UK’s trusted IT infrastructure partner since 2003
Servnet
FinanceToolsConfiguratorGet in Touch
Technical data sheet · PowerEdge XE rack-scale · 1 OU liquid-cooled tray, 2 × Grace

Dell PowerEdge XE8712Specifications, configurations & comparisons

The XE8712 is not a stand-alone server: it is a 1 OU, direct-liquid-cooled compute tray that only runs inside Dell’s IR7044 or IR7050 Open Rack v3 cabinet, drawing DC power from the rack busbar and coolant from the rack manifold. Each tray pairs two 72-core NVIDIA Grace CPUs with four Blackwell Ultra GPUs. This page separates what you buy per tray from what the rack has to supply.

Rack-scale2 socketsNVIDIA GraceiDRAC10
4GPUs per trayBlackwell Ultra · 192 GB HBM3e each
2Grace CPUs per tray72 cores each · up to 3.1 GHz
960GBLPDDR5X, soldered480 GB per CPU · no DIMM slots
36trays per 50 OU rackminimum 3 · 144 GPUs at the maximum
33kWper rack power shelf6 × 5500 W AC PSUs · none in the tray
10kWliquid heat capture per OUblind-mate 21-inch DLC manifolds

Every figure checked against the Dell PowerEdge IR7000, XE8712 Technical Guide (Rev. A01, January 2025)

Form factor
Rack-scale · 2 sockets
Processors
NVIDIA Grace
Drive bays
Up to 2 · 1 drive-bay layout
Expansion
4 × PCIe · 1 × OCP 3.0
GPUs
Up to 4 · 1 listed
Boot
M.2 and BOSS · 2 boot options
Management
iDRAC10
01 — Overview

The XE8712 at a glance

Our read of Dell’s documentation: where the XE8712 is strong, and the limits worth knowing before you spec one.

Strengths

  • Four GPUs and two 72-core Grace CPUs in a single 48 mm Open Rack unit, with up to 36 trays in an IR7050 — the 144 Blackwell GPUs per rack that Dell’s spec sheet quotes.
  • No power or coolant cabling at the tray: it has no power supplies, and it blind-mates onto the rack’s DC busbar and liquid manifold as it is pushed home.
  • Liquid cold plates carry the CPU and GPU heat, so Dell sets no CPU or GPU thermal restrictions and allows ambient air up to 40 °C.
  • Front I/O flexibility for a rack-scale tray: up to four FH HL PCIe Gen5 x16 slots across six riser configurations, an OCP 3.0 Gen5 x16 slot, and Dell’s CX7 or CX8 fabric options.
  • Managed like current PowerEdge: iDRAC10, OpenManage Enterprise with Power Manager, and Red Hat Enterprise Linux or Ubuntu Server LTS.

Watch-outs

  • It cannot run on its own — it needs an IR7044/IR7050 rack with power shelves, busbar, DLC manifolds, a CDU and switches, and Dell sets a minimum of three trays per rack.
  • Local storage is thin by design: two E3.S NVMe bays and one M.2 boot drive, no SAS, SATA or U.2 support and no HBA or PERC — and the spec sheet lists the E3.S bays as not available at launch.
  • Loud: Dell measured up to 79 dB(A) at the bystander position under maximum load at 35 °C, advises against siting it near people, and calls for industrial-grade ear protection.
  • The guide contradicts itself in places — tray size and weight (30 kg or 33.07 kg), the Grace TDP (600 W in Table 1, “Grace 300W” in the acoustic set-ups) and the DC bus voltage (48, 49/50, 51 or up to 54 V) — so confirm the facility design with Dell.
  • Static L-bracket rails only: no cable arm, no strain-relief bar, no servicing in the rack, and Dell advises two people to lift the tray.
Processors1 SKU
Cores / CPU72
Drive layouts1
Max bays2
GPUsup to 4
Rack power parts1 · to 5,500 W
02 — Processors

1 supported processor, mapped

Every tray carries two NVIDIA Grace processors, and Dell lists no other processor for it. Each has 72 cores and runs at up to 3.1 GHz; Dell does not publish thread count, base clock or cache. The processors reach the GPUs over a coherent NVLink CPU-GPU interconnect that Dell rates at 900 GB/s.

