Dell PowerEdge R640 Review — Gen 14 1U 10-NVMe HPC & Scale-Out Server
Purpose-Built 1U Platform for Dense Scale-Out and HPC Workloads
Dense Scale-Out Virtualization — Dual 28-core Xeon Scalable processors with 24 DIMM slots and up to 3 TB RAM pack maximum compute and memory density into a single rack unit for ESXi, Hyper-V, and KVM hypervisor hosts
High Performance Computing (HPC) — Low-latency Intel UPI inter-CPU fabric, up to 10 NVMe SSDs, and 48 PCIe Gen 3 lanes per processor deliver the throughput and memory bandwidth required by scientific simulation, financial modeling, and engineering analysis
Private Cloud Infrastructure — Dell EMC ReadyNode configurations pre-validated for VMware vSAN, Red Hat OpenShift, and Nutanix simplify hyper-converged deployments with known, tested Bill of Materials
Database Servers — 24 DDR4 channels, support for Intel Optane DC Persistent Memory up to 7.68 TB, and NVMe flash storage dramatically reduce OLTP latency and I/O wait for Oracle, SQL Server, and PostgreSQL
AI/ML Inference — Up to three NVIDIA T4 single-width GPUs in 1U plus Intel AVX-512 and Intel Deep Learning Boost in the CPU accelerate inferencing workloads without requiring a dedicated AI appliance
Software-Defined Storage (SDS) — Up to 12 hot-plug 2.5-inch drives plus optional rear bays and full NVMe support make the R640 an ideal Ceph, Lustre, or vSAN data node for scale-out storage clusters
Web Serving & Edge CDN — 1U form factor combined with iDRAC9 remote management enables deployment of 40+ R640 nodes in a standard 42U cabinet — ideal for content delivery and geographically distributed web infrastructure
Dual 2nd Generation Intel® Xeon® Scalable — Up to 28 Cores Per Socket
2nd Gen Xeon Scalable (Cascade Lake-SP) — Dual LGA 3647 sockets supporting Bronze, Silver, Gold, and Platinum Cascade Lake-SP SKUs up to Xeon Platinum 8280 (28C/2.7 GHz/205 W TDP) for workloads demanding maximum computational throughput
Up to 28 Cores Per Socket — 56 Total — The R640 supports the full Platinum tier, including the Xeon Platinum 8280 and 8276 in both standard and Low Power (L) variants; paired dual-socket configs deliver up to 56 cores and 112 threads in 1U
Up to 2933 MT/s Memory Bus — 2nd Gen Xeon Scalable unlocks DDR4-2933 speeds (vs 2666 MT/s on 1st Gen) with a single DIMM per channel, providing significant memory bandwidth increases for latency-sensitive analytics and in-memory database workloads
Intel C620 Series Chipset — Two-chip CPU + PCH platform with ACPI 4.0, PCIe 3.0, xHCI USB 3.0, Intel Active Management Technology 12.0, and Intel Trusted Execution Technology for enterprise security compliance
3 × Intel UPI Links at 10.4 GT/s — Gold and Platinum Gold/Platinum SKUs with one or two active UPI links at up to 11.2 GT/s on the R640; inter-socket coherent bandwidth far exceeds QPI from the R630 generation
Intel AVX-512 & Deep Learning Boost — AVX-512 FMA units accelerate floating-point compute and VNNI instructions accelerate INT8 neural network inference natively in-CPU for AI-assisted analytics workloads
Speed Select Technology (SST) — SST-PP (-Y SKUs) allows dynamic selection of core count, base frequency, and TDP profile to match real-time power constraints in capped-power rack environments
24-Slot DDR4 — Up to 7.68 TB with Intel® Optane™ Persistent Memory
24 DDR4 DIMM Slots — 12 slots per CPU organized into 6 channels × 2 DIMMs per channel; full 24-slot population requires dual-processor configuration; 6 channels per CPU represents a 50% increase over the R630's 4-channel E5-2600 memory architecture
Up to 3 TB LRDIMM — 24 × 128 GB LRDIMMs achieve 3 TB maximum; 1.53 TB achievable with 24 × 64 GB RDIMMs at full RDIMM capacity for performance-optimized configurations
Up to 7.68 TB with DCPMM — 12 × Intel Optane DC Persistent Memory 512 GB DIMMs combined with 12 × 64 GB LRDIMMs achieve 7.68 TB total addressable memory for in-memory database and persistent analytics deployments
