Dell PowerEdge T640 8-Bay 3.5"

Configure Your System:

Currently Configured: $3,587.70
1 Processor Required
Series
Category
CPU (2x included in build)
Component Guide Processor. This step works in two passes. Pick a processor series first, which narrows the list, then choose the CPU that fits your workload and licensing.

Step 1: Choose a series

Both generations drop into the same T640 motherboard, so this is a price and features decision, not a compatibility one.

Intel Xeon Gen 1 (Skylake-SP, 4100/5100/6100/8100 series) is the value play. Mature, plentiful, and the best price per core in the lineup.

Intel Xeon Gen 2 (Cascade Lake-SP, 4200/5200/6200/8200 series) adds higher clocks per tier and 2933 MT/s memory support on Gold 6200 SKUs. Worth the premium when per-core performance is load-bearing; otherwise Gen 1 does the same job for less.

Step 2: Choose your CPU

The T640 is a rack server in a tower suit: it carries R740-class compute into buildings without a rack, and the CPU advice follows the workload rather than the form factor.

Running Windows Server 2022 or 2025? Two 8-core Gold CPUs land exactly on the 16-core base license; go bigger and you are buying extra core packs before the server performs any differently. The Gold 6134 (8 cores, 3.2 GHz) is the pick where per-core performance matters.

Virtualizing, or feeding GPUs? The Gold 6230 (20 cores, 2.1 GHz) is the balanced default, and VMware's 16-core-per-socket license minimum makes the 8-core strategy pointless on vSphere anyway. GPU compute workloads mostly want the accelerators fed, which is a core-count and memory-bandwidth job, not a clock-speed one.

Populate both sockets for any serious build: a single-CPU T640 loses half its memory slots and the PCIe slots that route through CPU2, and on the tower people buy for GPUs, those slots are the point.

Fair warning on any CPU above 150W: the performance heatsink and fan kit are required, not optional. Confirm they are in your build before checkout.
2 Heat Sink Optional
Component Guide Heat Sink. The T640 ships with a standard heatsink or a performance heatsink, and the tower format works in your favor: a chassis this size moves air with big, slow fans, which is why a T640 can carry serious compute in a room where people work.

The standard heatsink covers CPUs up to 150W TDP in a normal environment.

The performance heatsink is required above 150W, and we also spec it in two cases: any GPU configuration, where the accelerators compete for the same airflow, and warm environments; intake air above 30°C tightens the math, and server closets without dedicated cooling hit that number more often than people expect.

The T640 is the tower people put GPUs in, so plan accordingly: if a GPU is in the build or on the roadmap, take the performance heatsinks now. The cost difference is small, and it keeps the fans slow, which in a shared space is a feature you will hear the absence of.
3 Memory (RAM) 24 DIMM slots, modules added in sets of 2 Required
Total Installed Memory
RAM Clock Speed
RAM Configuration
Component Guide Memory. The T640 has 24 DDR4 DIMM slots, 12 per CPU, across 6 channels per socket at 2 DIMMs per channel. Fill all 24. This is our standing recommendation and we push it hard.

There are two benefits. The first is performance: full population at 2 DPC keeps all twelve memory channels working at maximum depth, and on memory-bandwidth-sensitive workloads like virtualization and GPU-fed compute it consistently outperforms a partial population running at higher clock. The second is cost per gigabyte: 24 smaller modules hit the same capacity for meaningfully less than 12 larger ones, and 768 GB as 24× 32 GB is the best cost per GB in the lineup.

We know the instinct is to leave slots open for a later upgrade, and on paper that is sensible. In practice it rarely plays out well. One year from now you are hunting for DIMMs that match the rank and speed of what is already in the box, and a mismatch either clocks the whole server down or refuses to train. The server is in production by then, so adding memory means downtime everyone notices.

Buy the capacity now, in one matched set. If you outgrow it, the right move is usually another node rather than another DIMM.

384 GB (24× 16 GB) for file, application, and light virtualization duty
768 GB (24× 32 GB) is our default for a serious virtualization host or GPU compute box
1.5 TB (24× 64 GB) for dense consolidation or in-memory workloads

Past 1.5 TB you are into 128 GB LRDIMMs, which reach 3 TB but cost substantially more per GB. Most buyers are better served by a second host.
4 RAID Controllers Optional
Dell 14th Gen PCIe
Component Guide RAID. The controller decision comes down to one question: how much does local write performance matter? The cache size is what you are paying for, and the T640's bay count raises the stakes: the more bays you fill, the more work the controller does.

PERC H740P (8 GB NV cache, battery-backed) is the production storage pick, and on a heavily populated T640 it moves from option to default. Write-intensive or transactional workloads, SQL Server, heavy logging, or any array past 8 drives: the 8 GB non-volatile cache earns its price.

