HPE GPU Server Solutions

Unlock the full potential of your AI, machine learning, and high-performance computing workloads with GPU servers from HPE ProLiant. Purpose-built for demanding data analytics, deep learning, and HPC environments, these enterprise-grade systems deliver exceptional processing power, memory bandwidth, and scalability. Whether you're training large language models, running real-time inference, or managing complex data pipelines, serverandparts.com has the HPE configuration to match your workload and budget. Browse our full selection of in-stock HPE GPU servers and find the performance your business needs — without the wait.

Shop Now

Filters

Price Range

Min ($)

Max ($)

$0$10,000+

Memory4

16 DIMM slots (16 DIMM channels per processor), up to 2TB total DDR5, speeds up to 5600 MT/s; only HPE DDR5 5600 MT/s SKUs can be selected with 5th Generation Intel processors; registered ECC DDR5 only124 DDR5 DIMM slots, 12 DIMM channels per processor, up to 3TB total, speeds up to 5600 MT/s; HPE DDR5 Smart Memory132 DIMM slots, HPE DDR5 Smart Memory, up to 4TB total (2TB per processor); RAS features include Advanced ECC, online spare, mirroring, combined channel and ADDDC16 DDR5 DIMM slots; DDR5 speeds of 4800 or 6400 MT/s1

Networking1

Supports the latest NVIDIA InfiniBand, Ethernet and BlueField adapters, including NDR/NDR200 InfiniBand and BlueField DPU options; NVIDIA OVX certified1

Form Factor5

Acoustics1

LWA,m 9.5 B; LpAm 80 dBA (tested in performance configuration at 100% CPU and GPU TDP)1

Air Filter1

Bezel filter kit with removable air filter; HPE recommends replacement every three to six months (consumable, not covered under warranty)1

CPU Cores1

72 Arm Neoverse V2 cores (GH200) or 144 Arm Neoverse V2 cores (GH200 NVL2); base frequency 3.1GHz, all-core SIMD frequency 3.0GHz1

More technical filters41

Cache1

64KB i-cache + 64KB d-cache L1; 1MB L2 per core; 114MB L3 (GH200) or 228MB L3 (GH200 NVL2)1

Chassis2

10DW (double-wide GPU) chassis with one drive bay for drive cage options; ships with one 4SFF x4 U.3 NVMe drive cage by default14DW (double-wide GPU) chassis with one drive bay for drive cage options; ships with an 8SFF x4 U.3 NVMe drive bay by default1

Compute Performance (GH200 NVL2)1

FP64 68 teraFLOPS; FP64 Tensor Core 134 teraFLOPS; FP32 Tensor Core up to 1979 teraFLOPS; BFLOAT16 Tensor Core up to 3958 teraFLOPS with sparsity; FP8 Tensor Core up to 7916 teraFLOPS with sparsity; INT8 Tensor Core up to 7916 TOPS with sparsity1

Cooling4

Air cooled1Air cooled, front to back; 52 CFM idle entry, 133 CFM idle typical with DPU, 240 CFM operating maximum at 25C ambient1Air cooled, with a thermal design tuned for dense GPU configurations1Air cooling, with support for Direct Liquid Cooling (DLC)1

Cooling and Environment1

Continuous operation from -5C to 55C; withstands high-dust and high-vibration environments; designed to qualify for NEBS Level 31

Dimensions2

Approximately 15 inches wide; 16 inches deep without bezel, 20 inches deep with bezel filter1Height 87.5mm (3.44"), Width 448mm (17.64"), Depth 789mm (31.08")1

Drive Bays3

8 SFF (SAS/SATA/NVMe) with optional Universal Media Bay; 4 LFF with optional optical disk drive and DisplayPort/USB kit; 8 EDSFF x4 NVMe drive cage option1Up to 2 SFF HDD/SSD, or up to 6 EDSFF E3.S drives; self-encrypting drives supported1Up to 8 EDSFF NVMe Gen5 E3.S drives in a dual-superchip GH200 NVL2 configuration; maximum internal storage 122.8TB (8x 15.36TB). Hardware RAID controllers and hot-pluggable drives are not supported in this generation1

