Dell PowerEdge R940 8-Bay 2.5"
Configure Your System:
1 Processor Required
On series: four-socket operation is what separates Gold and Platinum from the rest of the Xeon Scalable range, so the list here starts where two-socket lists end. 6100-series parts are 1st Gen (Skylake-SP); 6200 and 8200-series are 2nd Gen (Cascade Lake-SP) with higher clocks per tier and Optane PMem support. Both generations are proven in this chassis.
On selection, match the CPU's character to the workload rather than chasing the biggest number. Transactional databases and ERP platforms reward clock speed and cache: the frequency-optimized Golds are often the faster real-world choice for them than a bigger, slower Platinum. Massive consolidation and in-memory platforms reward core density and memory bandwidth, which is where the high-core-count Platinums earn their premium. Four of anything on this page is a serious amount of compute; the question is what shape.
All four CPUs must match exactly: same model, same stepping. We build them that way, and it is worth remembering if you ever expand a partially populated unit.
Not sure which shape your workload wants? Tell us what it runs at quote time, or call 1-800-778-1545 and we will spec it against the actual application.
2 Heat Sink Optional
The standard heatsink covers the moderate-wattage CPUs in the list in a normal datacenter environment.
The performance heatsink is required for the top-wattage CPUs, and a four-socket buyer is rarely shopping the bottom of the list; plan on performance heatsinks on all four sockets for any seriously configured unit. We also spec them on densely populated NVMe configurations and at ambient inlet temperatures above 30°C.
The cost difference is small against the price of the processors they protect, and it is a lot cheaper than a thermal event on the most important server in the building. Confirm all four are in the quote and move on.
3 Memory (RAM) 48 DIMM slots, modules added in sets of 4 Required
Fill all 48 slots. On this platform the case is performance first: full population keeps every channel on every socket working, which is exactly what the in-memory databases, ERP platforms, and analytics engines this chassis attracts are bandwidth-hungry for. Cost per gigabyte follows the same logic as everywhere else: 48 smaller modules reach a given capacity for meaningfully less than 24 larger ones.
One rule that matters more here than on any two-socket server: keep the configuration identical across all four sockets. These workloads are NUMA-sensitive, and an unbalanced memory map means some cores wait on remote memory while others do not. One matched set, evenly distributed, is the whole discipline.
768 GB (48× 16 GB) is the sensible floor for a server in this class
1.5 TB (48× 32 GB) is our default, and the best cost per GB in the lineup
3 TB (48× 64 GB) for large in-memory databases and analytics
Past 3 TB you are into 128 GB LRDIMMs, and the R940 is one of the few platforms where that conversation is routine rather than exotic; buyers at that capacity chose this chassis precisely because the workload cannot be split. Call 1-800-778-1545 and we will spec it.
4 RAID Controllers Optional
PERC H740P (8 GB NV cache, battery-backed) is the default, and on this platform it is barely a decision. Transactional databases, ERP platforms, and analytics scratch space all lean on the controller's non-volatile cache, and the H740P's 8 GB is what keeps local storage out of the workload's way. Against the price of the rest of this configuration, choosing a smaller controller is saving in exactly the wrong place.
PERC H730P (2 GB cache) is defensible only when local storage is an afterthought: a boot volume and little else, with the real data on a SAN or NVMe. If that describes your build, it will do the job.
Entry controllers do not belong in this chassis, full stop.
Mind the array math wherever the data lands locally: RAID 1 or 10 for transaction logs and hot data, RAID 6 for large-capacity volumes, consistent drive models within each array, and a hot spare on any array this business depends on.
Running your data path over NVMe instead? Those drives connect over PCIe and bypass the controller entirely; redundancy happens in software there, and the controller here only matters for whatever SAS or SATA drives remain. Call 1-800-778-1545 and we will spec the whole storage path against the workload.