Intel Optane DC Persistent Memory Modes — Memory Mode uses DRAM as a transparent DCPMM cache for any application without code changes; App Direct Mode exposes persistent DCPMM byte-addressable storage for PMem-aware databases and key-value stores
NVDIMM-N (Non-Volatile DIMM) — Up to 12 × 16 GB NVDIMMs (192 GB total) provide DRAM-speed access with battery-backed NAND persistence across system power loss events for write-intensive journal and log workloads
DDR4-2933 Performance — Single DIMM per channel configurations reach 2933 MT/s; dual-DIMM per channel runs at 2666 MT/s; both offer significant bandwidth improvements over the DDR4-2400 peak of the R630 generation
Advanced RAS Features — SDDC, Memory Mirroring, Single/Multi-Rank Sparing, Demand and Patrol Scrubbing, VMware Fault Resilient Memory (FRM) Zone, and Dell Memory Address Parity protection ensure data integrity across all operating modes
Up to 10 × 2.5-Inch NVMe SSDs — 76.8 TB Max Capacity in 1U
10 × 2.5-Inch Chassis (Mixed NVMe) — Up to 10 × 2.5-inch hot-swap drives supporting SAS, SATA, and up to 8 NVMe PCIe SSDs simultaneously; maximum raw capacity 76.8 TB with SAS/SATA/SSD mix
10 × 2.5-Inch All-NVMe Chassis — Dedicated 10-slot NVMe backplane supports 10 × CPU Direct Attach PCIe SSDs for maximum sequential throughput and minimum I/O latency; max 64 TB all-NVMe configuration
8 × 2.5-Inch SAS/SATA Chassis — Eight-bay configuration with optional ODD slot supports SAS, SATA, and NVMe drives for cost-optimized deployments balancing density and bay count in 1U
4 × 3.5-Inch SAS/SATA Chassis — Four-bay large-form-factor chassis for high-capacity SAS/SATA HDD deployments; max 64 TB with NL-SAS or SATA HDDs without sacrificing 1U form factor
2 Optional Rear Drive Bays — The 10-bay chassis supports an additional 2 × 2.5-inch rear-accessible SAS/SATA/NVMe drives for 12 total drives; max 15.36 TB in rear bays alone with high-density SSDs
CPU Direct-Attach PCIe NVMe — NVMe drives connect directly to the processor PCIe lanes without going through a controller or PCIe bridge, delivering predictable single-digit microsecond I/O latency for real-time databases and caches
NVMe + PERC Mixed Configurations — SAS/SATA drives behind a PERC hardware RAID controller can coexist with NVMe drives in CPU Direct Attach mode to simultaneously run structured storage and NVMe caching tiers in the same chassis
BOSS M.2 Module and Internal Dual SD Module for Dedicated OS Boot
BOSS Card (Boot Optimized Storage Subsystem) — Dedicated PCIe card hosts two M.2 SATA SSDs on an isolated RAID controller, completely separate from all data drive bays — ensures OS boot does not consume a hot-plug drive slot
240 GB or 480 GB M.2 SSDs — BOSS supports two 6 Gbps SATA M.2 drives, each in 240 GB or 480 GB capacity, for OS boot media; hardware RAID 1 mirrors both drives for automatic boot volume redundancy
HWRAID 1 Mirror — Both BOSS M.2 drives form a hardware RAID 1 volume controlled by the BOSS card's integrated controller; the OS sees a single logical RAID 1 disk with no software RAID stack required
Recommended for Full OS Deployments — BOSS is the recommended boot solution for Windows Server and Linux OS installs where a full operating system occupies the boot drive; keeps all 10–12 data bays free for workload data
IDSDM microSD for Hypervisor Boot — Internal Dual SD Module supports up to 2 × 16/32/64 GB microSD cards in hardware-mirrored IDSDM mode for VMware ESXi and other hypervisor-only boot images that fit on small flash media
vFlash Option — Separate 16 GB vFlash microSD card on the IDSDM module provides persistent vFlash remote file access and iDRAC Virtual Flash partition for OS deployment scripts and temporary ISO storage
Independent of Data Drive Bays — Whether BOSS or IDSDM is used for OS boot, all 10 front NVMe or SAS/SATA drive bays remain fully available for workload storage with no capacity sacrifice
PERC H740P, H730P, H330, HBA330, and S140 Software RAID
PERC H740P (Premium) — 12 Gbps hardware RAID controller with 8 GB NV cache supporting RAID 0/1/5/6/10/50/60; delivers peak IOPS for mixed SAS+NVMe workloads and large-scale virtualization storage arrays