PERC H730P (2 GB cache, battery-backed) covers mixed and read-heavy workloads well on smaller arrays, and it is the most common spec on builds where local storage is a supporting act rather than the star.

PERC H330 (no cache) is entry-tier hardware RAID for light duty. Fine for a domain controller or file server; wrong for anything transactional, and wrong in front of a full backplane.

S140 software RAID is dev/test only, and SATA drives only. Not a production recommendation.

Mind the RAID math whatever you pick: RAID 1 or 10 for OS and transactional volumes, RAID 6 over RAID 5 for large-capacity arrays, consistent drive models within each array, and a hot spare if this tower is the only server in the building.

And if this T640 is joining a vSAN, Storage Spaces Direct, or Ceph cluster, you do not want hardware RAID at all; those stacks need direct drive access through an HBA330. Call us at 1-800-778-1545 and we will quote it.
5 Storage Drives Select up to 8 drives (0/8 Slots Used) Optional
Recommended Drives
Show New SAS HDDs 3.5"
Show Refurbished SAS HDDs 3.5"
Show Refurbished SATA HDDs 3.5"
Show New SAS SSDs 2.5"
Show New SATA SSDs 2.5"

Drive trays are included and matched to your chassis automatically

Component Guide Drive. Drives are the number one failure point on any server, and this configuration exists to hold them in large format.

This is the T640 where spinning disk is the point. Eight 3.5" bays exist for capacity per dollar: file archives, backup targets, media, surveillance retention, and every workload where terabytes matter more than IOPS. Large-capacity 3.5" drives deliver cost per terabyte that no SSD matches. If the workload is random-IO-heavy instead (a real database, dense virtualization with local storage), the 16-Bay 2.5" T640 with SSDs is the better tool; eight spindles cannot deliver those IOPS. A practical middle path we quote often: a pair of SSDs for the OS and hot data, spinning capacity behind them.

Buy new. We steer you away from refurbished on drives specifically. Rebuilds on large-capacity drives run long, and a rebuild is exactly when you do not want a second drive with unknown hours in the array.

The RAID math is non-negotiable on big spinning drives: RAID 6 over RAID 5, no exceptions, plus a hot spare if this tower is the only server in the building; nobody is driving a replacement drive out on Saturday night.

Keep the OS off your bays: a Dell BOSS card, available in the Add Ons section below, gives you mirrored M.2 boot drives and leaves all eight bays for the capacity you bought this chassis to hold.
6 Remote Access Required
Component Guide Remote Access. The T640 ships with iDRAC9 either way; the license determines what you can do with it remotely.

The dividing line is the remote console. iDRAC9 Enterprise gives you full remote KVM and virtual media, meaning you can reach the server's screen, mount an ISO, and rebuild an OS from anywhere, even when the operating system is down. Express does not include the remote console, so any problem below the OS level means someone physically at the machine.

If this T640 sits down the hall and someone on site can reach it without ceremony, Express is a legitimate answer. The case for Enterprise is who actually fixes the server: if that person is an outside IT provider or a head-office admin who is not in the building, Enterprise turns every below-the-OS problem from a site visit into a remote session, and one avoided trip pays for the license. A T640 also tends to carry more weight than a typical tower (GPU compute, a serious virtualization stack, an office's entire storage), and the more the box matters, the shorter the Enterprise conversation gets. It also adds automated firmware updates, two-factor authentication, and secure erase.

Deployments of 5 or more units should take Enterprise on all of them; fleet management without remote console access does not work in practice.
7 Power Supply 2x included in build Required
Component Guide Power Supply. Every option here is a dual hot-swap redundant pair, so redundancy and serviceability are already handled. The only decision is wattage, and the T640's range is wide because the chassis is: the same tower carries everything from quiet office duty to multi-GPU compute.

The 750W pair is our default for balanced builds: dual Gold CPUs, full memory, and a backplane of SSDs sit comfortably inside it at the 50 percent efficiency sweet spot.

The 1100W pair is where top-bin builds belong: high-wattage CPUs with every bay populated, or any single-GPU configuration.

The 1600W pair and above is GPU territory. The T640 is the tower people buy to put accelerators in, and multi-GPU builds need the headroom; we spec the top tiers on any build with two or more cards.

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; the price step is small. Watch the estimated TDP counter at the bottom of the page as you add components to see where your build lands.

Two tower-specific notes. If a GPU might be added later, buy the higher-wattage pair now; the swap is painless but the second shipment is not free. And since a tower usually plugs into wall power rather than conditioned rack feeds, put the two cords on different circuits where the building allows it, with at least one on a UPS.
8 Network Cards Required
Component Guide Network Card. Two decisions here: speed and port type, and the tower deployment picture shapes both.

