The Dell PowerEdge R840 8-Bay 2.5" is the 14th generation 4-socket 2U rack server: the scale-up platform in Dell's 14th gen lineup for workloads that have genuinely exhausted dual-socket compute and memory headroom. Built on the Intel Purley platform with 2nd Generation Intel Xeon Scalable (Cascade Lake) processors, it carries up to four CPUs, up to 6 TB of memory across 96 DDR4 DIMM slots, eight 2.5" SFF hot-swap bays, up to 8 PCIe Gen3 slots, iDRAC9 management with Silicon Root of Trust, and up to four Dell Flex Slot power supplies. This is the Dell answer for SAP HANA scale-up, Oracle large-instance databases, mission-critical virtualization at extreme VM density, and any application designed to scale vertically rather than horizontally.

Refurbished and configured to order. The 8-Bay 2.5" variant is the standard R840 configuration: maximum 4-socket compute paired with eight SFF bays for OS, application binaries, and hot dataset staging, with primary bulk storage expected on SAN, NFS, or distributed file systems. It is the right starting point for most 4-socket deployments where local storage is a supporting role rather than the main event.

To configure a build, call 1-800-778-1545 or use the quote form below. Every refurbished unit ships under our 180-day warranty with 12+ hour burn-in testing, and volume pricing starts at 5 units. R840 builds benefit from a design conversation early: workload architecture, SAP or Oracle licensing implications, power budget at 4-socket TDP, and thermal validation all matter before hardware selection.

Where the R840 Fits in the Family

The R840 is a fundamentally different platform from the dual-socket R640 and R740. Where the R740 tops out at 56 cores (28+28) and 1.5 TB of standard memory across 24 DIMM slots, the R840 carries up to 112 cores across four sockets and up to 6 TB of memory across 96 DIMM slots. It is the 14th gen platform for workloads that do not scale horizontally: the workloads where a single OS instance needs to see all the cores and all the memory.

Be direct about the 4-socket decision: most enterprise workloads do not require 4-socket servers. The dual-socket R640 and R740 handle the vast majority of virtualization, database, and application serving workloads at materially lower cost and complexity. The R840 makes sense when one of the following is genuinely true:

• A specific workload requires scale-up rather than scale-out. SAP HANA on a single certified server. Oracle Database Enterprise where licensing economics favor fewer sockets with more cores each. Microsoft SQL Server Enterprise where per-core licensing makes one high-core-count server cheaper than several smaller ones.
• The application is not horizontally scalable. Legacy enterprise applications, in-memory analytics platforms, or single-instance databases that cannot be sharded across nodes.
• Maximum single-chassis memory capacity is a genuine architectural requirement. 6 TB in a 2U chassis is meaningful when the working set has to fit in a single server's RAM.
• Per-socket Oracle or SQL licensing creates the right economics. Oracle Database Enterprise charges per physical core; a 4-socket server with 4x 24-core CPUs licenses 96 cores under one server count. The same cores split across two dual-socket servers count as two servers. This is a discussion to have with your Oracle licensing team before committing to architecture.

If the workload can distribute across multiple dual-socket nodes without licensing penalty or architectural friction, the R740 is almost always more cost-efficient. The R840 is a precision tool for scale-up requirements, not a default upgrade from the R740. When 8 SFF bays is not enough local storage alongside 4-socket compute, the R840 24-Bay 2.5" is the higher-density variant of this same platform.

Storage - 8 SFF Bays

Eight 2.5" SAS/SATA hot-swap bays in the front of the chassis. The R840's primary differentiation is compute and memory scale-up, not storage density. Eight SFF bays is correctly sized for the common 4-socket workload pattern: OS, application binaries, and hot dataset staging, with primary data living on SAN, NFS, or a distributed file system. For workloads that need large local storage alongside 4-socket compute, the 24-Bay variant is the right starting point rather than this chassis.