PERC H730P (Value Performance) — 12 Gbps hardware RAID with 2 GB NV cache; proven choice for mixed HDD+SSD RAID arrays requiring hardware-accelerated parity calculations at mid-range cost
PERC H330 / HBA330 (Entry) — H330 provides RAID 0/1/5/6/10/50/60 without NV cache for entry-level RAID; HBA330 operates in full JBOD pass-through mode for software-defined storage or OS-managed RAID stacks without PERC overhead
H750 / HBA350i (Adapter-Only) — Add-in card versions of PERC H750 RAID and HBA350i SAS HBA available for configurations where the mini PERC slot is occupied or a second controller is needed alongside an existing controller
PERC S140 (Software RAID) — Intel Lewisburg chipset-based software RAID supports RAID 0/1/5/10 on SATA and NVMe drives without consuming a PCIe slot; ideal for small SATA RAID deployments or pure NVMe JBOD configurations
H840P / HBA355e (External) — External 12 Gbps SAS RAID (H840P) and HBA355e non-RAID pass-through adapter for connecting MD-series SAS expansion shelves, LTO tape libraries, and external JBOD enclosures
Dedicated Mini PERC Slot — The system board mini PERC slot is reserved exclusively for the internal RAID controller — it does not consume any of the three general-purpose PCIe x16 expansion slots available on the R640
Up to 3 × NVIDIA T4 Single-Width GPUs or 1 × FPGA in 1U
Up to 3 × NVIDIA T4 GPUs — Three single-width (low-profile) NVIDIA T4 GPU accelerators in 1U deliver up to 240 INT8 TOPS for AI inference, video transcoding, virtual desktop infrastructure (VDI), and real-time analytics workloads at unprecedented 1U density
NVIDIA T4 — 16 GB GDDR6 — Each NVIDIA T4 provides 16 GB GDDR6 memory, CUDA, Tensor Cores (1st Gen), and support for multi-instance GPU (MIG) partitioning in newer firmware for running multiple isolated inference tenants per card
70 W TDP Per GPU — At 70 W TDP the NVIDIA T4 operates within standard 1U power and thermal constraints, enabling three-GPU configurations without requiring additional liquid cooling or specialty rack equipment
GPU Direct Storage — NVMe SSD data paths can be configured for NVIDIA GPUDirect Storage, bypassing CPU DRAM for direct GPU-to-NVMe DMA transfer to accelerate data pipelines for deep learning training and inference preprocessing
Up to 1 × FPGA Accelerator — Field-Programmable Gate Array cards supported via a PCIe x16 expansion slot; FPGA-based acceleration suits network packet processing, compression, encryption offload, and custom algorithm inference
PCIe Gen 3 x16 GPU Slots — Multiple low-profile PCIe Gen 3 x16 slots with 16 GT/s bandwidth serve GPU or FPGA cards; CPU-direct attachment ensures no bandwidth bottleneck on the PCIe fabric for high-throughput accelerator workloads
Chassis Thermal Optimization — Dell EMC's 1U chassis airflow is engineered to cool up to three T4 GPUs alongside two high-TDP Xeon CPUs using the standard hot-plug fan configuration without third-party cooling accessories
Up to 3 × Full x16 PCIe Gen3 Slots Across Four Riser Configurations
Riser 1A — Dual LP x16 (Processor 1) — Two low-profile/half-length PCIe Gen 3 x16 slots (Slot 1 and Slot 2) connected to Processor 1 for maximum 1U expansion density when full-height cards are not required
Riser 1B — Single LP x16 (Processor 1) — One low-profile PCIe Gen 3 x16 slot connected to Processor 1 when the second-card slot is traded for the PCIe SSD rear-drive access bay configuration
Riser 2A — Single LP x16 (Processor 2) — One low-profile PCIe Gen 3 x16 slot connected directly to Processor 2 for dual-CPU workloads where NUMA locality of expansion cards to their respective CPUs reduces latency
Riser 2B — Full-Height 3/4-Length x16 (Processor 2) — One full-height, three-quarter-length PCIe Gen 3 x16 slot connected to Processor 2 for standard full-height PCIe cards such as dual-port 25GbE NICs, HBAs, or FPGA accelerators
