case for you. Anything below 165 Watts and you’re probably okay to use the 1U low profile heatsinks with the GPU air shroud.

The second downside is no one-size-fits-all solution. What if the included power cable doesn’t work with your gpu? What if you only need riser 3 and not the others? What if your server already has 1U low profile heatsinks?

I see these bundles and wonder who is buying them. It’s probably best to track down the part numbers you need and order what  you want on an individual basis, versus the kit.

Overall I think the kit is a suggestion of parts rather than hard and fast rules for what it actually takes to “enable” your server for GPU support. I’ve built plenty of servers with GPUs and haven’t always used the GPU air shroud – and quite frankly haven’t had a single problem.