5 Storage Drives Select up to 8 drives (0/8 Slots Used) Optional
Buy SSDs, without much debate on this platform. The workloads that justify four sockets are latency-sensitive, and SAS SSDs are the production pick for the controller-attached bays: higher endurance, dual-port connectivity, and sustained-write behavior that holds up under the load this chassis exists to carry. Spinning disk has no real role here; if bulk capacity is part of the picture, it belongs on a storage platform like the R740xd, not in these bays.
The R940 also takes direct-attach NVMe in a portion of its bays, and for databases and in-memory platforms that is where the chassis earns its reputation: PCIe-attached flash feeding four sockets is the configuration to beat. On NVMe, mind endurance ratings: mixed-use (3 DWPD) where the writes land, read-intensive (1 DWPD) for capacity tiers. Remember NVMe bypasses the RAID controller, so redundancy for those drives is planned in software.
Buy new, and on this server treat it as a rule rather than advice. A refurbished drive arrives with unknown miles on its endurance budget, and this is the last chassis on which to discover that mid-quarter. Our refurbished drives have honest roles elsewhere in the catalog; this is not one of them.
Boot belongs on a Dell BOSS card, available in the Add Ons section below, keeping every front bay on mission duty.
Tell us the workload at quote time and we will spec the split between NVMe and SAS, or call 1-800-778-1545.
6 Remote Access Required
Take Enterprise. The R940 is bought to carry the workloads a business cannot lose, and iDRAC9 Enterprise is the tool that keeps a bad night short: full remote KVM and virtual media mean you can reach the server's screen, mount an ISO, and rebuild from anywhere, even when the operating system is down. Express lacks the remote console, so any problem below the OS level means someone physically at the rack, and the workloads this chassis carries do not wait politely for the drive in.
Enterprise also brings automated firmware updates, two-factor authentication, and secure erase, which matters on a server that will retire holding the most sensitive data in the building.
The license is a rounding error against this configuration's price, and the first incident it shortens pays for it many times over. This is the easiest recommendation on the page.
7 Power Supply 2x included in build Required
Add it up before choosing: four CPUs at full tilt, 48 DIMMs, and a bay full of flash is a fundamentally different load than anything a two-socket server presents. The lower-wattage pairs that anchor other configurators are not the default here.
The 1600W pair is our default for a seriously configured R940: four high-wattage CPUs, full memory, and a loaded backplane, with the 50 percent load target intact.
Step down a tier only for genuinely moderate builds: lower-wattage CPUs and restrained storage.
Step up to the top tier when the build maxes every axis at once, or when your datacenter's power redundancy scheme demands that one PSU carry the entire load alone with margin to spare, which is exactly the scheme most facilities running four-socket flagships enforce.
Size for roughly 50 percent load at your expected peak: a PSU at half capacity runs at peak efficiency, runs cooler, and lasts longer than one working near its limit. When the build sits between tiers, size up; against this build sheet the price step is a rounding error. Watch the estimated TDP counter at the bottom of the page as you add components to see where your build lands.
8 Network Cards Required
A four-socket flagship concentrates a rack's worth of work into one box, which concentrates a rack's worth of traffic into this card. Treat 10 Gb as the minimum: the quad 10 Gb options (Intel X710, X550, Broadcom 57840S) are the default recommendation, giving production, backup, and management traffic their own links. The 2x 10 Gb + 2x 1 Gb combo cards are the economy pick only when a SAN carries the storage traffic on its own fabric. Quad 1 GbE has no business on this platform; the server will saturate it before breakfast.
If the R940 is serving as a database or analytics engine for many clients, consider 25 GbE from the Add Ons section; the card is a small line item on this build sheet, and it removes the network from the list of things that can bottleneck four sockets.
Then port type, and this is the one that catches people. SFP+ ports take fiber or DAC cables and talk to datacenter-grade switches. The -T models (X540, X550) are 10GBASE-T copper on ordinary RJ45 cabling. Neither is better; they must simply match your switch. Order SFP+ against a copper-only switch and the server racks up with nothing to plug into.
Call 1-800-778-1545 and we will spec the whole path, switch to server.