Dedicated NDC/OCP Slot — A dedicated PCIe x8 Gen 3 slot is reserved for the network daughter card (NDC) — accommodates the selected NIC option without consuming any general-purpose expansion slot
Dedicated Mini PERC Slot — An additional dedicated PCIe x8 Gen 3 slot is reserved for the internal PERC RAID controller; three general-purpose slots remain free regardless of RAID controller installed
Symmetrical x16 on All Slots — Unlike the R630 which mixed x16/x8/x16 PCIe link widths, the R640 provides PCIe x16 links on all three general-purpose riser slots, eliminating bandwidth asymmetry across card positions
NDC Up to 2 × 25GbE — Isolated iDRAC9 Dedicated Management Port
4 × 1GbE NDC — Four-port Gigabit Ethernet network daughter card for standard host network connectivity; all four ports independent of the iDRAC9 management NIC, providing maximum uplink redundancy at entry-level bandwidth
2 × 10GbE + 2 × 1GbE NDC — Mixed-speed NDC with two 10GBASE-T or SFP+ ports for storage/VM traffic plus two 1GbE ports for dedicated management or heartbeat network segments
4 × 10GbE NDC — Quad 10GbE NDC for high-density hypervisor hosts running vSAN, NFS, or iSCSI storage traffic alongside virtual machine network traffic without consuming a PCIe expansion slot
2 × 25GbE NDC — Dual 25 Gigabit Ethernet NDC for maximum bandwidth per port; pairs with 25GbE top-of-rack switches for spine-leaf fabric designs in HPC and high-frequency trading environments
iDRAC9 Dedicated 1GbE Management Port — Completely isolated out-of-band management NIC that operates independently of all host NDC ports — management access is preserved even when all data NICs are saturated or the OS is down
InfiniBand Option — Mellanox ConnectX InfiniBand adapters installable in PCIe Gen 3 expansion slots for RDMA-capable HPC cluster fabric with EDR InfiniBand at 100 Gb/s per port for MPI message-passing workloads
NPAR / NIC Partitioning — 10GbE and 25GbE NDC options support NIC partitioning (NPAR) allowing virtualization of the physical NIC into multiple logical partitions with bandwidth guarantees for traffic isolation in mixed-workload environments
Hot-Plug Redundant PSUs — 495 W to 1600 W Platinum and Titanium
1+1 Hot-Plug Redundancy — Two hot-swappable PSU bays with 1+1 redundancy; a failed PSU is replaceable under full system load without interrupting any running workload or hypervisor instance
495 W Platinum — Right-sized for single-socket Silver or Bronze configurations with light storage and no GPU accelerators; 1908 BTU/hr heat dissipation for the most thermally constrained rack environments
750 W Platinum & Titanium — Dual-socket Gold configurations with up to 8 drives; 750 W Titanium variant achieves higher efficiency rating for facilities targeting ENERGY STAR or 80 PLUS Titanium power contracts
1100 W Platinum — Handles fully-populated dual-socket Platinum configurations with 10 drives and PCIe expansion cards; 380 V HVDC variant available for China and Japan DC-bus rack deployments at 4100 BTU/hr
1600 W Platinum — Maximum PSU for three NVIDIA T4 GPU configurations, high-TDP Xeon Platinum dual-socket configs, and fully loaded NVMe + PCIe card systems; 6000 BTU/hr heat dissipation rating
DC Power Options — 1100 W 48 VDC PSU for data centers running –48 V telecom DC bus; 750 W Mixed Mode DC and HVDC variants for China-specific power infrastructure configurations
1% Power Monitoring Accuracy — iDRAC9 + Node Manager power monitoring reports actual watt draw at 1% accuracy versus the industry standard 5%, enabling fine-grained power capping and rack PDU circuit-breaker protection
Up to 8 Hot-Plug Fans — Intelligent Cooling Across 10–40°C Ambient Range
Up to 8 Hot-Plug Fans — Doubled fan count versus the R630's seven fans; redundant N+1 fan configuration ensures continuous operation through a single fan fault without requiring a service call
1×6 Fan Module Architecture — Each fan unit is a single replaceable module containing six fan blades; the compact 1×6 design reduces air turbulence and acoustic noise compared to the R630's 2×3 paired modules