Drive options span the full 14th gen SFF portfolio: SAS SSDs in mixed-use and read-intensive endurance tiers (480 GB through 7.68 TB), SATA SSDs for cost-optimized boot and OS roles, SAS HDDs at 10K and 15K for moderate-IOPS data, and self-encrypting drive (SED) variants for compliance-regulated deployments. Common R840 8-Bay storage profiles in production:

• SAP HANA appliance. A boot pair via BOSS, with the 8 front bays carrying mixed-use SAS SSDs in RAID 10 for HANA log and shared volumes. Primary HANA data volumes mirror to external storage; local SSDs handle log persistence and warm-data staging.
• Oracle Database with ASM on SAN. BOSS for OS plus Oracle Grid Infrastructure binaries, front bays available for local Fast Recovery Area or archive log staging. Primary database storage on Fibre Channel or iSCSI SAN via an FC HBA in PCIe expansion.
• Mission-critical VMware cluster node. BOSS for ESXi boot, front bays unused or populated as a vSAN cache tier. Primary VM storage on a shared SAN datastore. The R840's 4-socket compute drives high VM density per host with the storage layer abstracted by vSphere.
• SQL Server Enterprise consolidation host. BOSS for OS, front bays carrying RAID 1 SSD pairs for tempdb and RAID 10 SAS SSDs for log files. Primary SQL data on SAN. Eight bays is sufficient for SQL's local-disk patterns when primary data is networked.

Boot Drives

BOSS module for boot. Dual mirrored M.2 SATA SSDs on a dedicated PCIe card, hardware RAID 1, cold-swap. It keeps the OS off the front bays, frees all eight front bays for data storage, and provides hardware-mirrored boot redundancy without consuming a front bay or a RAID controller channel. On a platform where the 8-bay storage budget is already tight against scale-up workload patterns, dedicating two front bays to OS mirroring is wasteful. We include BOSS by default on R840 quotes unless you specify otherwise.

Storage Controllers

The R840 8-Bay supports the 14th gen PERC family:

• PERC H740P (8 GB NV cache, battery-backed). The production storage default for write-intensive or transactional workloads where local storage matters. Full hardware RAID 0/1/5/6/10/50/60. Right pick for SQL Server log files or Oracle redo logs staged on local SSD.
• PERC H730P (2 GB cache, battery-backed). A solid general-purpose choice for mixed or read-heavy workloads where the larger H740P cache is not load-bearing.
• PERC H330 (no cache, entry-tier hardware RAID). For light workloads where the storage layer is not a performance factor.
• HBA330 (pass-through HBA). For software-defined storage stacks (vSAN, Storage Spaces Direct, Ceph, ZFS). No hardware RAID; clean SAS pass-through. Right pick when the storage abstraction is the hypervisor or distributed file system, not the controller.
• S140 (software RAID via chipset). Acceptable for development and test only. We do not quote S140 for production data on a 4-socket platform.

The H740P NV cache is flash-backed rather than dependent on a battery wear item, which is one of the genuine 14th gen advantages over the 13th gen H730P lineage. For the full PERC controller reference shared across the 14th gen line, the R740 8-Bay 2.5" page covers the controller family in the dual-socket context.

Processors

Up to four 2nd Generation Intel Xeon Scalable (Cascade Lake-SP) processors in the LGA 3647 Purley platform. Up to 28 cores per CPU across four sockets is up to 112 cores and 224 threads maximum. TDP ranges from the Gold 5000 series through the Platinum 8000 series, roughly 85W to 205W per CPU. The 4-socket configuration uses Dell's CPU expansion design carrying sockets 3 and 4 plus their associated memory; production R840 deployments are almost always 4-socket, because a 2-socket build forfeits the platform's entire value proposition and the R740 does the same job for less.

CPU options we quote for production 4-socket builds:

• Gold 6230 (20 cores, 125W, DDR4-2933). The common production sweet spot: 80 cores total at 4-socket, a manageable thermal envelope, and balanced single-thread performance. Right pick for general 4-socket virtualization and database consolidation.
• Gold 6248 (20 cores, 150W, DDR4-2933). Higher base frequency than the 6230 at a higher TDP. 80 cores total. Good fit for Oracle and SQL Server where single-thread performance matters within the per-core licensing model.
• Platinum 8260 (24 cores, 165W, DDR4-2933). 96 cores total. Our standard maximum-performance specification when core count drives licensing economics.
• Platinum 8280 (28 cores, 205W, DDR4-2933). 112 cores total, the maximum core count for the platform. 4x 205W in a 2U chassis is thermally aggressive and requires confirmation of inlet temperature spec and PSU sizing. We validate thermal headroom on every 8280-class quote.