Speed first. Quad 1 GbE (Intel i350 or Broadcom 5720) covers a T640 doing straightforward office duty over a 1 Gb switch. But a T640 is usually bought to do more than straightforward office duty, and a virtualization host, backup target, or GPU compute box will saturate 1 GbE quickly: the 2x 10 Gb + 2x 1 Gb combo cards are the sweet spot there, with 10 Gb carrying the data and 1 Gb left for management. Quad 10 Gb (Intel X710, X550) is the right architecture when this tower anchors a small cluster.

Then port type, and in an office this decision is usually made for you. Most office and small-business switches are copper RJ45, which means the -T models (X540, X550, 10GBASE-T) are the natural fit; they run on the Cat6 cabling the building already has. SFP+ ports take fiber or DAC cables and belong with datacenter-grade switches; order SFP+ against a copper-only office switch and the server arrives with nothing to plug into. Check the switch before you check this box.

Fair warning that applies to every option here: a 10 Gb card does not deliver 10 Gb unless the switch, cabling, and the device on the other end keep up. In an office, the switch is usually the first thing to check.

Need additional network cards? They are in the Add Ons section below, or call 1-800-778-1545 and we will spec the whole path.
9 Operating System Required
Operating System
Component Guide Operating System. We install, license, and burn-in test your OS before shipping, so the server arrives ready to join the network, which matters double when there is no IT department on site to finish the job.

Windows Server 2022 is the version to buy today; 2019 belongs on a new build only when an application vendor requires it. Standard licensing is per core with a 16-core base, which is why our processor guidance steers Windows buyers to two 8-core CPUs: the 16-Core license covers that build exactly. Past 16 total cores, you need the 24-Core option. On CALs, count your users and round up.

Fair warning on Evaluation Editions: free and fully functional for 180 days, then they expire. Right for labs, wrong for production. 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.

Proxmox VE is what we increasingly quote for buyers done with VMware licensing costs: open-source KVM virtualization with clustering and live migration, no per-core math, and first-class GPU passthrough, which suits the compute duty this tower often pulls.

Ubuntu Server LTS covers Linux application and GPU compute duty with five years of free updates, and it is the default base for most machine learning stacks.

No OS is there for teams imaging from their own deployment infrastructure.

Not sure which fits? Tell us the workload at quote time, or call 1-800-778-1545.

WServer Warranty


Dell PowerEdge T640 8-Bay 3.5" Drives [14th Gen] Detailed Review

The refurbished Dell PowerEdge T640 8-Bay 3.5" is the 14th gen flagship tower server: a 5U dual-socket platform carrying the full enterprise envelope (24 DIMM slots symmetric, up to 8 PCIe Gen3 slots, up to four 300W GPUs, NVDIMM-N persistent memory, and up to 2400W power supplies) in a tower chassis built for office and remote-site deployments that need datacenter-class compute without rack infrastructure. The 8-Bay 3.5" configuration is the one we reach for when bulk LFF capacity is the storage priority: eight hot-swap 3.5" front bays for SAS, SATA, or Nearline SAS drives alongside the platform's flagship compute envelope.

This is the right tower for branch-office virtualization hosts running 30 to 60 VMs, remote-site SQL or Exchange servers that need serious capacity, modest GPU-accelerated workloads such as CAD, inference, and VDI in office environments, and any deployment where rack space is unavailable but datacenter-class platform headroom is required. In positioning it is the tower equivalent of the R740 and R740xd: same socket, same memory topology, same PCIe envelope, same iDRAC9, in a rack-convertible tower chassis.

To configure a build, call 1-800-778-1545 or use the quote form on this page and we will respond within 24 hours. Every refurbished T640 ships after a 12+ hour burn-in covering every memory channel, every PCIe slot, and every drive bay, backed by our standard 180-day warranty with 1-Year, 2-Year, and 3-Year Premium options available. Volume pricing applies at 5 units and above.

Where the T640 8-Bay 3.5" Fits in the Family

The T640 sits at the top of Dell's 14th gen PowerEdge tower line, above the entry-tier Dell PowerEdge T340 8-Bay 3.5" entry tower (single-socket, modest envelope) and the mid-range Dell PowerEdge T440 8-Bay 3.5" tower (dual-socket, 16-DIMM asymmetric memory, 1 TB ceiling, single GPU). The T640 is the only 14th gen tower with the full flagship platform: 24 symmetric DIMM slots, a 3 TB memory ceiling, four-GPU support, eight PCIe Gen3 slots, and NVDIMM-N persistent memory.

Within the T640 chassis line we stock two configurations: this 8-Bay 3.5" LFF variant for bulk capacity, and the Dell PowerEdge T640 16-Bay 2.5" SFF configuration for IOPS-leaning, higher-VM-density workloads. The platform underneath is identical; the difference is the storage profile. Choose this 8-Bay LFF for file serving, capacity-tier databases, backup repositories, or a four-GPU build that also needs bulk local storage. Choose the 16-Bay 2.5" SFF when transaction IOPS, high VM count, or optional NVMe matter more than raw capacity.