Open-Loop + Closed-Loop Hybrid Control — Pre-computed fan speed tables (open-loop) initialize cooling based on system BOM; real-time closed-loop sensor feedback from CPUs, DIMMs, PCH, GPU cards, and inlet air adjusts dynamically
Standard 10–35°C / Extended 5–40°C — Standard operating range 10–35°C; expanded Fresh Air Cooling mode extends continuous operation to 5–40°C ambient with configuration-specific limits and de-rating above 35°C
GPU Thermal Support — Chassis airflow validated for three NVIDIA T4 cards alongside dual high-TDP CPUs; internal component placement optimized to prevent GPU thermal throttling under sustained AI inference loads
Data Center Audio Awareness — Dell EMC acoustic design targets tone prominence thresholds in addition to raw dBA levels; fan speed algorithms balance thermal performance with psychoacoustical comfort metrics for open-floor-plan data centers
Dell Active Power Controller (DAPC) — Default BIOS thermal profile balances workload performance with fan and CPU power; Performance Optimized and Maximum Performance profiles available for latency-critical workloads
Front and Rear Port Layout for Console, Management, and Data Access
Front USB 2.0 — 1 × USB 2.0 port on the right control panel for local installation media, USB keys, and field service tools without removing the 1U chassis from the rack
Front USB 3.0 (Optional) — An optional front USB 3.0 port can be added via a USB module cable connected to the system board internal USB port; not available on the 4×3.5-inch or 10×2.5-inch storage configurations
Front iDRAC Direct (Micro-AB USB) — Dedicated Micro-AB USB port with LED indicator for direct laptop or tablet connection to iDRAC9 local management without requiring network access during installation or troubleshooting
Front VGA — 1 × VGA port on the right bezel panel for monitor connection during BIOS configuration, POST diagnostics, and emergency OS console access at the rack without routing cables to the rear
Rear USB 3.0 × 2 — Two USB 3.0 (5 Gbps) ports on the rear panel for persistent peripheral devices, external USB storage, OS installation media, or service keyboard attachment
Rear VGA & Serial — 1 × VGA and 1 × DB-9 serial port on the rear panel for headless remote console, legacy serial management devices, and BIOS access via serial console redirect through iDRAC9
Optional LCD Bezel — LCD bezel option embeds a status display and navigation buttons into the front bezel for reading system ID, IP address, iDRAC status, and alert summaries directly at the server without a laptop
Cyber Resilient Architecture — Silicon Root of Trust Through System Erase
Silicon Root of Trust — Cryptographic fingerprint burned into iDRAC9 silicon at the factory validates every firmware image from power-on before any CPU instruction executes; hardware-anchored trust chain cannot be spoofed in software
Cryptographically Signed Firmware — Dell-issued certificate on every firmware package; Lifecycle Controller rejects unsigned or modified firmware at install time, protecting against supply-chain firmware injection attacks
Secure Boot — UEFI Secure Boot enforces OS bootloader signature verification; prevents rootkits, malware, and unauthorized OS images from executing before the operating system security stack is loaded
TPM 1.2 / 2.0 / TCM 2.0 — Plug-in TPM or TCM chip (TCM for China compliance) provides hardware-rooted key storage, platform attestation, Intel TXT, and hardware-backed encryption key escrow for BitLocker and vTPM
Server Lockdown — OpenManage Enterprise-enforced lockdown prevents all firmware, BIOS, and hardware configuration changes from any interface (iDRAC, RACADM, BIOS) until explicitly unlocked by an authorized admin
System Erase — NIST 800-88-compliant cryptographic and physical erase of all persistent storage — HDDs, SSDs, NAND, NVDIMM flash, and optionally system memory — for secure server decommissioning or hardware transfer
