Aqron Serv
AI80 D8G
Up to eight 600 W PCIe GPUs, two 5th Gen Intel Xeon processors and up to 6 TB of memory in one 4U server — for AI training, inference and compute-intensive workloads. Shown with eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs.
Up to 8 GPUs,
600 W each
Eight PCIe 5.0 x16 slots for double-width GPUs — from NVIDIA RTX 4500 Ada to H200 NVL and RTX PRO 6000 Blackwell Server Edition.
Up to 6 TB
DDR5 ECC
24 DIMM slots, 8 memory channels per CPU. Up to 5600 MT/s.
Storage and
expansion
8 hot-swap E3.S NVMe bays, dedicated M.2 boot carrier, 8 extra PCIe 5.0 x8 slots and OCP 3.0 for up to 400GbE.
Redundant power
and cooling
Four hot-swap Titanium PSUs in 2+2 redundancy on independent A/B feeds. Hot-swap fan modules, front-to-rear airflow.
Reference
configuration
The 8-GPU build shown on this page. Every component can be changed within the platform options listed in the specifications.
8× RTX PRO 6000 Blackwell, 2× Xeon Platinum 8558P
Eight 600 W GPUs with two 48-core processors, 2 TB of memory and 122.88 TB of NVMe storage. Uses high-performance cooling and four 3,200 W power supplies, as required for 8 × 600 W GPUs.
- Processors
- 2× Intel Xeon Platinum 8558P · 48 cores / 96 threads, 2.70 GHz, 350 W
- Memory
- 2 TB — 16× 128 GB DDR5-6400 ECC RDIMM, 2Rx4
- GPU
- 8× NVIDIA RTX PRO 6000 Blackwell Server Edition, 96 GB GDDR7 ECC, passive, 600 W
- Storage
- 8× 15.36 TB E3.S PCIe Gen5 NVMe, 1 DWPD, with PLP — 122.88 TB raw
- Boot
- 2× 960 GB M.2 22110 NVMe, SED / TCG Opal
- Network
- 2× 10GbE RJ45 onboard · Intel E810-XXVDA2 OCP 3.0, 2× SFP28 25GbE
- Power
- 4× 3,200 W 80 PLUS Titanium hot-swap PSUs, 2+2 redundancy
- Cooling
- High-performance: 12 front and 4 rear hot-swap fan modules
- Rack
- Tool-less sliding rail kit
- Software
- No operating system pre-installed
With GPUs above 350 W, maximum operating inlet temperature is 30 °C.
Data centre
requirements
What the 8 × 600 W GPU build needs from your rack, power and cooling.
- Rack space
- 4U, 19-inchRail kit included in the reference configuration; cable management arm available as an option.
- Power
- 200–240 V AC, A/B feedsFour hot-swap Titanium PSUs, two per independent feed. With 4× 3,200 W, total system load must stay within 6,400 W to keep 2+2 redundancy.
- Inlet temperature
- Up to 30 °CFor GPUs above 350 W. Platform range 10–35 °C.
- Cooling
- Front-to-rear airflowHigh-performance cooling with redundant hot-swap fan modules, required above 300 W per CPU or per GPU.
- Network
- 2× 10GbE RJ45 onboardPlus OCP 3.0 (PCIe 5.0 x16) and 8 PCIe 5.0 x8 slots for adapters up to 400GbE or InfiniBand HDR.
- Management
- Dedicated 1GbE portAqron iMC out-of-band management: Redfish, IPMI 2.0, SNMPv3, HTML5 console, LDAP / Active Directory.
Engineered in Taiwan.
Supported in the GCC.
AQRON designs, assembles and tests its hardware in Taiwan. Sales, technical support and installation are handled by our GCC offices.
NVIDIA RTX PRO 6000
Blackwell Server Edition
Passive, dual-slot data-centre GPU on the NVIDIA Blackwell architecture. Eight cards per system, each with 96 GB of ECC GDDR7 and support for Multi-Instance GPU.
RTX PRO 6000 Blackwell SE
Technical
specifications
Platform specifications. Available options depend on the number of installed CPUs and the cooling configuration.