Processor landscape — cores × TDP

1 SKUs · 1 family
Dell PowerEdge XE8712: every supported processor by cores and TDP081624324048566472600 WCores per processorNVIDIA Grace (GB200 NVL4) — 72 cores · 600 WNVIDIA Grace (GB200 NVL4)72 cores · 8.3 W/core — top SKU and most efficient
NVIDIA Grace 1Bubble size = L3 cache · hover a bubble for the full SKU

Processor explorer

filter · sort · compare
ProcessorCores / threadsBaseL3 cacheTDPW / coreMemory
NVIDIA Grace (GB200 NVL4)72600 W8.3
  • Table 1 gives a maximum TDP of 600 W; the acoustic test configurations describe the parts as “GB200 with Grace 300W”. Both figures are Dell’s own.
  • The two processors sit under the tray’s left and right direct-liquid-cooling cold plates rather than fan-cooled heatsinks.
  • Riser slots 1 and 2 and the OCP slot are wired to CPU0; riser slots 4 and 6 and the M.2 interposer (slot 5) to CPU1.
03 — Memory

Memory built onto the processor board — no DIMM slots

There are no DIMM slots to fill. Each Grace has 480 GB of LPDDR5X with ECC soldered down, 960 GB per tray, and each of the four GPUs carries 192 GB of HBM3e — 768 GB of GPU memory per tray by simple multiplication. Capacity is fixed at the factory: Dell supports neither DIMM nor CXL 2.0 memory expansion.

Integrated memory

fixed at the factory

960 GB LPDDR5X (480 GB per CPU, soldered)

  • Dell does not publish an LPDDR5X speed for this tray.
  • Table 2 calls the CPU memory “LP DDR5”; Table 1 and the memory chapter say LPDDR5X.
04 — Storage

1 drive-bay layout, up to 2 bays

Dell sizes local storage for operating-system deployment and application logging. Two front E3.S NVMe bays sit on their own backplane behind a Gen5 PCIe switch, each drive on a Gen5 x4 link with hot-plug, and a single M.2 boot drive rides on an interposer in slot 5. Dell also supports diskless trays.

Headline layouts

front panel view

Front panel

2 × EDSFF NVMe at the front — 2 bays in total.

SAS/SATANVMeSAS/SATA + NVMe

Controllers & boot

  • Direct NVMe (CPU-direct via Gen5 PCIe switch) — Onboard (no RAID). No HBA or PERC supported; SATA/SAS/U.2 not supported
  • 1 x M.2 boot (single M.2 drive on M.2 interposer, slot 5, PCIe Gen4 x4, BOSS flatbread-compatible connector; not hot-swap) — boot
  • Diskless configuration also supported — boot
  • No SATA, SAS or U.2 drives and no HBA or PERC: the NVMe drives reach the CPUs directly through the PCIe switch.
  • Dell’s spec sheet marks the E3.S bays as not available at product launch, and the guide names no E3.S capacities; confirm availability when ordering.
  • With no drives, a vented blanking plate replaces the drive cage while the backplane stays fitted.
  • The M.2 interposer carries one drive on PCIe Gen4 x4 in the slot table (the acoustic test set-ups list an “M.2 Gen5 x4” boot drive); it is front-accessible and tool-less but not hot-swap, so boot is a single device rather than a mirrored pair.
05Accelerators & I/O

GPUs, PCIe and networking

The GPUs are not PCIe cards. Four Blackwell Ultra GPUs — Dell’s wording; its acoustic section calls the board an NVIDIA GB200 NVL4 HPM — are liquid-cooled like the two Grace CPUs, and the inside view also shows an NVLink hard-bridge. Each GPU has 192 GB of HBM3e. The guide gives no GPU TDP, clock or throughput figures, so none appear here. 4 × Blackwell Ultra per tray, 192 GB HBM3e each · up to 144 per rack at 36 trays.