Power-Off Security & Cover Latch — BIOS configurable power-button disable prevents unauthorized power-off; integrated tooled cover latch and optional front-panel security bezel with lock restrict physical drive and component access
iDRAC9 with Lifecycle Controller, RESTful Redfish API, and Quick Sync 2
iDRAC9 Embedded Controller — Dedicated management processor with its own power rail, NIC, and firmware runs fully out-of-band from the host OS for 24×7 inventory, alerting, power monitoring, and remote console access even with the server off
Lifecycle Controller 3.0 — Agent-free provisioning, OS deployment, firmware updates, hardware configuration, and system log collection via iDRAC web UI or RACADM CLI without needing an installed operating system
iDRAC RESTful API with Redfish — DMTF Redfish 1.0 standards-based JSON REST API enables full server lifecycle automation from Ansible Playbooks, Python scripts, Terraform providers, and ServiceNow workflows
Quick Sync 2 BLE/Wi-Fi (Optional) — Bluetooth Low Energy and Wi-Fi bezel module allows iDRAC9 inventory, RACADM commands, and firmware update triggers from a mobile device using Dell OpenManage Mobile app at the rack front
OpenManage Enterprise Console — Single-pane fleet management console for automated discovery, firmware drift alerts, policy-based configuration push, power usage dashboards, and warranty tracking across all R640 nodes
Integrations — VMware, Ansible, Microsoft, ServiceNow — OMIVV for vCenter manages R640 health and firmware from within the vSphere web client; OpenManage Ansible Modules enable Infrastructure-as-Code server provisioning for DevOps teams
SupportAssist Embedded — Automated case creation with Dell Support upon failure detection; predictive storage, fan, and DIMM failure alerts reduce mean time to repair with AI-based health scoring before downtime occurs
Windows, Linux, VMware ESXi, Oracle Linux, and Citrix XenServer Certified
Windows Server 2019 / 2016 with Hyper-V — Certified for Windows Server Standard, Datacenter, and Essentials LTSC editions; Hyper-V host with Windows Admin Center provides browser-based cluster and storage management
VMware ESXi — Full VMware Hardware Compatibility Guide (HCG) certified; integrates with OpenManage Integration for VMware vCenter (OMIVV) for automated health, firmware baseline compliance, and lifecycle management from vSphere
Red Hat Enterprise Linux — RHEL certification for long-term enterprise Linux workloads including OpenShift Container Platform hosting; iDRAC Service Module (iSM) provides host-to-iDRAC communication for coordinated health reporting
SUSE Linux Enterprise Server — SLES certified for SAP HANA scale-up deployments (with DCPMM configurations achieving multi-terabyte in-memory SAP HANA instances in 1U), file/print, and SUSE-standardized enterprise infrastructure
Canonical Ubuntu Server LTS — Ubuntu LTS support enables cloud-native Kubernetes nodes, OpenStack compute, and developer-facing workloads on certified bare-metal hardware with long-term kernel and security update guarantees
Oracle Linux — Oracle Linux certification for Oracle Database, Oracle RAC, and middleware workloads requiring Oracle Unbreakable Enterprise Kernel (UEK) on certified PowerEdge hardware with vendor co-support
Citrix XenServer / Hypervisor — Citrix Hypervisor supported for virtual application delivery, virtual desktop infrastructure, and multi-tenant hosted cloud platform deployments on 1U dense compute nodes
ReadyRails II Sliding and ReadyRails Static for Universal 4-Post and Telco Racks
ReadyRails II Sliding Rails — Support full-extension of the R640 out of the rack for in-place servicing of DIMMs, drives, PCIe cards, and fans without removing the 1U chassis to a bench; critical for dense racks where adjacent servers block access
Tool-Less 4-Post Installation — Drop-in installation into 19-inch EIA-310-E compliant square or unthreaded round-hole 4-post racks with no cage nuts or tools; sliding rail front adjustability range 630–883 mm (square), 616–876 mm (round)