High-TDP quad-socket builds require high-performance heatsinks and specific airflow configurations. For CPUs in the upper TDP range, we strongly recommend the high-performance heatsink option to maintain stability under sustained 4-socket loads. All four sockets must carry the same processor SKU; mixed-SKU population is not supported.

Memory

96 DDR4 DIMM slots: 24 per CPU socket, six channels per socket at 2 DIMMs per channel. Maximum capacity is 6 TB with 64 GB LRDIMMs across all 96 slots. For SAP HANA and large in-memory database deployments, this single-chassis memory capacity is the primary justification for the R840's cost premium over dual-socket alternatives.

Memory speed follows standard Cascade Lake population rules: DDR4-2933 capable DIMMs run at full rated speed at 1 DIMM per channel on supported Gold and Platinum SKUs, and step down to DDR4-2666 at full 2 DPC population. This is the expected behavior, not a defect, and it is the right tradeoff for most workloads: the capacity gain from full population outweighs the one-bin speed reduction except on the most bandwidth-sensitive workloads. For maximum memory bandwidth on SAP HANA and similar bandwidth-bound workloads, populate at 1 DPC (48 DIMMs total, 12 per CPU) and accept the lower capacity ceiling. RDIMM and LRDIMM cannot be mixed, and balanced symmetric population across all four sockets is required for optimum performance.

Optane Persistent Memory is supported on the Cascade Lake L-series CPUs, which extends the effective memory ceiling well beyond the 6 TB LRDIMM limit for App Direct and Memory Mode deployments. This is the feature that lets a single R840 hold an in-memory dataset that previously required custom hardware. Confirm L-series CPU selection at quote time if PMem is part of the design.

Networking and PCIe Expansion

Dell Network Daughter Card (NDC) mezzanine for primary networking, the same architecture used across the 14th gen line. The NDC does not consume a PCIe expansion slot. NDC options include 4x 1 GbE, 2x 10 GbE plus 2x 1 GbE, 4x 10 GbE, and 2x 25 GbE, chosen by the network fabric the server connects into. Most R840 deployments standardize on 10 GbE or 25 GbE given the workload class.

PCIe expansion is up to 8 PCIe Gen3 slots with all four CPUs populated, with the exact slot map depending on riser configuration. The slot budget is what makes the R840 viable for scale-up workloads that also need substantial I/O: dual FC HBAs for redundant SAN connectivity, additional NICs for converged or storage networks, and a boot card all coexist. If your design needs more simultaneous high-bandwidth PCIe cards than the 2U R840 riser map can deliver, the 3U R940 8-Bay 2.5" provides more expansion slots within the same 4-socket platform family.

GPU Support

The R840 supports selective GPU acceleration rather than primary GPU compute. The 2U chassis and 4-socket thermal budget accommodate a limited number of single-width accelerators in the right riser configuration, suitable for inference, VDI acceleration, or analytics offload alongside the CPU workload. It is not a GPU training platform: if dense double-width GPU compute is the primary workload, a purpose-built GPU platform is the right answer rather than a 4-socket scale-up server. The honest framing is that GPUs in an R840 ride alongside its CPU and memory workload, they are not the reason to buy the chassis.

Management - iDRAC9 Generation

iDRAC9 Enterprise is the production management baseline, and on a 4-socket mission-critical platform it is rarely optional. It delivers remote KVM, virtual media mounting, predictive analytics, Active Health System telemetry, and full Lifecycle Controller firmware management with OpenManage Enterprise integration. We quote iDRAC9 Enterprise explicitly with any R840 build.

Silicon Root of Trust is standard: a hardware-anchored chain of trust that verifies iDRAC firmware, BIOS, and bootloader against cryptographic measurements. For SAP HANA, Oracle, and SQL Server deployments subject to compliance audit (SOC 2, PCI DSS, HIPAA, FedRAMP), Silicon Root of Trust provides documented platform-attestation evidence required in modern compliance frameworks. TPM 2.0 is supported and we recommend including it on every production build. NUMA topology visibility through iDRAC9 is meaningful on a 4-socket platform, where cross-socket memory access carries a latency penalty versus same-socket access; iDRAC9 surfaces the topology data and the workload (hypervisor NUMA scheduling, database affinity settings) does the actual placement. For SAP HANA and Oracle in particular, NUMA tuning is a standard part of production deployment.