Storage: 8 x 3.5" LFF Bays for Flagship-Tier Tower Capacity

The 8-Bay 3.5" chassis provides eight front-accessible hot-swap 3.5" drive bays for SAS, SATA, or Nearline SAS drives. The backplane is SAS/SATA only; this LFF chassis does not support front NVMe (NVMe lives on the 16-Bay 2.5" and specialist 24-Bay 2.5" variants). With eight 22 TB Nearline SAS drives, raw capacity reaches 176 TB; in RAID 6 with one hot spare, usable capacity lands near 110 TB. That is real bulk-storage density backed by the T640's flagship compute envelope, which matters when the tower does more than serve files: dense VM hosting with capacity-tier storage, SQL databases with multi-TB data sets, and application servers with large content stores.

For boot, the T640 uses a BOSS PCIe card (Boot Optimized Storage Solution) at the rear of the system: up to two 80 mm or 110 mm M.2 SATA devices in hardware RAID 1, the same BOSS-S1 module used across the R740 and R740xd. Putting the OS on BOSS keeps all eight front bays free for data and removes the OS from the data array entirely. IDSDM (Internal Dual SD Module) and an internal USB option exist for hypervisor-only boot, but BOSS is the right call for production. BOSS drives are cold-swap on this platform; replacement requires downtime.

Drive guidance for the 8-Bay 3.5": for bulk capacity we spec 12 TB, 16 TB, 20 TB, or 22 TB Nearline SAS 7.2K drives. RAID 6 is mandatory on any array of 8 TB and larger drives, because the rebuild window on large NL-SAS arrays carries real double-disk-failure risk. For mixed workloads, a SAS SSD pair (1.92 TB or 3.84 TB) for cache or hot data alongside six NL-SAS bulk drives is a clean layout. External SAS shelf expansion is supported through the H840 external controller for deployments that outgrow eight bays.

Storage Controllers: PERC H740P Is the Default

The T640 supports the standard 14th gen flagship PERC family in a dedicated controller slot that keeps all eight PCIe expansion slots free:

  • PERC H740P (8 GB NV cache, battery-backed): the production default for write-intensive or mixed read/write workloads on this chassis.
  • PERC H730P (2 GB NV cache, battery-backed): solid general-purpose choice for read-leaning workloads.
  • PERC H330 (no cache): entry-tier hardware RAID for light workloads only.
  • HBA330 (pass-through HBA): for software-defined storage stacks such as Storage Spaces Direct, Ceph, and ZFS.
  • PERC S140 (software RAID via the C620 chipset): dev and test only. We do not quote S140 for production.
  • PERC H840 (external, 8 GB cache): for SAS shelf expansion beyond the internal eight bays.

For the 8-Bay 3.5" our default recommendation is the PERC H740P. The 8 GB non-volatile cache earns its place on bulk-capacity workloads with mixed read/write patterns: backup-target ingest, file-server cold writes, and modest database transaction logging. For software-defined storage builds the HBA330 pass-through is the right call; the T640's eight PCIe slots make it workable as a tower hyperconverged node, though for serious clustered storage we still point customers to rack platforms.

Processors: 14th Gen Skylake-SP and Cascade Lake-SP

The T640 is built on Intel's LGA 3647 socket and takes up to two Xeon Scalable processors from the 1st generation Skylake-SP family or the 2nd generation Cascade Lake-SP family. Same socket, drop-in compatible with a BIOS update. For any new T640 deployment in 2026 we spec 2nd gen Cascade Lake: better performance per watt, hardware Spectre and Meltdown mitigations, and access to the widely available Refresh SKUs (Gold 6230R, Gold 6248R, Gold 6258R).

The platform supports up to 28 cores per socket (Platinum 8280) and accepts CPUs up to 205W TDP, meaningfully higher than the T440's 150W mainstream ceiling. Common specs: the Gold 6230 (20 cores, 2.1 GHz, 125W) for balanced virtualization and database workloads; the Gold 6248R (24 cores, 3.0 GHz, 205W) where clock speed matters; and the Platinum 8280 (28 cores, 2.7 GHz, 205W) for maximum-density VM hosting. Dual Gold 6248R is a common build for serious tower virtualization.

For the 8-Bay 3.5" specifically, the LFF bays bias the deployment toward bulk-capacity workloads where memory and storage matter more than raw core count, so dual Gold 6230 (40 cores total) is our most common spec, stepping to dual Gold 6248R (48 cores) for compute-heavier mixes. The T640's thermal envelope genuinely supports dual 205W CPUs without acoustic compromise; the chassis was designed for serious dual-socket operation and ships with the high-performance heatsinks those CPUs require. Single-socket configurations are supported but cut memory to 12 DIMMs and PCIe to three slots, which is rarely the right call on a flagship-tier tower; if single-socket is enough, the T440 is the better-positioned platform.