NVIDIA Blackwell Ultra (GB200 NVL4 HPM, 4-way)

192GB HBM3e per GPU · 4 per tray

I/O topology

2 sockets
CPU 1PCIe 5.00 DIMMCPU 2PCIe 5.00 DIMMUp to 4 × FH HL PCIe Gen5x16 · front risers 1, 2, 4, 6M.2 interposer · slot 5PCIe Gen4 x4 · one boot driveOCP 3.0 · slot 3Gen5 x16 · 2 × 25GbE SFP28 cardiDRAC10 + AST2600 BMC2 × RJ45 iDRAC · HMC for the HPM

All expansion is at the front of the tray. Up to four full-height half-length PCIe Gen5 x16 slots sit on front risers, next to an OCP 3.0 slot at Gen5 x16 and the M.2 interposer. PCIe runs at Gen5, 32 GT/s.

Network adapters Dell lists

4 options
AdapterPortsForm
Mellanox OCP 3.0 2 x 25GbE SFP282 × 25GbEOCP 3.0
NVIDIA ConnectX-7 (CX7) x16 PCIe - up to 4 × PCIe
NVIDIA ConnectX-8 (CX8) OSFP - up to 2 × PCIe
NVIDIA BlueField-3 (B3240) SuperNIC / DPU FHHL - 1 × PCIe
  • Riser configurations RC0, RC5 and RC6 give four x16 slots (1, 2, 4 and 6); RC2 and RC7 give three (2, 4 and 6); RC1 gives two (2 and 4).
  • Slots 1 and 4 are rated 75 W and slot 6 150 W; slot 2 is 150 W in the four-slot configurations and 75 W in RC1, RC2 and RC7. The OCP slot is rated 86 W.
  • Dell’s fabric options read “4 x16 CX7 or 2 CX8 OSFP ports”, alongside one BlueField-3 SuperNIC (FHHL); its noisier acoustic test tray carried a BlueField-3 B3240 and two CX-7 cards.
  • The OCP feature table allows ports up to 100 GbE, yet the only OCP card Dell lists for this tray is a Mellanox 2 × 25GbE SFP28.
  • Fitting a BlueField-3 DPU raises fan speed, and with it the noise level.
06 — Power & cooling

No supplies of its own — powered by the rack

The tray has no power supplies. The IR7000 rack carries ORv3 33 kW power shelves (PS33), each holding six 5500 W AC supplies, and they feed a 1400 A busbar that the tray plugs into at the back. How many shelves a rack needs depends on the tray count and the redundancy model, and Dell publishes a table for each.

050010001500200025003000350040004500500055005500 W >97.5% peak (shelf: 97.2% peak / 95.0% at full load) — No PSUs in the XE8712 tray. IR7000 PS33 power shelf holds 6 x 5500 W AC PSUs (200-277 V, 50-60 Hz; single or three phase input; 50 V/49 V 112 A output), feeding a 48 V (up to 54 VDC) ORv3 1400 A busbar; 33 kW / 674 A per shelf. Shelf count scales with node count: AI N+1 = 2 shelves (1-2 nodes) to 8 shelves (29-33 nodes); HPC half-shelf redundancy = 2 shelves (1-5 nodes) to 8 shelves (31-36 nodes). Delta and Lite-On 21-inch 33 kW 60 A shelves listed.5500DCWatts · dashed line = output at 100–120 V low line
AC
  • AI deployments (N+1, one spare shelf): 2 shelves for 1–2 trays, 3 for 3–6, 4 for 7–13, 5 for 14–18, 6 for 19–23, 7 for 23–28 (Dell’s ranges overlap at 23) and 8 for 29–33.
  • HPC deployments (half-shelf redundancy, three spare PSUs): 2 shelves for 1–5 trays, 3 for 6–10, 4 for 11–15, 5 for 16–20, 6 for 21–26, 7 for 27–30 and 8 for 31–36 — the only table that reaches the 36-tray maximum.
  • Shelf input is 200–277 V single-phase or three-phase delta, or 346/480 V three-phase star, at 50/60 Hz through a 7-pin 60 A connector. The shelf has no input fusing of its own, so upstream per-phase circuit breakers are needed; this guide’s breaker rating is still marked as unconfirmed, so take the rating from Dell.
  • Each shelf outputs 33 kW at 674 A and weighs 38 kg with six PSUs. Efficiency is given as 97.2% peak and 95.0% at full load in the shelf feature list and input table, but as more than 97.5% peak in the PSU table; the spec sheet cites up to 264 kW per rack.
  • Voltage wording varies by section: the busbar is a 48 V ORv3 design, the shelf output is listed at 49/50 V and 51 V, and the tray takes up to 54 VDC.
  • Heat leaves through 21-inch blind-mate DLC manifolds with two quick-disconnects per OU and up to 10 kW of capture per OU. Dell names the Motivair MCDU-50 and Vertiv CDU 1350 as in-row CDU examples and refers CDU choice to its sales team.
07 — Management, security & OS