ReadyRails Static Rails — Broader rack compatibility and smaller mounting footprint than sliding rails; support 19-inch square, round, and threaded 4-post racks, plus 2-post Telco center-mount and flush-mount configurations
Threaded Rack Mounting — Both sliding and static rails support tooled installation in threaded-hole 4-post racks; screws not included (thread sizes vary by rack vendor) — static rails offer greater adjustability range in threaded configurations
Telco 2-Post Support — ReadyRails Static B4 supports both center-mount and flush-mount configurations in 2-post Telco racks (not supported by sliding rails B6) for telecom, edge, and co-location deployments with non-standard rail systems
Optional Cable Management Arm — Sliding rails support a CMA for organized rear cable bundles and server extension without cable disconnection; minimum rail depth without CMA is 714 mm for sliding, 622 mm for static
1U Chassis Profile — 42.8 mm (1.69 in.) height occupies exactly one rack unit; 808.5 mm (31.8 in.) depth fits all standard 1000 mm-depth rack enclosures with adequate rear clearance for cabling and PSU pull
R640 vs R630 — More Memory, More NVMe, GPU Support, and iDRAC9
2nd Gen Xeon Scalable vs E5-2600 v4 — Cascade Lake-SP replaces Broadwell-EP; comparable core counts with significantly higher per-core IPC, 50% more memory bandwidth (6-channel vs 4-channel), AVX-512, and Intel Deep Learning Boost unavailable on v4
24 DIMM Slots — Same Count, Far Higher Capacity — Both platforms share 24 DIMM slots, but the R640 supports 128 GB LRDIMMs, DCPMM up to 512 GB/DIMM, and NVDIMM-N — technologies not supported on R630 — for radical memory density improvements
3× PCIe Gen3 x16 vs Mixed x16/x8/x16 — R640 provides symmetrical x16 link width on all three general-purpose slots; R630's center slot was limited to x8, creating a bandwidth bottleneck for storage and GPU cards placed in the center position
Up to 10 NVMe Drives vs 4 NVMe — R640 supports a dedicated 10-NVMe backplane for all-flash NVMe; R630 was limited to 4 PCIe SSDs via a PCIe bridge card — a 2.5× NVMe density improvement in the same 1U form factor
GPU Support (New) — R640 adds up to 3 NVIDIA T4 single-width GPUs; the R630 had no GPU accelerator option in its 1U chassis, making the R640 a fundamentally different platform for AI inference and VDI workloads
iDRAC9 vs iDRAC8 — iDRAC9 adds Silicon Root of Trust, Redfish REST API, Quick Sync 2 wireless management, Server Lockdown, System Erase, and DCPMM health monitoring — management capabilities unavailable on R630's iDRAC8
BOSS M.2 Boot Module (New) — R630 had no BOSS equivalent; R640's BOSS card provides dedicated 240/480 GB M.2 SATA RAID 1 OS boot, preserving all 10–12 data drive bays entirely for workload storage
| Feature | R630 (Gen 13) | R640 (Gen 14) |
|---|---|---|
| Processor Family | Xeon E5-2600 v3 / v4 | 2nd Gen Xeon Scalable (Cascade Lake) |
| Chipset | Intel C610 | Intel C620 |
| DIMM Slots | 24 slots | 24 slots + DCPMM / NVDIMM |
| Max Memory Speed | Up to 2400 MT/s | Up to 2933 MT/s |
| Max NVMe Drives | 4 via PCIe bridge | 10 CPU Direct Attach |
| PCIe Slot Widths | x16 / x8 / x16 (mixed) | x16 / x16 / x16 (all equal) |
| GPU Accelerators | Not supported | Up to 3× NVIDIA T4 or 1× FPGA |
| Boot Storage | Not supported | BOSS M.2 240/480 GB RAID 1 |
| Intel Optane DCPMM | Not supported | Up to 12 × 512 GB DIMMs |
| Remote Management | iDRAC8 | iDRAC9 |
ProSupport Plus with Automated SupportAssist for Maximum R640 Uptime
ProSupport Plus — Dell's highest-tier service plan with 24×7 automated SupportAssist monitoring, predictive hardware failure alerts, and priority access to senior engineers for critical P1 incidents affecting production R640 workloads
SupportAssist for PowerEdge — Automated diagnostics and case creation reduce IT effort by up to 72% per Dell research; predictive failure scoring identifies fan, drive, DIMM, and PSU degradation before unplanned outage events
ProSupport — 24×7 phone support and next-business-day or next-day on-site hardware repair with priority part dispatch for R640 environments requiring human-escalated issue resolution with defined SLA response windows