9 Operating System Required
Windows Server 2022 is the version to buy today, and on this chassis the edition question matters more than usual: Datacenter edition licenses unlimited virtual machines on the host, and at the consolidation density a four-socket flagship invites, Datacenter usually beats Standard's per-VM math decisively. Windows licenses by core across all four sockets, so the core counts you chose in the processor step flow straight into the license bill; we will price both editions against your configuration at quote time.
Fair warning on Evaluation Editions: free for 180 days, then they expire. Labs only. If the server is going into service, buy the license now.
VMware ESXi 8.0 is the one to deploy for virtualization duty; 6.7 and 7.0 are past end of support and only belong on hosts joining an existing cluster. Licensing is yours to bring, and at four sockets it is a serious line item; price it before you commit to the platform.
Ubuntu Server LTS and the enterprise Linux family cover the database, ERP, and analytics duty this chassis is frequently bought for, with five years of free updates on the LTS side.
Proxmox VE runs well here for buyers consolidating away from VMware economics, though most four-socket buyers arrive with the platform already decided.
No OS is there for teams imaging from their own deployment infrastructure.
Tell us the workload at quote time, or call 1-800-778-1545.
WServer Warranty
Dell PowerEdge R940 8-Bay 2.5" Drives [14th Gen] Detailed Review
The Dell PowerEdge R940 is the 14th-generation 4-socket 3U rack server, Dell's scale-up flagship and the successor to the 4U R930. The 8-Bay 2.5" is the mainstream configuration: four 2nd Generation Intel Xeon Scalable processors, up to 48 DDR4 DIMM slots, 13 PCIe Gen3 expansion slots, and eight 2.5" hot-swap front bays sized for OS and hot application data. This is a refurbished enterprise platform, fully tested and ready for production scale-up workloads.
Where the 2U R840 maximizes 4-socket compute density in less rack space, the R940's 3U chassis trades a rack unit for I/O headroom: more PCIe slots, more expansion paths, and the Processor Expansion Module that unlocks the full 4-socket slot count. For SAP HANA, large Oracle and SQL Server consolidation, and in-memory analytics where 4-socket compute has to coexist with multiple high-bandwidth cards, the R940 is the platform we quote over the R840.
To configure a build, call 1-800-778-1545 and our account team will scope processors, memory population, storage controllers, and PCIe allocation against your workload. Every chassis ships with our 180-day warranty after a 12+ hour burn-in, and volume pricing starts at 5 units. We return a validated configuration within 24 hours.
Where the R940 Fits in the Family
The R940 sits at the top of Dell's 14th-generation rack lineup. Below it, the R740 8-Bay 2.5" handles dual-socket workloads that do not need 4-socket scale-up. Beside it, the R840 8-Bay 2.5" delivers the same 4-socket compute in a denser 2U chassis. The R940's reason to exist is expansion: when a 4-socket build also needs multiple Fibre Channel HBAs, InfiniBand, NVMe expansion cards, or GPUs at the same time, the 2U R840 runs out of slots first. Choose the R940 when PCIe slot count is a genuine architectural constraint, not just when you need four sockets.
Storage - 8 SFF Bays
Eight 2.5" SAS/SATA hot-swap bays. The 8-Bay R940 is a compute-and-expansion platform, not a storage-dense one. These bays are sized for the operating system, transaction logs, and hot application data, with bulk capacity living on SAN, NVMe expansion cards, or the higher-density R940 24-Bay 2.5" chassis. Maximum raw capacity at the 8-bay front is roughly 61 TB with 7.68 TB SSDs, though most scale-up deployments run far less local storage than that.
For boot, we quote the Dell BOSS-S1 card: dual mirrored M.2 SATA SSDs on a dedicated controller, presented as a hardware RAID 1 volume. It keeps the OS off the front bays and frees all eight 2.5" slots for data. BOSS drives are not hot-swap, which is the right tradeoff for a boot device that should rarely be touched in production.