Run it, secure it, choose the OS

Management

  • iDRAC10 — Datacenter licence offered as an upsell, adding telemetry streaming, thermal, GPU/DPU and air-and-liquid cooling management
  • Aspeed AST2600 BMC (DC-SCI compatible) with a BMC-to-NVIDIA HMC link for the GPU board
  • iDRAC RESTful API with Redfish · Zero-Touch Provisioning · iDRAC Service Module
  • Front 2 × RJ45 iDRAC ports and a USB Type-C iDRAC Direct port
  • OpenManage Enterprise with Power Manager, Update Manager and SupportAssist plug-ins
  • Power-shelf PMC: own Ethernet port, Redfish and OpenBMC, monitored and updated through OpenManage Enterprise

Security

  • TPM module on the board · chassis intrusion switch
  • Secure Enterprise Key Management and Secure Component Verification
  • System Lockdown (iDRAC10)
  • Silicon-based root of trust, multi-factor authentication and role-based access (spec sheet)
  • Firmware to be flashed only from Dell’s own website

Operating systems

  • Red Hat Enterprise Linux (Tier-1 in Dell’s management matrix)
  • Canonical Ubuntu Server LTS
  • The spec sheet lists RHEL only; Windows Server and VMware ESXi are not listed
08 — Compare

How the XE8712 compares with its family

Dell’s liquid-cooled XE range comes in three build styles. The XE8712 and XE9712 are single-unit Grace trays with the GPUs on the processor board; the XE9780L and XE9785L are 3 OU IR7000 nodes carrying eight-GPU HGX boards (or AMD Instinct MI355X on the XE9785L) beside Xeon 6 or EPYC 9005 processors; the XE9680L and XE9685L are 4U liquid-cooled HGX servers. The bars use each model’s own spec file.

XE8712XE9712XE9780L/XE9780LAPXE9785LXE9680LXE9685LCores / CPUXE8712: 7272XE9712: 7272XE9780L/XE9780LAP: 128128XE9785L: 192192XE9680L: 6464XE9685L: 192192DIMM slotsXE8712: 00XE9712: 00XE9780L/XE9780LAP: 3232XE9785L: 2424XE9680L: 3232XE9685L: 2424Max memoryXE8712: 0.938 TB0.938XE9712: 0.469 TB0.469XE9780L/XE9780LAP: 4 TB4XE9785L: 3 TB3XE9680L: 4 TB4XE9685L: 3 TB3Max drive baysXE8712: 22XE9712: 88XE9780L/XE9780LAP: 1616XE9785L: 1616XE9680L: 88XE9685L: 88Max GPUsXE8712: 44XE9712: 44XE9780L/XE9780LAP: 88XE9785L: 88XE9680L: 88XE9685L: 88Largest PSUXE8712: 5,500 W5,500XE9712: 5,500 W5,500XE9780L/XE9780LAP: 33,000 W33,000XE9785L: 33,000 W33,000XE9680L: 3,000 W3,000XE9685L: 3,000 W3,000
This modelSiblings and successorEach value from that model’s own verified spec file · bars scale per column
09 — Use cases

Three builds that suit the XE8712

Three tray builds. Our configurator holds a short parts list for the XE8712 — the Grace pair, its soldered LPDDR5X, the four GPUs, the E3.S bays and the 25GbE OCP card — so each build names the rack items and fabric adapters that are added on the quote.

HPC

FP64 simulation tray for a full 36-tray rack

Dell puts GPU-accelerated FP64 HPC — weather, seismic, CFD, genomics, oil and gas — first among this tray’s workloads, and its HPC power table (half-shelf redundancy) is the one that runs to 36 trays in a 50 OU rack on eight shelves. Scratch data belongs on the cluster file system, so this build leaves the E3.S cage blanked and boots from the single M.2 drive.