Memory: 24 DIMMs Symmetric, Up to 3 TB

The T640 has 24 DDR4 DIMM slots in a fully symmetric topology: CPU1 owns 12 slots, CPU2 owns 12 slots, six channels per CPU at two DIMMs per channel. This is the same flagship memory topology as the R740 and R740xd, and a real upgrade over the T440's asymmetric 10-plus-6 layout. Symmetric population gives NUMA-aware applications balanced per-socket bandwidth, which matters for VM density, large databases, and analytics.

Memory speed reaches 2933 MT/s at 1 DIMM per channel on Cascade Lake, dropping to 2666 MT/s at 2 DIMMs per channel under full population; mixed-speed configurations run at the slowest installed DIMM. Skylake-SP tops out at 2666 MT/s regardless. The 2933 ceiling at 1 DPC is a genuine delta over the T440's flat 2666 MT/s and matters for memory-bandwidth-sensitive workloads.

Maximum memory is 3 TB with 24 x 128 GB LRDIMMs (3DS); with 64 GB LRDIMMs or 64 GB RDIMMs the ceiling is 1.5 TB; single-socket configurations max at 1.5 TB across 12 DIMMs. We typically ship T640 systems in the 384 GB to 768 GB range for tower virtualization, stepping to 1.5 TB for serious VM density. The 3 TB ceiling is rarely needed outside in-memory database workloads.

NVDIMM-N persistent memory is supported: up to 12 x 16 GB NVDIMM-N modules (one per channel), totaling 192 GB of persistent memory. It requires both CPUs installed and follows specific population rules (NVDIMM-N may be mixed with RDIMM but not with LRDIMM). NVDIMM-N is genuine storage-class memory, flash-backed with a backup battery so data survives power events. This is unique to the T640 in Dell's tower line; the T440, T550, and T560 support no persistent memory in any form. For tower deployments running write-intensive transactional databases, SAP HANA, or Storage Spaces Direct with a persistent metadata tier, NVDIMM-N is the platform-justifying feature.

Networking and PCIe Expansion

The T640 ships with two onboard 10 GbE BASE-T LOM ports (Broadcom 57416), a meaningful step up from the T440's 2 x 1 GbE baseline. On a flagship-tier tower, 10 GbE is the standard rather than an upsell, and it is enough for most SMB and remote-site virtualization with iSCSI or NFS storage networking without adding a PCIe NIC.

For more, the T640 takes rNDC (rack Network Daughter Card) options including dual 10 GbE SFP+, dual 25 GbE SFP28 (Mellanox ConnectX-4 Lx), and quad-port 1 GbE. The chassis carries up to 8 PCIe Gen3 expansion slots plus a dedicated PERC slot with both CPUs installed; single-CPU configurations expose only 3 PCIe slots. That slot count is a real advantage over the T440's five slots and leaves room for 10/25/40/100 GbE NICs, HBAs, and GPUs together. For dense virtualization we typically pair the onboard 10 GbE with a 25 GbE Mellanox ConnectX-4 Lx card.

GPU Support: Up to Four 300W Accelerators

The T640 supports up to four 300W GPU accelerators, the strongest GPU envelope of any Dell tower in any generation (the 15th gen T550 maxes at two; the 16th gen T560 supports up to six but at lower per-GPU power). The four-GPU configuration is a chassis-level option that must be specified at purchase, because it requires specific cooling and PCIe routing that cannot be retrofitted. Qualified cards have included the NVIDIA Tesla V100, T4, A10, A30, A40, A100, and RTX series, plus AMD MI-series accelerators; the qualified-card list shifts over time, so we confirm it at quote time.

One platform constraint matters: with NVMe storage configurations the GPU ceiling drops to two cards. The 8-Bay 3.5" chassis is SAS/SATA only, so the full four-GPU envelope is available here. If a deployment needs both NVMe storage and four GPUs, the platform forces a choice and a rack platform is the better answer. The four-GPU envelope makes this chassis a real option for office-deployed GPU work: branch AI and ML inference, CAD render nodes for engineering offices, dense VDI (60 to 100 light desktops per host), and modest on-prem ML training. This is the T640's strongest differentiator over the single-GPU T440 and the clearest reason to choose tower over rack when the deployment specifically needs office-deployable multi-GPU compute.

Management: iDRAC9 Generation

Out-of-band management is iDRAC9, standard across 14th gen PowerEdge. We recommend the iDRAC9 Enterprise license for any production T640: virtual console redirection, virtual media, automated firmware updates through the Lifecycle Controller, group management via OpenManage Enterprise, and SupportAssist proactive diagnostics. iDRAC9 Express lacks virtual console and is insufficient for remote troubleshooting at branch or unattended sites, which is exactly where flagship towers tend to live.