Storage Controllers
The R940 has a platform quirk worth knowing before you order: it accepts full-height PERC adapters only, in slot 1 (primary) and slot 6 (secondary), and does not support the mini-PERC form factor used elsewhere in the 14th gen line. Plan the controller choice around those two dedicated slots.
The controller lineup: PERC H330 (no cache, entry tier for light workloads), PERC H730P (2 GB cache, battery-backed, the solid general-purpose choice for mixed or read-heavy work), and PERC H740P (8 GB NV cache, battery-backed, our production default for write-intensive or transactional workloads where local storage is load-bearing). For software-defined storage stacks such as vSAN, Storage Spaces Direct, or Ceph, the HBA330 pass-through HBA is the right call. The PERC H840 is available when you need to drive external SAS enclosures. We do not quote S140 software RAID for production; it is a dev and test option only.
Processors
Up to four 2nd Generation Intel Xeon Scalable (Cascade Lake-SP) processors, up to 28 cores each, for as many as 112 cores and 224 threads in a fully populated quad-socket build. One platform detail drives most R940 configurations: a 2-CPU build runs without the Processor Expansion Module and behaves as a 2-socket, 24-DIMM machine with 7 PCIe slots. The PEM installs automatically with 4 CPUs, lighting up the full 48 DIMM slots and all 13 PCIe slots. If your design needs the slot count or the memory capacity, it needs four processors, not two.
Our most common balanced specification is four Gold 6230 (20 cores, 2.1 GHz, 125W) for 80 cores total, which keeps thermals and licensing manageable. For maximum per-socket performance we quote the Platinum 8260 (24 cores, 165W) or the top-bin 28-core parts. On the high-TDP SKUs, confirm the high-performance heatsink and fan configuration is specified; a missed heatsink on a 165W-plus CPU is the single most common stability problem we see on scale-up builds under sustained load.
Memory
Six memory channels per CPU, 2 DIMMs per channel, for 12 DIMM slots per socket and 48 slots across a fully populated 4-socket system. Maximum capacity is 6 TB using 64 GB LRDIMMs. With Intel Optane DC Persistent Memory in the mix, the platform reaches up to 15.36 TB combined across DCPMM and LRDIMM, which is what makes the R940 a genuine SAP HANA and large in-memory database host rather than just a high-core-count server.
Population rules matter here. For best performance, populate all six channels per CPU evenly, either 6 or 12 DIMMs per socket. Registered ECC DDR4 (RDIMM and LRDIMM) only. RDIMM is the right choice for the best balance of frequency, capacity, and rank flexibility; LRDIMM is what you step up to when you need the 64 GB-and-larger modules that get you to the 6 TB ceiling. Remember the PEM rule: half the memory slots are physically on the expansion module, so a 2-CPU build tops out at 24 DIMMs.
Networking and PCIe Expansion
Networking runs through a Flexible Rack Network Daughter Card (rNDC) that does not consume a PCIe slot. Options are 4x 1GbE, 4x 10GbE, 2x 10GbE plus 2x 1GbE, or 2x 25GbE. For most scale-up database and virtualization hosts we quote the 2x 25GbE or 4x 10GbE option; the 1GbE variants are reserved for management-plane-only or legacy environments.
PCIe expansion is the R940's headline advantage. A 2-socket build supports 7 PCIe Gen3 slots (slots 1 through 7). Adding the third and fourth processors brings the Processor Expansion Module online with two additional risers, for six more slots (8 through 13) and 13 total. That slot budget is why the R940 wins over the R840 for architectures that stack dual FC HBAs for redundant SAN paths, an InfiniBand or RoCE cluster interconnect, and NVMe expansion all in the same chassis.
GPU Support
Depending on riser configuration, the R940 accommodates up to 4 double-width GPUs or up to 8 single-width GPUs. This is real GPU capacity for a general-purpose scale-up server, suited to database acceleration and mixed analytics-plus-compute workloads. A clarifying note for buyers cross-shopping: the dedicated GPU-database variant is the R940xa (a separate 4U platform with a 1:1 CPU-to-GPU design and 32 front bays). The standard R940 covered here is the 3U server. If your workload is GPU-first rather than compute-first, the R940xa is worth a separate conversation.