  • 2 × NVIDIA Grace, 72 cores each — fixed on the board
  • 2 × 480 GB LPDDR5X, soldered — 960 GB per tray
  • 4 × NVIDIA Blackwell Ultra GPUs, 192 GB HBM3e each
  • Mellanox 2 × 25GbE SFP28 OCP 3.0
  • No E3.S drives — cage replaced by a vented blank
  • Red Hat Enterprise Linux 9
  • ProSupport Plus, 5 years

Power shelves, busbar, DLC manifolds, CDU and fabric switches are rack items, and the CX7/CX8 fabric adapters and the M.2 boot drive are specified on the quote because the configurator doesn’t list them for this tray. Dell lists RHEL without a release number, and a rack needs at least three trays.

Open this build in the configurator →
AI training

Scale-out training tray

Dell’s guide says a Llama 3.0 model with 405 billion parameters fits in a single XE8712, which pairs 960 GB of LPDDR5X with 768 GB of HBM3e. Training across trays needs the fabric: Dell lists CX7 or CX8 fabric ports for each tray and names Z9864F-ON (64 × 800 Gb) and Z9664F-ON (64 × 400 Gb) Ethernet or Quantum-2 QM9700 InfiniBand switches for the rack. AI racks follow the N+1 shelf table, which stops at 33 trays.

  • 2 × NVIDIA Grace, 72 cores each
  • 2 × 480 GB LPDDR5X — 960 GB
  • 4 × NVIDIA Blackwell Ultra GPUs — 768 GB HBM3e in total
  • 2 × 7.68 TB EDSFF E3.S NVMe Gen5 for application logs
  • Mellanox 2 × 25GbE SFP28 OCP 3.0
  • Canonical Ubuntu Server 24.04 LTS
  • ProSupport Plus, 3 years

The spec sheet lists the E3.S bays as not available at launch and the guide names no drive capacities, so drives are confirmed at quote. Fabric adapters and switches are priced separately, and Dell lists Ubuntu Server LTS without naming a release.

Open this build in the configurator →
Fine-tuning & inference

Tuning and inference tray with a BlueField-3

AI tuning and inferencing is the third workload Dell names for the tray. A BlueField-3 SuperNIC can go in a front riser, but Dell warns it lifts fan speed: its test tray with a B3240 beside two CX-7 cards measured 67 dB(A) at 25 °C, against 51 dB(A) for the same tray without the DPU. Two larger E3.S drives hold local model files.

  • 2 × NVIDIA Grace, 72 cores each
  • 2 × 480 GB LPDDR5X, soldered
  • 4 × NVIDIA Blackwell Ultra GPUs
  • 2 × 15.36 TB EDSFF E3.S NVMe Gen5
  • Mellanox 2 × 25GbE SFP28 OCP 3.0
  • Canonical Ubuntu Server 22.04 LTS
  • ProSupport, 3 years

The BlueField-3 card and any CX7/CX8 adapters are quoted separately — our configurator doesn’t list them for this tray. Dell recommends ProSupport Plus for PowerEdge XE systems; plain ProSupport is shown here as the lighter option.

Open this build in the configurator →
10 — Full specifications

Dell PowerEdge XE8712 specifications

Tray & cooling

Form factor1 OU compute tray for Dell IR7044 (44 OU) or IR7050 (50 OU) ORv3 21-inch racks
Open Rack unit1 OU = 48 mm high, 537 mm wide
DimensionsTable 2: 46.2 × 537 × 833.82 mm · appendix: 46.45 mm high, 560.07 mm (Xa) or 537 mm (Xb) wide, 788.53 mm hook to chassis rear, 840.85 mm hook to IMM alignment cup
Weight30 kg (Table 2) · 33.07 kg fully loaded (appendix)
CoolingDirect liquid cooling for CPUs and GPUs (left and right cold plates, inner manifolds with leak sensors); air for other parts
FansEight, in four pairs, at the rear · HPR Gold in Table 25 and the spec sheet, “standard” in Table 2
AmbientUp to 40 °C · no CPU or GPU thermal restrictions
RailsStatic L-bracket rails for ORv3 · no cable arm or strain-relief bar