Power and Cooling

Dell Flex Slot power supplies, with the R840 supporting 2 or 4 PSUs depending on configuration. A fully loaded R840 with 4x Gold 6230 (125W each), 96 DIMMs, and 8 SSDs draws roughly 1,200 to 1,600W at sustained peak. With 4x Platinum 8280 at 205W each, the draw rises toward 1,800W sustained. PSU sizing recommendations by configuration:

• 2x 1100W Platinum (lower-TDP 4-socket). Adequate for 4x 125W Gold configurations with modest memory and storage. Provides redundancy at the low end of the TDP range.
• 2x 1600W Platinum (typical production). The standard production redundant configuration for most R840 builds. Provides full 1+1 redundancy across common CPU configurations including 165W Platinum.
• 4x 1600W Platinum (maximum redundancy at high TDP). 2+2 redundancy for high-availability builds at 205W Platinum CPU configurations. Required when high availability and high TDP combine.

Power redundancy at 4-socket scale matters more than at dual-socket scale, because the workloads that justify an R840 (SAP HANA, Oracle, mission-critical SQL) are workloads where unplanned downtime carries documented cost. We recommend the redundant PSU configuration on every production R840 build. At 4-socket high TDP, the thermal envelope is real: confirm rack cooling and inlet temperature for the specific CPU SKU at quote time. We validate thermal and power configurations as part of every R840 quote.

Physical Specs and Platform Notes

• Form factor: 2U rack chassis, the same external height as the R640 and R740 but substantially more complex internally to carry four sockets and 96 DIMM slots. Plan chassis depth and cable management arm clearance into the rack layout.
• PCIe expansion: up to 8 PCIe Gen3 slots depending on riser configuration, with full-height and low-profile options across the riser map. The 4-socket population is what unlocks the full slot count.
• Parts availability: strong. 14th gen launched in 2018 and shares its processor and memory ecosystem with the high-volume R640 and R740, so CPUs, DIMMs, PERC controllers, PSUs, and fans are abundant on the new and refurbished market. Dell ProSupport remains available on the platform.
• Accessories we recommend: the BOSS boot card on every production build, and the Dell ReadyRails sliding rail kit for racking. The matching rail kit for this chassis is the Dell PowerEdge R840 2U B15 Sliding Ready Rail Kit, which we can include on the quote.
• Platform notes: all four sockets must carry identical CPU SKUs; full 96-DIMM population steps memory speed down one bin; high-TDP CPUs require the high-performance heatsink option; and the cable management arm consumes rear clearance worth confirming against rack depth before deployment.

Our Assessment

Where it excels: The R840 8-Bay is the right answer for genuine scale-up workloads in the 14th gen Dell family. SAP HANA scale-up appliances where the in-memory dataset has to fit in one server. Oracle Database Enterprise consolidation where per-core licensing economics favor fewer sockets carrying more cores. SQL Server Enterprise consolidation onto a single high-core-count license unit. Mission-critical virtualization at extreme VM density where one host with 112 cores and 6 TB of RAM replaces a rack of smaller nodes. These are the workloads where the 4-socket premium pays for itself.

Where to look instead: If the workload distributes cleanly across dual-socket nodes, the R740 8-Bay 2.5" or the R640 8-Bay 2.5" delivers the same work at materially lower cost. If you need large local storage alongside 4-socket compute, the R840 24-Bay 2.5" is the right configuration. If multiple high-bandwidth PCIe cards have to coexist with 4-socket compute, the 3U R940 8-Bay 2.5" adds the expansion the 2U chassis cannot. HPE shops evaluating the equivalent 4-socket 2U platform should look at the HPE ProLiant DL560 Gen10 8-Bay 2.5".

Bottom line: The R840 8-Bay is a purpose-built scale-up server, not a general-purpose enterprise box. Buy it when a specific workload genuinely needs four sockets, maximum single-chassis memory, or per-core licensing consolidation, and buy a dual-socket R740 for everything else. For the customer who has confirmed the scale-up requirement, the R840 8-Bay is the 14th gen Dell platform that delivers it in 2U at a refurbished price point well below the current-generation equivalent. This is the paragraph to put in the procurement justification: four sockets, up to 112 cores, up to 6 TB of memory, validated and burned in, under warranty.

Where the R840 Fits in 2026

The R840 is the 14th gen 4-socket platform, launched in 2018 on the Intel Purley architecture with Cascade Lake refresh in 2019. It sits two generations behind the 15th gen Ice Lake platforms and three behind the 16th gen Sapphire Rapids and Emerald Rapids platforms. Notably, Dell did not carry the 4-socket-in-2U envelope forward in the same form in later generations, which makes the R840 a distinctive answer for organizations that want 4-socket density in 2U specifically.

What is specific to the R840 in 2026: the platform is mature, the workloads it serves have not fundamentally changed (SAP HANA still scales up, Oracle licensing economics still favor fewer sockets with more cores, SQL Server Enterprise per-core licensing still rewards consolidation), and the per-core acquisition cost is meaningfully lower than the current generation for the same workload envelope. For organizations adding 4-socket capacity to existing 14th gen estates where standardization reduces operational complexity, the R840 8-Bay delivers genuine production work at significantly reduced cost. It is not the newest platform and it is not obsolete; it is the correct tool for a specific scale-up pattern when budget is a meaningful design constraint.

Honest Limitations

• 4-socket only makes sense for scale-up. A 2-socket R840 is technically supported but rarely the right call; if two sockets are sufficient, the R740 does the same job for less. Do not buy four sockets you will not use.
• Full 96-DIMM population drops memory speed one bin. DDR4-2933 capable DIMMs run at DDR4-2666 at full 2 DPC. For HANA or bandwidth-sensitive workloads, populate at 1 DPC (48 DIMMs) for full speed and accept the lower capacity ceiling.
• 8 SFF bays is not a storage-dense configuration. This chassis expects primary bulk data on SAN, NFS, or distributed storage. If you need high-density local SSD alongside 4-socket compute, move to the 24-Bay variant.
• High-TDP 4-socket thermals require validation. 4x 205W Platinum in a 2U chassis is thermally aggressive. Confirm inlet temperature spec, rack cooling capacity, and PDU sizing before deployment. We validate thermal configurations on every quote.
• Not a primary GPU compute platform. The PCIe slot map and thermal budget support selective acceleration, not dense GPU training. For GPU-first workloads, a purpose-built GPU platform is the right answer.
• 14th gen generational caveats apply. PCIe Gen3 rather than Gen4, a DDR4-2933 ceiling, and iDRAC9 rather than the newer management generation. These are expected for the platform's age and are not defects, but they are real if your requirement is current-generation I/O bandwidth.

Workload Fit

Where to Look Instead

• Dual-socket is sufficient? The R740 8-Bay 2.5" is the 2U dual-socket workhorse, and the R640 8-Bay 2.5" is the 1U dual-socket option. Both cost materially less than the R840 and handle most virtualization, database, and application workloads.
• Need high-density local storage with 4-socket compute? The R840 24-Bay 2.5" is the same platform with three times the SFF bay count.
• Need more PCIe expansion than 2U allows? The R940 8-Bay 2.5" is the 3U 4-socket platform with more slots, and the R940 24-Bay 2.5" combines maximum expansion with maximum SFF storage.
• HPE shop at the same 4-socket 2U tier? The HPE ProLiant DL560 Gen10 8-Bay 2.5" is the HPE counterpart, same generation and equivalent workload positioning.
• Racking the server? The R840 2U B15 Sliding Ready Rail Kit is the matching rail kit for this chassis.

Ready to Configure?

R840 configurations start with a design conversation. Tell us the workload (SAP HANA, Oracle, SQL Server, virtualization, or in-memory analytics), the licensing context (per-core, per-socket, or ULA), the CPU and core target, the memory target including any Optane Persistent Memory requirement, the storage configuration (local SSD pattern plus external SAN or NFS), PSU redundancy preference, PCIe expansion requirements, and quantity. We respond within 24 hours with a validated configuration including thermal and power-budget confirmation. Every refurbished unit ships with the Wholesale Servers 180-day warranty and 12+ hour burn-in testing, and volume pricing starts at 5 units. Call 1-800-778-1545 or use the quote form below.