/ 01Processor
| Sockets | 1 or 2 fifth-generation Intel Xeon Scalable processors, FCLGA4677; dual-CPU configurations use the same CPU model |
| Supported CPUs | Silver 4514Y (16C, 2.00 GHz, 150 W) · Silver 4516Y+ (24C, 2.20 GHz, 185 W) · Gold 6542Y (24C, 2.90 GHz, 250 W) · Gold 6548Y+ (32C, 2.50 GHz, 250 W) · Gold 6554S (36C, 2.20 GHz, 270 W) · Platinum 8562Y+ (32C, 2.80 GHz, 300 W) · Platinum 8558P (48C, 2.70 GHz, 350 W) · Platinum 8568Y+ (48C, 2.30 GHz, 350 W) · Platinum 8570 (56C, 2.10 GHz, 350 W) · Platinum 8592+ (64C, 1.90 GHz, 350 W) |
| Note | Base clock and TDP per CPU, at the default profile |
/ 02Memory
| DIMM slots | 24; 12 slots per installed CPU, 8 memory channels per CPU |
| Type | DDR5 ECC RDIMM / 3DS RDIMM, multi-vendor |
| Module capacities | 32 · 64 · 96 · 128 · 256 GB |
| Maximum capacity | 6 TB with 24× 256 GB; up to 3 TB with one CPU |
| Speed | Up to 5600 MT/s; depends on CPU, DIMM type and population |
| Population | 8 equal DIMMs per CPU, one per channel; matched between CPUs |
/ 03GPU
| Slots | Up to 8 double-width PCIe GPUs, up to 600 W per GPU; 8 GPUs require two CPUs |
| Options | NVIDIA RTX 4500 Ada Generation 24 GB, 210 W · RTX 6000 Ada Generation 48 GB, 300 W · RTX PRO 6000 Blackwell Server Edition 96 GB, 600 W · H100 PCIe 80 GB, 350 W · H200 NVL 141 GB, 600 W |
| NVLink | H200 NVL supports 2-GPU or 4-GPU NVLink groups; bridge kit selected to match GPU slot spacing |
| Kits | GPU cooling and power kits follow GPU model |
/ 04PCIe expansion
| GPU slots | 8× PCIe 5.0 x16 |
| Add-in slots | 8× low-profile PCIe 5.0 x8 |
| OCP | 1× OCP 3.0, PCIe 5.0 x16 |
| Topology | GPU and add-in slots connect through PCIe switches; CPU uplink bandwidth is shared |
| One-CPU builds | Up to 4 GPU slots and 4 add-in slots; two CPUs enable all slots |
/ 05Storage
| Drive bays | Up to 8 hot-swap E3.S NVMe SSDs, PCIe 4.0 / 5.0 x4 per drive depending on SSD model |
| Drive options | 1.92 · 3.84 · 7.68 · 15.36 TB enterprise SSD |
| Raw capacity | Up to 122.88 TB with 8× 15.36 TB; usable capacity depends on RAID and formatting |
| Features | Power-loss protection; optional self-encrypting drives, TCG Opal support depends on SSD model; software RAID depends on OS |
/ 06Boot storage
| Carrier | Dedicated M.2 22110 NVMe carrier for 1 or 2 SSDs, separate from front drive bays |
| Drive options | 480 GB · 960 GB · 1.92 TB · 3.84 TB enterprise M.2 SSD |
| Mirroring | Two-drive RAID 1 boot mirror where supported by OS; ESXi needs a compatible boot RAID solution; optional SED / TCG Opal by SSD model |
/ 07Networking
| Onboard | 2× 10GbE RJ45 |
| Optional adapters | 10/25GbE SFP28 · 100GbE QSFP28 · 200GbE QSFP56 · 400GbE QSFP-DD (OCP 3.0 / PCIe 5.0 x16) · InfiniBand HDR 200 Gb/s QSFP56 |
| x16 adapters | Choose an OCP 3.0 model or fit a PCIe card in a free GPU slot; GPU slot use reduces GPU quantity |
| RDMA | RoCE v2 on supported Ethernet adapters; optics and cables available separately |
/ 08Power
| Power supplies | 4 matching hot-swap Titanium-efficiency PSUs, 2+2 redundancy |
| PSU rating | 4× 2,000 W · 4× 2,400 W · 4× 3,200 W |
| Input | 200–240 V AC; 2 PSUs connected to each independent A/B power feed |
| Limits | Total configured system load must stay within the rated output of one 2-PSU bank. For 8× 600 W GPUs: 4× 3,200 W PSUs and high-performance cooling; full system power budget within 6,400 W to maintain 2+2 redundancy |
/ 09Cooling
| Standard | Up to 300 W per CPU / 300 W per GPU |
| High-performance | Up to 350 W per CPU / 600 W per GPU; required above standard limits |
| Airflow | Redundant hot-swap fans, front-to-rear airflow |
/ 10Management & security
| BMC | Aqron iMC, dedicated 1GbE RJ45 port for out-of-band management |
| Features | HTML5 remote console and virtual media · Redfish API, IPMI 2.0 and SNMPv3 · remote BIOS / BMC update and event logs · power, temperature and fan monitoring · role-based access, LDAP / Active Directory |
| Security | TPM 2.0, UEFI Secure Boot and signed firmware updates |
| Local I/O | 2× USB 3.2 Gen 1, VGA and serial console |
/ 11OS & hypervisor
| Options | No OS, or Ubuntu Server · Red Hat Enterprise Linux · Rocky Linux · Windows Server · VMware ESXi |
| Note | Release, driver and GPU feature support depend on selected configuration |
/ 12Rack, environment & service
| Chassis | 4U, 19-inch rack |
| Rails | Optional rails and cable management arm |
| Inlet air | 10–35 °C; maximum 30 °C for GPUs above 350 W; limit depends on configuration |
| Support | 1-, 3- or 5-year support |
Specifications are subject to change without notice. NVIDIA, RTX and Blackwell are trademarks of NVIDIA Corporation. Intel and Xeon are trademarks of Intel Corporation.
Request
AI80 D8G
Tell us what you plan to run and where it will be installed. We will propose a configuration from the platform options and send a commercial proposal.