Processor & memory

Processors2 × NVIDIA Grace per tray · 72 cores each · up to 3.1 GHz
Maximum TDP600 W (Table 1) · acoustic configurations list Grace 300 W
CPU memory480 GB LPDDR5X with ECC per CPU, soldered · 960 GB per tray
ExpansionNo DIMM slots · no DIMM or CXL 2.0 memory expansion

Accelerators

GPUs4 × NVIDIA Blackwell Ultra per tray, on the GB200 NVL4 HPM
GPU memory192 GB HBM3e per GPU
CPU–GPU linkNVLink, 900 GB/s of coherent memory
Per rackUp to 144 GPUs at 36 trays (spec sheet)

Storage

Front baysUp to 2 × EDSFF E3.S 1T NVMe, Gen5 x4, hot-plug · not available at launch (spec sheet)
ControllerNone — CPU-direct NVMe via a Gen5 PCIe switch · no HBA, PERC, SAS, SATA or U.2
Boot1 × M.2 on an interposer (slot 5, PCIe Gen4 x4 in Table 21; acoustic set-ups say Gen5 x4), not hot-swap · diskless supported

Expansion & networking

PCIeUp to 4 × FH HL PCIe Gen5 x16 (32 GT/s), front risers only
Riser configurationsRC0, RC5, RC6: 4 slots · RC2, RC7: 3 slots · RC1: 2 slots
OCP1 × OCP 3.0 Gen5 x16 (slot 3, 86 W) · Mellanox 2 × 25GbE SFP28
FabricDell’s wording: “4 x16 CX7 or 2 CX8 OSFP ports”
DPU1 × BlueField-3 SuperNIC, FHHL

Power (rack-level)

Tray PSUsNone — up to 54 VDC from the rack busbar
Power shelfORv3 PS33: 33 kW, 674 A · 6 × 5500 W AC PSUs · 38 kg
Shelf input200–277 V single or three-phase delta · 346/480 V star · 7-pin 60 A inlet
BusbarORv3, 1400 A, 48 V
Per rackUp to 264 kW (spec sheet)

Rack (IR7000)

HeightIR7044: 2290 mm · IR7050: 2578 mm (guide maxima) · spec sheet: 2286 / 2574 mm
Width × depth750 × 1200 mm frame · 1345 mm with doors (spec sheet: 1340 mm)
PayloadIR7044: 1600 kg max, 1515 kg usable · IR7050: 1900 kg max, 1804 kg usable
Tray countMinimum 3, maximum 36 per 50 OU rack
Liquid21-inch UQDB06 blind-mate manifolds · 2 QDs per OU · up to 10 kW per OU
SwitchesManagement: Dell S5248-ON · fabric: Dell Z9664F-ON, Z9864F-ON, NVIDIA Quantum-2 QM9700
rPDUVertiv VP4UU1A0, 3P+N+E IP44 inlet

Ports & warranty

Front1 × USB 3.0 Type-A · 1 × USB 2.0 Type-C (iDRAC Direct) · 2 × RJ45 iDRAC · 1 × Mini-DisplayPort
RearNo I/O — fans, busbar connector and coolant quick-disconnects
WarrantyThree years hardware, standard for PowerEdge XE
11 — FAQ

XE8712 questions, answered from the documentation

Can the Dell PowerEdge XE8712 run as a standalone server?

No. It is a 1 OU compute tray with no power supplies of its own: it takes DC power from the busbar and coolant from the DLC manifolds of a Dell IR7044 or IR7050 Open Rack v3 cabinet, blind-mating to both as it slides in. The rack also carries the power shelves, management and fabric switches and a rack PDU, with a CDU alongside, and Dell requires at least three trays per rack.

How many GPUs and CPUs does each XE8712 tray have?

Two NVIDIA Grace processors with 72 cores each and four Blackwell Ultra GPUs with 192 GB of HBM3e each. Dell rates the NVLink link between CPUs and GPUs at 900 GB/s of coherent memory. The guide lists a single processor and GPU complement, so those parts are the same in every tray.

How many XE8712 trays fit in a rack?