Embedded Management5

HPE iLO 6 ASIC, iLO RESTful API with Redfish, UEFI Class 3 (UEFI boot mode only, no legacy mode), HPE Active Health System, RESTful Interface Tool (iLOREST)1HPE iLO 6 with Intelligent Provisioning; HPE Compute Ops Management1HPE iLO 6; HPE Compute Ops Management1HPE iLO 6; HPE Compute Ops Management; HPE OneView1HPE iLO 7 with a redesigned GUI and Redfish API; HPE Compute Ops Management; HPE OneView Standard, with optional iLO Advanced and OneView Advanced1

Embedded NIC1

Broadcom BCM5720 Ethernet 1Gb 2-port BASE-T Adapter standard, upgradeable with a range of networking options1

Expansion1

PCIe Gen51

Fans3

4 hot-plug fan assemblies, each consisting of one dual-rotor 92x56mm fan and two single-rotor 40x28mm fans; a 110W fan upgrade is recommended to cool higher-temperature 600W GPUs14 standard or high-performance hot-plug fans, rear-mounted, drawing air through the front filter bezel16 dual-rotor hot-plug fans with N+1 redundancy by default1

Field Upgrades1

No field upgrade kits. HPE does not support changes to the number of superchips, drives, PCIe or OCP slots outside the factory1

GPU Interconnect1

NVIDIA 2-way NVLink Bridge supported between two adjacent GPUs to increase inference performance1

GPU Support4

Up to 10 double-wide GPUs at up to 600W each, or up to 16 single-wide GPUs1Up to 3 single-wide GPUs, or 1 double-wide GPU1Up to 4 double-wide or 8 single-wide GPUs. Note: 0-GPU configurations are not allowed, and GPU installation after shipment must be performed by HPE Services1Up to 4 single-wide or 2 double-wide GPUs in the front cage with the GPU CTO server1

High-Bandwidth Memory1

144GB HBM3e up to 4.9TB/s (GH200) or 288GB HBM3e up to 9.8TB/s (GH200 NVL2)1

Internal Boot3

HPE NS204i-u Gen11 NVMe hot-plug Boot Optimized Storage Device1Hot-pluggable or internal high-availability RAID1 NVMe M.2 boot options1Up to two M.2 devices (PCIe Gen3); recommended for boot rather than local cache. Boot options set in HPE iLO from M.2 NVMe, EDSFF NVMe, or network PXE1

LPDDR5X Memory1

480GB at 512GB/s (GH200) or 960GB at 1024GB/s (GH200 NVL2)1

Maximum Heat Dissipation1

3048 Watts (10.4k BTU/hr) for a fully populated GH200 NVL2 system1

Maximum Weight (fully populated)1

Single superchip 26kg (58lb); dual superchip 31kg (69lb)1

Memory Population2

DIMMs must be selected in quantities of 1, 2, 4, 6, 8, 12 or 16; DIMM blanks required in each empty slot1Intel Xeon 6700-Series with E-cores limits DIMM quantities per CPU to 1, 2, 4, 8 and 16; 6500/6700-Series with P-cores allows 1, 2, 4, 8, 12 and 161

Mounting1

Wall, desk and rack mounting options; ruggedised option kit available1

NVIDIA NVLink-C2C CPU-to-GPU Bandwidth1

900GB/s (GH200) or 1800GB/s (GH200 NVL2)1

Network Options1

Choice of OCP or stand-up card, supporting a range of NIC adapters1

Operating Temperature1

10C to 35C (50F to 95F) at sea level, with altitude derating of 1.0C per 305m above sea level to a maximum of 3050m; extended ambient ranges supported on approved hardware configurations1

PCIe Slots4

6 PCIe Gen5 expansion slots; refer to HPE QuickSpecs for slot configuration rules1Up to 2x x16 PCIe Gen5 slots plus 1x OCP 3.0 slot1Up to 4 full-height half-length x16 Gen5 PCIe slots, or 2 PCIe plus 2 OCP 3.0 x16 slots, depending on factory configuration1Up to 4x x16 PCIe Gen5 slots plus 2 OCP slots; primary riser carries PCIe 5.0 slots 2 and 3, secondary riser carries slots 5 and 61

Ports — Front1

1x iLO Service Port, 1x USB 3.2, front panel LEDs for Unit ID, NIC status, health and power1