ProSupport One for Data Center — Single point of contact support contract covers the entire data center infrastructure stack — R640 servers, Dell EMC storage, and networking — under one unified agreement with dedicated account team
ProDeploy Enterprise Suite — Managed deployment services ranging from Basic Deployment (hardware rack-and-stack) to ProDeploy Plus (OS install, firmware baseline, BIOS configuration, and application configuration) for R640 clusters
Residency Services — Embedded Dell engineers available on-site for data migration planning, vSAN ReadyNode cluster buildout, DCPMM App Direct Mode configuration, and knowledge transfer to internal IT operations teams
Dell Education Services — Instructor-led and self-paced online training covering iDRAC9, Lifecycle Controller, OpenManage Enterprise fleet management, and DCPMM configuration for teams deploying Intel Optane-enabled R640 clusters
Frequently Asked Questions — Dell PowerEdge R640
The Dell PowerEdge R640 supports up to 3 TB of DDR4 LRDIMM RAM across 24 DIMM slots (12 per CPU) at speeds up to 2933 MT/s. With DDR4 RDIMMs the maximum is 1.53 TB. Uniquely among Gen 14 1U platforms, the R640 also supports Intel Optane DC Persistent Memory (DCPMM) in up to 12 DIMM slots, reaching 7.68 TB total addressable memory when paired with 12 LRDIMMs. Up to 12 NVDIMM-N modules (192 GB total) are also supported for persistent memory use cases requiring battery-backed data retention. Configure your R640 memory at ECS.
The Dell PowerEdge R640 supports up to 10 CPU Direct-Attach NVMe PCIe SSDs in the all-NVMe 10-bay chassis configuration, delivering up to 64 TB all-NVMe capacity. In the mixed SAS/SATA/NVMe 10-bay configuration, up to 8 of the 10 bays can be NVMe with the remainder as SAS or SATA drives, achieving up to 76.8 TB mixed capacity. The NVMe drives connect directly to processor PCIe lanes — not through a controller bridge — for consistent sub-100 microsecond I/O latency. A dedicated 10-NVMe backplane is required for all-NVMe deployments.
Yes. The Dell PowerEdge R640 supports up to 3 × NVIDIA T4 single-width GPU accelerators in 1U — or up to 1 × FPGA accelerator card. The NVIDIA T4 is a 70 W low-profile card providing 16 GB GDDR6 memory, Tensor Core AI inference acceleration, and NVIDIA vGPU support for virtual desktop infrastructure. Three T4 GPUs deliver up to 240 INT8 TOPS of inference throughput in a single 1U chassis. The R640 is the only Gen 14 1U general-purpose platform with validated multi-GPU support. For more information, visit Dell.com/GPU for the latest accelerator compatibility details.
Yes. Express Computer Systems stocks professionally reconditioned refurbished Dell PowerEdge R640 servers tested, cleaned, and configured to your exact processor, memory, and storage specifications — ready to deploy for HPC, virtualization, NVMe storage, or AI inference workloads at significant cost savings versus new gear. Shop refurbished Dell R640 servers at ECS.
The Dell PowerEdge R640 (Gen 14) succeeds the R630 (Gen 13) with 2nd Gen Xeon Scalable Cascade Lake-SP processors replacing the E5-2600 v3/v4 family with 50% more DDR4 memory bandwidth (6 channels vs 4), support for Intel Optane DC Persistent Memory and NVDIMM-N not available on R630, up to 10 CPU direct-attach NVMe SSDs versus 4 via PCIe bridge on R630, symmetrical PCIe Gen 3 x16 on all three expansion slots removing the R630’s x8 bottleneck in the center slot, GPU accelerator support for up to 3 NVIDIA T4 cards absent from R630, iDRAC9 replacing iDRAC8 with Silicon Root of Trust and Redfish RESTful API, and BOSS M.2 boot module — all in the same 1U rack chassis footprint.
Ready to Deploy the Dell PowerEdge R640?
Express Computer Systems offers professionally reconditioned Dell PowerEdge R640 servers configured to your exact specifications. Our expert team tests every unit and backs every order with our quality guarantee.
Start building your custom server today