Management - iDRAC9 Generation
iDRAC9 with Lifecycle Controller, Enterprise license required for production-grade remote management (virtual console, virtual media, automated deployment). The platform carries Silicon Root of Trust for boot integrity and supports a TPM 2.0 module, which is required for NIST, CMMC, FedRAMP, HIPAA, and PCI DSS compliance frameworks. Quick Sync 2 enables at-the-rack management from a phone or tablet over Bluetooth, and the platform integrates with Dell OpenManage Enterprise for fleet management. Specify TPM 2.0 up front if you are in a regulated environment; retrofitting it is avoidable friction.
Power and Cooling
Dell hot-swap redundant PSUs in 1100W, 1600W, 2000W, and 2400W tiers, configured as a 1+1 redundant pair. Size the supply to the processor count, GPU load, and PCIe card draw, not just the CPUs. The table below is a starting point for the 8-Bay; final sizing depends on your exact CPU, memory, and card mix.
| Configuration | PSU Recommendation | Est. Peak Draw |
|---|---|---|
| Balanced (4x Gold 6230, full RAM, no GPU) | 2x 1600W Platinum | ~1100W |
| Performance (4x Platinum 8260, full RAM, dual HBA) | 2x 2000W Platinum | ~1450W |
| Heavy (4x top-bin, full RAM, GPUs + expansion) | 2x 2400W Platinum | ~1900W |
A fully loaded 4-socket scale-up chassis is a real datacenter power draw. Confirm your rack PDU and circuit budget before specifying the 2400W tier.
Physical Specs & Platform Notes
- Form factor: 3U rack server, the more expandable successor to the 4U R930. The 3U-versus-2U-R840 decision is about PCIe slots and I/O paths, not compute or memory, which are equivalent between the two platforms.
- PCIe expansion: 7 slots in a 2-socket build (slots 1-7), 13 slots in a 4-socket build with the Processor Expansion Module (slots 8-13 added). A mix of x8 and x16 Gen3 electrical lanes across the risers.
- Parts availability: Strong. The R940 shares the broad 14th-gen PowerEdge parts ecosystem, and refurbished components, rails, and accessories are readily sourced. Third-party maintenance is the standard production support path in 2026 as Dell ProSupport on 14th gen approaches end of extended support.
- Accessories we recommend: the Dell BOSS-S1 boot card (dual M.2 SATA, hardware RAID 1) to keep the OS off the front bays, the LCD bezel for at-a-glance status and Quick Sync 2 management, 3U sliding rails, and a cable management arm for a chassis this deep.
- Platform notes: Full-height PERC only in slots 1 and 6, no mini-PERC. The Processor Expansion Module is mandatory for 4-socket operation and installs automatically with 4 CPUs. No 3.5" LFF drive support on this platform; it is a 2.5" SFF chassis.
Our Assessment
Where it excels: The R940 8-Bay is the right answer when 4-socket scale-up compute has to share the chassis with heavy PCIe expansion. SAP HANA scale-up nodes, large Oracle and SQL Server consolidation, and in-memory analytics that pair 6 TB of memory (or 15.36 TB with Optane) with redundant SAN connectivity and cluster interconnects are squarely in its lane. The 8-bay front is the correct storage choice here: boot plus hot data local, bulk capacity on SAN or NVMe expansion.
Where to look instead: If you need four sockets but minimal PCIe expansion, the denser R840 8-Bay 2.5" gives you the same compute in 2U. If two sockets cover your workload, the R740 8-Bay 2.5" is a far more cost-effective platform. If you need large local SAS/SATA or NVMe capacity alongside the 4-socket compute, move to the R940 24-Bay 2.5".
