Dell R630 vs R640: Choosing a 1U Compute Node

A 1U dual-socket node is a compromise machine. You get two sockets and 24 DIMM slots in half the height of a 2U server, and you pay for it in expansion slots, drive bays and thermal headroom. The R630 and R640 make that trade differently, and the differences show up fastest in memory-heavy and high-TDP builds.

The R630 is Dell's 13th-generation 1U server: two Socket 2011-3 processors from the Xeon E5-2600 v3 or v4 families, four memory channels per socket. The R640 is the 14th-generation successor: two Socket P (LGA 3647) processors from the 1st- or 2nd-generation Xeon Scalable families, six memory channels per socket. Both offer 24 DIMM slots and a maximum of three PCIe 3.0 slots.

Side by side

Item	PowerEdge R630 (13G)	PowerEdge R640 (14G)
Form factor	1U, dual socket	1U, dual socket
Socket	2 x Socket 2011-3	2 x Socket P (LGA 3647)
Processors	Xeon E5-2600 v3 / v4	Xeon Scalable 1st and 2nd gen
Max cores per socket	22	28
Max CPU TDP	145 W	205 W
Memory channels per CPU	4	6
DIMM slots	24	24
Max memory speed	2133 MT/s (v3), 2400 MT/s (v4)	2666 MT/s (1st gen), 2933 MT/s (2nd gen)
Front drive bays	4 x 3.5", 8 x 2.5", 10 x 2.5", or 24 x 1.8" SSD	4 x 3.5", 8 x 2.5", or 10 x 2.5"
Rear drive bays	None	None
NVMe	Up to four Express Flash bays on specific backplanes	NVMe-capable backplanes on the 10-bay configuration
PCIe	Gen 3, up to 3 slots	Gen 3, up to 3 slots
Onboard networking	rNDC, up to 10 GbE	rNDC, up to 25 GbE options
Boot	Internal dual SD (IDSDM), front bays	BOSS-S1 mirrored M.2, IDSDM, front bays
Management	iDRAC8	iDRAC9

Six channels versus four

Both servers have 24 DIMM slots, which makes them look equivalent on a parts list. They are not. With six channels per socket, the R640 reads and writes memory across 50 percent more paths than the R630 at the same DIMM count, and it does so at a higher rated speed. For a 1U node hosting 20 to 40 virtual machines, or running a database that lives mostly in RAM, that is usually the difference you feel first, ahead of core count.

Population rules follow from the channel count. On the R630, you add memory in groups of four per processor to keep every channel active. On the R640, groups of six. Migrating a memory inventory between the two without re-planning leaves channels empty and quietly costs performance.

Watch the processor SKU as well. Xeon Silver 4100-series parts cap memory at 2400 MT/s and carry two UPI links instead of three, so a Silver-based R640 will not show the platform's headline memory speed. And on both platforms, filling two DIMMs per channel can lower the operating speed depending on DIMM rank and CPU model, so check the population guidelines against the exact processor before ordering 24 modules.

1U constraints: thermals, slots, and boot devices

The R640 supports processors up to 205 W, which on paper opens up parts like the 24-core Gold 6248R. In 1U, that headroom comes with conditions: high-TDP processors require the high-performance heatsink, and Dell applies configuration and ambient-temperature restrictions to those builds. Fan speed, and therefore acoustics and fan power draw, rise accordingly. If you want core density in 1U without fighting the thermal envelope, a 165 W 28-core part such as the Gold 6238R is usually the more comfortable choice than a 205 W SKU. The R630 tops out at 145 W, which limits its ceiling but also keeps its thermal behavior predictable.

Expansion is tight on both. Three PCIe 3.0 slots is the maximum, and real riser configurations often deliver fewer usable slots depending on the drive backplane and PSU choice. Count your cards before committing: a dual-port 25 GbE adapter, a SAS HBA for external storage, and a fiber channel card will consume everything you have. The R640's rNDC options extend to 25 GbE, so onboard networking can absorb a requirement that would otherwise cost you a slot. The R630's rNDC tops out at 10 GbE.

Boot media is the other quiet difference. The R640 supports the BOSS-S1 card, which puts the hypervisor on a mirrored pair of M.2 SATA modules on a low-profile card, keeping all front bays free for data and avoiding the SD-card wear problems that plagued IDSDM-based ESXi installs. On the R630 the choice is between the internal dual SD module and giving up a pair of front bays to the operating system. There is no rear drive cage in a 1U R630 to move boot media into, which is exactly why the BOSS card mattered when it arrived on 14G.

Front bays deserve a mention too. The R630 offered an unusual 24 x 1.8-inch SSD configuration alongside the conventional 4 x 3.5-inch, 8 x 2.5-inch and 10 x 2.5-inch layouts, which is worth knowing about if you find one, though 1.8-inch SSD supply is thin now. The R640 sticks to 4 x 3.5-inch, 8 x 2.5-inch and 10 x 2.5-inch, with NVMe-capable backplanes on the 10-bay version.

Which node fits

Choose the R630 when the node is CPU-modest and predictable: hypervisor hosts with a small VM count, application and web tiers, DNS/DHCP/AD infrastructure, jump hosts, test and lab clusters, or scale-out roles where adding nodes suits the design better than making one node faster. It is also the sensible pick if you already run 13G hardware and hold spare rails, carriers, PSUs and PERC controllers.

Choose the R640 when memory bandwidth or capacity is the constraint, when the host needs NVMe on the front, when 25 GbE onboard saves a slot, when mirrored M.2 boot simplifies your build, or when the machine has to stay on a supported hypervisor release for several more years. Newer VMware and enterprise Linux versions have progressively dropped older processor generations, and E5-2600 v3/v4 is closer to that line than Xeon Scalable, so check the exact CPU on the vendor's compatibility list either way.

Parts do not carry across generations. 14G drive carriers, rails, risers and heatsinks are all different from their 13G equivalents, so standardizing on one generation across a rack is worth more than squeezing a slightly better spec out of a mixed fleet.

Where to go from here

Browse the available Dell PowerEdge R630 systems and Dell PowerEdge R640 systems to see which processor pairs, backplanes and network daughter cards are on hand. If your build is not on the shelf, request a quote and specify CPUs, memory, drive layout, controller, rNDC and boot device for that exact configuration. Send the workload and rack constraints along with the parts list and we will flag anything, thermal or otherwise, that will not behave the way the spec sheet implies.