The platform carries the full iDRAC9 security baseline: TPM 2.0, cryptographically signed firmware, Silicon Root of Trust, Secure Boot, System Lockdown (Enterprise plus OpenManage Enterprise), Quick Sync 2.0 mobile management, and System Erase data sanitization. For remote-site deployments with limited on-site IT, iDRAC9 Enterprise is the single most useful line on the BOM: remote console is what saves a site visit when something breaks.

Power and Cooling

The T640 supports a broader PSU range than any other Dell tower in our catalog. All are hot-plug and support redundant 1+1 operation:

Configuration PSU Recommendation Est. Peak Draw
Light (Gold 6230, 256 GB RAM, 4 NL-SAS, no GPU) 2x 750W Platinum ~420W
Balanced (dual Gold 6230, 512 GB RAM, 8 NL-SAS, no GPU) 2x 1100W Platinum ~640W
Heavy (dual Gold 6248R, 768 GB RAM, 8 NL-SAS, 2x 300W GPU) 2x 1600W Platinum ~1450W
Maximum (dual Platinum 8280, 1.5 TB RAM, 8 NL-SAS, 4x 300W GPU) 2x 2400W Platinum ~2100W

The 750W pair handles non-GPU light builds; 1100W is the right default for dual-socket Gold-tier compute without GPUs; 1600W is required for two-GPU configurations; 2400W is required for four-GPU configurations. The 2000W and 2400W PSUs derate at low line (100 to 120V AC), so any two-GPU-or-greater build should run on 200 to 240V AC for full output. Dual hot-plug redundant Platinum PSUs are mandatory for production; Titanium-tier SKUs are available where efficiency targets call for them. Cooling uses a redundant fan configuration, which is what lets the chassis carry dual high-TDP CPUs and multi-GPU loads while staying office-acceptable in most builds. Four-GPU plus dual 205W CPU configurations will run noticeably louder.

Physical Specs & Platform Notes

  • Form factor: 5U tower, rack-convertible with the optional rack conversion kit. Chassis depth roughly 726 mm; loaded weight near 35 kg with eight LFF drives and two PSUs. In rack mode it consumes 5U.
  • PCIe expansion: up to 8 PCIe Gen3 slots plus a dedicated PERC slot with both CPUs installed; slots 4 through 8 require the second processor, and single-CPU builds expose only 3 slots.
  • Parts availability: excellent. The T640 shares its platform, PERC family, BOSS module, iDRAC9, and PSUs with the high-volume R740 and R740xd, so spares and field-replaceable units are mature and widely stocked. Dell ProSupport on 14th gen is approaching end of extended support, so third-party maintenance is the standard production support path in 2026.
  • Accessories we recommend: the BOSS-S1 boot card for production boot; the rack conversion kit if rack deployment is planned (sold separately, add it to the BOM up front); the iDRAC9 Enterprise license; and a 25 GbE Mellanox ConnectX-4 Lx NIC for dense virtualization.
  • Platform notes: the LFF backplane is SAS/SATA only (no front NVMe on this chassis); BOSS is cold-swap; four-GPU support must be ordered at build time and cannot be retrofitted; and the four-GPU and NVMe options are mutually exclusive on the platform.

Our Assessment

Where it excels: the T640 8-Bay 3.5" is the right call when a deployment needs the flagship platform envelope (24 symmetric DIMMs, 3 TB memory, four GPUs, eight PCIe slots, NVDIMM-N) in a tower form factor with bulk LFF capacity as the storage priority. It is strong for branch-office virtualization hosts running 30 to 60 VMs with capacity-tier storage, remote-site SQL and Exchange servers with serious data sets, office-deployed GPU work (CAD, AI inference, dense VDI), and persistent-memory-aware workloads that need NVDIMM-N in a tower.

Where to look instead: if rack space is available and tower form factor is not required, the R740xd is the same platform in 2U and generally a better datacenter fit. If the workload fits a smaller envelope and budget is the constraint, the T440 is meaningfully cheaper and right-sized. If IOPS or NVMe matter more than bulk capacity, the T640 16-Bay 2.5" SFF configuration is the better choice on the same platform. All three are linked in the sections above and below.

Bottom line: this is the 14th gen tower to buy when you need flagship-tier platform headroom, want bulk LFF capacity, and require a tower form factor for office acoustics or sites with no rack. It is the last Dell tower built at the 24-DIMM, four-GPU, NVDIMM-N envelope, and there is no direct successor at that tier in 15th or 16th gen. If your deployment does not need the flagship envelope, we will tell you the T440 is the smarter buy; if you have rack space, we will point you to the R740xd. That is the call we make at quote time.

Where the T640 Fits in 2026

The T640 succeeds the 13th gen Dell PowerEdge T630 8-Bay 3.5" (13th gen flagship tower) (Broadwell, iDRAC8, 24 DIMMs at 2400 MT/s, two-GPU envelope, no NVMe). Moving up to the T640 brings the Skylake and Cascade Lake architecture, iDRAC9 with Silicon Root of Trust, faster memory at 2933 MT/s, four-GPU support, BOSS internal boot, and eight PCIe Gen3 slots. Buying a refurbished T630 in 2026 saves a little but gives up real platform value.

There is no direct flagship-tower successor in 15th or 16th gen. The 15th gen T550 tops out at 16 DDR DIMM slots (2 TB max) and two GPUs but brings PCIe Gen4 and 3rd Gen Xeon. The 16th gen Dell PowerEdge T560 12-Bay 3.5" (16th gen tower) moves to DDR5 at up to 4800 MT/s, PCIe Gen5, and BOSS-N1 NVMe boot, and supports up to six GPUs, but carries only 16 DDR5 DIMM slots (1 TB max) and no NVDIMM-N. For deployments that genuinely need 24 DIMMs, 3 TB of memory, or NVDIMM-N in a tower, the T640 is still the answer in 2026, and it is the last Dell tower built at that envelope. On support: Dell ProSupport for 14th gen is near end of extended support, so plan production coverage around third-party maintenance.

Honest Limitations

  • PCIe Gen3 ceiling. No Gen4 or Gen5 expansion. Modern Gen4 NICs and HBAs run at roughly half native bandwidth; H100, L40S, and other Gen4/Gen5 GPUs are throttled by the bus. Match GPUs to the platform: V100, T4, A10, A30, A40, and A100 are well-suited; H100 and Gen5 cards are bottlenecked.
  • No front NVMe on this chassis. The 8-Bay 3.5" backplane is SAS/SATA only. NVMe lives on the 16-Bay 2.5" configuration and specialist 24-Bay variants. For NVMe storage in a tower, choose the 16-Bay 2.5" or move to rack.
  • Four GPUs and NVMe are mutually exclusive. Per Dell's platform spec, NVMe configurations cap GPUs at two. This LFF chassis supports the full four-GPU envelope precisely because it has no NVMe; if both matter, the platform forces a choice.
  • Single-socket loses half the platform. Single-CPU T640 builds expose only 12 DIMMs and 3 PCIe slots. Single-socket is rarely the right call here; the T440 is better-positioned for single-socket tower needs.
  • 2400W PSU derates at low line. Two-GPU-or-greater builds should run on 200 to 240V AC. At 100 to 120V AC the top PSUs derate and may force a lower-spec build.
  • 5U footprint is large. Rack-converted, the T640 consumes 5U against the R740xd's 2U. For rack-dense sites the rack platforms are better-positioned; the tower makes sense for office and remote-site deployments.
  • BOSS is cold-swap. Boot-module replacement needs downtime. Hot-swap boot arrived with 15th gen (BOSS-S2) and NVMe boot with 16th gen (BOSS-N1).
  • iDRAC9 Express is insufficient for production. Always add Enterprise, especially at unattended sites. Remote console is the feature you miss most when something breaks with no on-site IT.
  • NVDIMM-N has population rules. Persistent-memory builds need both CPUs, cannot mix NVDIMM-N with LRDIMM, and require OS support for storage-class memory (Windows Server 2016 and later with the right drivers, or Linux with libnvdimm). Confirm at deployment.
  • Rack rails are a separate line item. The chassis is rack-convertible but the kit is not included. Add it to the BOM if rack deployment is planned.

Workload Fit

What the T640 8-Bay 3.5" Excels At Consider Alternatives For
Branch-office virtualization (30 to 60 VMs with capacity storage) SMB/ROBO under the T440 envelope (use the T440)
Remote-site SQL, Exchange, and database servers with large data sets Datacenter rack deployments (use the R740xd)
Office-deployed multi-GPU compute (AI inference, CAD, dense VDI) Bulk SFF or IOPS-leaning workloads (use the 16-Bay 2.5")
NVDIMM-N persistent-memory tower deployments NVMe storage requirements (16-Bay 2.5" or rack)
Tower file servers backed by serious compute (24 DIMMs, 3 TB max) Current-gen GPU compute at scale (R750xa, R760xa)
Tower hyperconverged nodes (Storage Spaces Direct, ZFS, modest Ceph) DDR5 memory-bandwidth-bound workloads (T560, R760)

Where to Look Instead

If your deployment does not fit the T640 8-Bay 3.5", these are the configurations we point customers to:

  • Dell PowerEdge R740xd 12-Bay 3.5" rack server: the same 14th gen platform in a 2U datacenter form factor with greater storage density and broader NVMe options. The better fit whenever rack space is available.
  • Dell PowerEdge R650 8-Bay 2.5" (15th gen rack): a generation newer, with Ice Lake CPUs, DDR4 at 3200 MT/s, and PCIe Gen4, for deployments that want a current-tier rack platform rather than a 14th gen tower. For the SFF version of this T640 platform, the 16-Bay 2.5" configuration linked above is the companion to consider.