Bottom line: This is a specialist scale-up platform, not a general-purpose workhorse, and it should be quoted as one. The typical buyer is consolidating a mission-critical database or running an in-memory workload that genuinely needs four sockets and a deep PCIe budget at once. If that is the workload, the R940 8-Bay is the most flexible 14th-gen platform Dell built for it. If it is not, one of the alternatives above will cost less and rack denser.
Where the R940 Fits in 2026
The R940 is a 14th-generation platform, two generations behind the 15th-gen R950-class scale-up servers and three behind current 16th-gen hardware. For new mission-critical greenfield deployments with long support horizons, newer generations are worth evaluating. But for refurbished scale-up procurement in 2026, where the workload is well understood and the budget is finite, the R940 delivers 4-socket Cascade Lake compute, 6 TB of memory, and a 13-slot PCIe budget at a fraction of new-platform cost. Dell ProSupport on 14th gen is approaching end of extended support, so plan on third-party maintenance as the production support path.
Honest Limitations
- Full-height PERC only, restricted to slots 1 and 6, with no mini-PERC support. This constrains how you allocate the slot budget when you also want HBAs and GPUs.
- A 2-CPU build is not a "smaller R940" in the way buyers sometimes assume: without the Processor Expansion Module it drops to 24 DIMM slots and 7 PCIe slots. The full platform requires four processors.
- The 8-bay front is genuinely limited for local storage. If you discover mid-deployment that you need bulk local capacity, you are looking at the 24-Bay chassis, not an upgrade path on this one.
- This is a power-hungry, deep 3U chassis. Confirm rack depth, PDU capacity, and cooling before committing, especially on GPU or full-expansion builds.
- It is a specialist platform. For workloads that do not need 4-socket scale-up, you are paying for capability you will not use.
Workload Fit
| The R940 8-Bay is right for | Consider alternatives for |
|---|---|
| ✅ SAP HANA scale-up nodes | ❌ Dual-socket workloads (use the R740) |
| ✅ Large Oracle / SQL Server consolidation | ❌ Storage-dense local deployments (use the R940 24-Bay) |
| ✅ In-memory analytics with Optane PMem | ❌ 4-socket compute with minimal PCIe needs (use the R840) |
| ✅ 4-socket compute plus heavy PCIe expansion | ❌ Rack-density-constrained datacenters (the R840 is 2U) |
| ✅ GPU-accelerated database (up to 4 double-width) | ❌ GPU-first workloads (evaluate the R940xa) |
Where to Look Instead
- Same compute, less rack space: Dell PowerEdge R840 8-Bay 2.5" delivers 4-socket compute in 2U for builds that do not need the R940's PCIe headroom.
- More local storage: Dell PowerEdge R940 24-Bay 2.5" is the maximum-density configuration of this platform.
- Storage-dense 2U scale-up: Dell PowerEdge R840 24-Bay 2.5" pairs 4-socket compute with 24 SFF bays in 2U.
- Dual-socket alternative: Dell PowerEdge R740 8-Bay 2.5" for workloads that do not need four sockets.
- Cross-vendor counterpart: the closest HPE scale-up platform in our catalog is the HPE ProLiant DL580 Gen9, a 4-socket flagship. Note two differences: it is a 4U Gen9-era platform (one hardware generation behind the 14th-gen R940), so cross-shop it on workload fit and budget rather than as a like-for-like generational match.
Ready to Configure?
Tell us your workload (HANA, Oracle, SQL Server, virtualization, or HPC), your processor and memory targets, your PCIe card requirements, and any compliance framework you fall under. Call 1-800-778-1545 and our account team will return a validated R940 8-Bay configuration within 24 hours. Every build is backed by our 180-day warranty and a documented 12+ hour burn-in, with volume pricing available from 5 units. No retail checkout: this is a configured, quoted scale-up platform, and we scope it with you before you commit.
Save Your Design
Click the Add to Quote button at the bottom of your screen to save your design as a draft order for future reference and to check for discounts, lead time, and availability. Most servers ship within 1-3 days.