Dell supports between 3 and 36 XE8712 trays in one 50 OU IR7050 rack. At four GPUs per tray, 36 trays give the 144 Blackwell GPUs per rack quoted in Dell’s spec sheet, which also cites up to 264 kW per rack.

How much memory does the XE8712 have?

Each tray has 960 GB of LPDDR5X with ECC — 480 GB soldered to each Grace processor — plus 192 GB of HBM3e on each of its four GPUs. There are no DIMM slots, and Dell supports no DIMM or CXL 2.0 memory expansion, so the capacity you buy is the capacity you keep.

How is the XE8712 powered?

From the rack. ORv3 33 kW power shelves, each holding six 5500 W AC power supplies, feed a 1400 A busbar and the tray takes up to 54 VDC from it. Dell sizes shelves by tray count: eight shelves serve 29–33 trays with N+1 redundancy for AI, or 31–36 trays with half-shelf redundancy for HPC.

Is the XE8712 liquid-cooled?

Yes, for the CPUs and GPUs. Two cold plates connect through inner manifolds to quick-disconnects at the rear of the tray, which blind-mate to the rack’s 21-inch DLC manifolds, and each OU can hand up to 10 kW to the liquid. Eight HPR Gold fans still cool the remaining parts, with ambient air allowed up to 40 °C.

What storage does the XE8712 support?

Up to two hot-plug EDSFF E3.S NVMe drives at the front, each on a Gen5 x4 link through a PCIe switch, and one M.2 boot drive on an interposer; diskless trays are supported too. There is no SAS, SATA or U.2 support and no HBA or PERC, and Dell’s spec sheet lists the E3.S bays as not available at launch.

What PCIe slots and networking does the XE8712 have?

Up to four full-height, half-length PCIe Gen5 x16 slots on front risers, with the count set by the riser configuration (Dell documents six), alongside one OCP 3.0 Gen5 x16 slot. For the fabric Dell lists CX7 x16 or CX8 OSFP options, plus one BlueField-3 SuperNIC, and a Mellanox 2 × 25GbE SFP28 OCP card.

How loud is the XE8712?

Dell describes it as a server for unattended data centres and advises against deploying it near people. In its tests the tray measured 51 to 67 dB(A) at the bystander position at 25 °C, depending on configuration, and 64 to 79 dB(A) under maximum load at 35 °C. The guide calls for industrial-grade ear protection when working near it.

Is the Dell PowerEdge XE8712 end of life?

Dell’s technical guide gives no end-of-sale or end-of-service date for the XE8712, and PowerEdge XE servers come with a three-year hardware warranty that ProSupport can extend. Our server end-of-life checker records vendor dates when Dell publishes them.

Sources & verification

Where these figures come from

  • Dell PowerEdge IR7000, XE8712 Technical Guide (Rev. A01, January 2025) — Tables 1–2 (features), Tables 3–4 and Figures 1–4 (views; gallery images from PDF pages 8–10), rack Tables 5–6, busbar and manifold overview, power-shelf Tables 7–15, switch and rPDU Tables 16–20, memory and storage chapters, riser Table 21, OCP Tables 22–23, thermal Table 25, acoustic Tables 27–28, iDRAC10 and management Tables 29–30, dimensions and weight Tables 31–32, service appendix.
  • Dell PowerEdge XE8712 Specification Sheet (November 2025) — Per-rack GPU count and power (144 GPUs, 264 kW), E3.S launch availability, rack and tray dimensions, OS list and security features.

Specifications are extracted from Dell’s published documents and independently re-checked line by line. Configurations are quote-based; availability and exact options are confirmed at quotation. Last reviewed 2026-09-11.

Get a quote for the Dell PowerEdge XE8712

Tell us the workload and we’ll spec and price it — or build it in the configurator first.

Business enquiries only — please use your company email. Gmail, Hotmail and other personal addresses are not accepted.

We only ship hardware to the countries in the dropdown. Software and maintenance services are global.

Enquiring about: Dell PowerEdge XE8712

Response within 4 business hours · No obligation

Quote for
Dell PowerEdge XE8712
0800 987 4111

Talk to a UK specialist

Get expert advice or a no-obligation quote — servers, storage, networking, maintenance, finance and cloud. We reply the same working day.

or call 0800 987 4111