Ready to Configure?

Tell us your workload, target memory capacity, drive count and capacity per drive, single-socket or dual-socket, whether GPU acceleration is needed and how many cards, and whether NVDIMM-N persistent memory is in scope. We will translate that into a specific build and a firm quote.

Call 1-800-778-1545 or submit the quote form on this page and we will respond within 24 hours. Every T640 we ship is tested with a 12+ hour burn-in and backed by a 180-day warranty, with extended 1-Year, 2-Year, and 3-Year Premium coverage available. Volume pricing applies at 5 units and above.

Estimated TDP Draw: 0W
0 lines

Dell PowerEdge T640 8-Bay 3.5"

8-Bay 3.5"

Subtotal $3,587.70
Estimated TDP Draw 0W
Build total $3,587.70

Choosing Memory for Your Dell PowerEdge T640

The T640 has 24 DDR4 DIMM slots, 12 per CPU, across 6 channels per socket at 2 DIMMs per channel. Fill all 24. This is our standing recommendation and we push it hard. There are two benefits. The first is performance: full population at 2 DPC keeps all twelve memory channels working at maximum depth, and on memory-bandwidth-sensitive workloads like virtualization and GPU-fed compute it consistently outperforms a partial population running at higher clock. The second is cost per gigabyte: 24 smaller modules hit the same capacity for meaningfully less than 12 larger ones, and 768 GB as 24× 32 GB is the best cost per GB in the lineup. We know the instinct is to leave slots open for a later upgrade, and on paper that is sensible. In practice it rarely plays out well. One year from now you are hunting for DIMMs that match the rank and speed of what is already in the box, and a mismatch either clocks the whole server down or refuses to train. The server is in production by then, so adding memory means downtime everyone notices. Buy the capacity now, in one matched set. If you outgrow it, the right move is usually another node rather than another DIMM. 384 GB (24× 16 GB) for file, application, and light virtualization duty 768 GB (24× 32 GB) is our default for a serious virtualization host or GPU compute box 1.5 TB (24× 64 GB) for dense consolidation or in-memory workloads Past 1.5 TB you are into 128 GB LRDIMMs, which reach 3 TB but cost substantially more per GB. Most buyers are better served by a second host.

Choosing Your iDRAC License

The T640 ships with iDRAC9 either way; the license determines what you can do with it remotely. The dividing line is the remote console. iDRAC9 Enterprise gives you full remote KVM and virtual media, meaning you can reach the server's screen, mount an ISO, and rebuild an OS from anywhere, even when the operating system is down. Express does not include the remote console, so any problem below the OS level means someone physically at the machine. If this T640 sits down the hall and someone on site can reach it without ceremony, Express is a legitimate answer. The case for Enterprise is who actually fixes the server: if that person is an outside IT provider or a head-office admin who is not in the building, Enterprise turns every below-the-OS problem from a site visit into a remote session, and one avoided trip pays for the license. A T640 also tends to carry more weight than a typical tower (GPU compute, a serious virtualization stack, an office's entire storage), and the more the box matters, the shorter the Enterprise conversation gets. It also adds automated firmware updates, two-factor authentication, and secure erase. Deployments of 5 or more units should take Enterprise on all of them; fleet management without remote console access does not work in practice.

Choosing Your T640 Power Supply

Every option here is a dual hot-swap redundant pair, so redundancy and serviceability are already handled. The only decision is wattage, and the T640's range is wide because the chassis is: the same tower carries everything from quiet office duty to multi-GPU compute. The 750W pair is our default for balanced builds: dual Gold CPUs, full memory, and a backplane of SSDs sit comfortably inside it at the 50 percent efficiency sweet spot. The 1100W pair is where top-bin builds belong: high-wattage CPUs with every bay populated, or any single-GPU configuration. The 1600W pair and above is GPU territory. The T640 is the tower people buy to put accelerators in, and multi-GPU builds need the headroom; we spec the top tiers on any build with two or more cards. 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; the price step is small. Watch the estimated TDP counter at the bottom of the page as you add components to see where your build lands. Two tower-specific notes. If a GPU might be added later, buy the higher-wattage pair now; the swap is painless but the second shipment is not free. And since a tower usually plugs into wall power rather than conditioned rack feeds, put the two cords on different circuits where the building allows it, with at least one on a UPS.

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.