Ports — Rear1

2x USB 3.0, 1x 1Gb iLO dedicated NIC, 1x serial, 1x VGA, 1x rear Unit ID indicator1

Power1

Three independent power domains, two of which are dedicated to the GPUs, delivering power efficiently and reliably to both the system and the accelerators1

Power Supplies4

500W Flex Slot Platinum; 1000W Flex Slot Titanium (high line 200-240VAC only); 1800W-2200W Flex Slot Titanium; hot-plug. Mixing two different power supplies is not supported1HPE 1800W-2200W Flex Slot Titanium hot-plug, up to 96% efficiency, C-14 inlet. Single GH200: two, three or four supplies (N+0, N+1, N+2); dual GH200: four supplies (N+1). Rated line voltage 200-240VAC per supply1HPE Flex Slot hot-plug power supplies, multiple output options for right-sizing; standard 6-foot IEC C-13/C-14 jumper cord included with each standard AC power supply option kit1HPE ProLiant DL110 700W Flex Slot Platinum (94% efficiency); 700W Flex Slot -48VDC; 900W-1000W Titanium (96% efficiency); all reverse airflow, hot-plug1

Processor4

One 4th or 5th Generation Intel Xeon Scalable processor (5nm), up to 60 cores, up to 270W TDP1One AMD EPYC 8004 (Siena, Zen4c) processor, 8 to 64 cores, TDP up to 200W; or one AMD EPYC 8005 (Sorano, Zen5) processor, 8 to 84 cores, TDP up to 225W1Two 4th or 5th Generation Intel Xeon Scalable processors, up to 64 cores per processor, up to 350W TDP1Two Intel Xeon 6 processors (6500 Series with P-cores, or 6700 Series with P- or E-cores), up to 144 cores, up to 250W TDP1

Recommended Support1

HPE Tech Care Service or HPE Complete Care Service for enhanced response and proactive support5

Security5

HPE Silicon Root of Trust anchoring server firmware to the Intel Secure Processor, bezel lock kit, intrusion detection cable kit, TPM1HPE Silicon Root of Trust with zero-trust firmware integrity checking, default IDevID certificates, embedded TPM 2.0, optional locking bezel kit, optional intrusion detection kit; iLO 7 supports quantum-resistant (PQC) algorithms and signs Gen12 firmware with CNSA 2.0-approved algorithms1HPE Silicon Root of Trust, HPE iLO 6, IDevID, Platform Certificates, chassis intrusion detection, bezel lock kit and Kensington lock, TPM1HPE Silicon Root of Trust, Secure Boot and Secure Start1HPE Silicon Root of Trust, Secure Boot, TPM 2.0, IDevID, Platform Certificates, chassis intrusion detection1

Standard Warranty1

3-year parts, 3-year labour, 3-year onsite support with next business day response5

Storage1

SFF NVMe and EDSFF (E3.S) drive options; NS204i-u front cage kit available for two hot-pluggable NVMe M.2 boot devices1

Storage Controllers1

HPE MR216i-p Gen11 and MR416i-p Gen11 plug-in controllers; HPE Smart Array E208e-p SR Gen10 external controller1

Superchip1

NVIDIA GH200 Grace Hopper Superchip; single-superchip or dual-superchip NVIDIA GH200 NVL2 configuration with NVLink between the two GH2001

Supported Operating Systems1

Initial certification plans cover Red Hat Enterprise Linux 9.4, Canonical Ubuntu 24.04, and SUSE Linux Enterprise Server 15 SP5 QU11

Target Workloads2

Enterprise AI inferencing at scale, focused fine-tuning, high-performance computing1Machine learning, deep learning, AI training and inference, advanced engineering applications, graphics-intensive workloads1

Top Workloads1

Edge AI and computer vision, containers, hybrid cloud, VDI, virtualisation1

Validated GPU Options1

NVIDIA RTX PRO 6000 Blackwell Server Edition (up to 8 double-wide), NVIDIA H200 NVL, NVIDIA H100 NVL, NVIDIA L40S, NVIDIA L20, NVIDIA L4 (single-wide)1

Video1

Video modes up to 1920 x 1200 @ 60Hz (24 bpp)1

Showing 1-5 of 5 results

Sort by: