UP  TO 2X EPYC 7003

128 CORES

DESIGNED FOR

RTX5090

POWERED BY
MAX MEMORY

8192 GB DDR4

PCI EXPRESS 4.0

16 GT/s

DESIGNED FOR

4U-E10G-ROMEX2

The Most Versatile 4U GPU Server Solution

#Ai #ML #HPC

The ultimate HPC/AI server designed to supercharge your computing capabilities. Powered by AMD EPYC™ 7003 processor, this server delivers unparalleled performance, making it the perfect choice for high-performance computing (HPC) and artificial intelligence (AI) applications.

Featuring support for up to 12 GPUs depending on how configured. Supports up to 12x GPU at PCIe 4.0 X8 link or 6x GPU at PCIe 4.0 X16 link. Built to work with even the longest and thickest GPUs.

Technical Specifications:
  • 4U Rackmount 19” inch form factor
  • Dual Socket SP3, supports AMD EPYC™ 7003
  • 32x DIMM slots (2DPC), MAX 8192GM DDR4 RDIMM
  • Up To 10 Dual-Slot PCIe4.0 X16* (GPU 9,10 @ X8)
  • 8x Hot-Swap 3.5” drive bays (OPTIONAL)
  • 2x Hot-Swap U.2 NVMe drive bays
  • Supports 1x M.2 (PCIe4.0 x4) (UPGRADABLE)
  • Front Or Rear AC Power Input
  • 6400W (3+1 1600W Platinum CPRS)
  • 2x RJ45 (10GbE) by Intel® X550-AT2
  • Remote Management (IPMI)
  • 6x Hot-Swap Fans
    • 3x 7200RPM Counter rotating
    • 3x 7200RPM High Static Pressure Fans
Dimensions: 940x 440 x 176.5 mm

Base Configuration price including CPU’s & RAM:

From: $4,908.80

Configure & Order Online

Maximum Performance

Achieve the highest performance with two AMD EPYC™ 7003 CPUs and up to 8 RTX 5090 AI GPUs for AI, ML, and HPC computations. Maximize the PCIe 4.0 and high-speed DDR4 memory to reach the peak efficiency.

Up-To 5 Year Warranty

Peace of mind  knowing that you are covered for up to 5 years warranty. A mission critical product, OCTOSERVER guarantees reliability, durability and dedicated support for your AI and HPC needs.

High Density Compute

Get the maximum compute density in a 4U housing with 8 GPUs and 32 DIMM slots for the memory, which can hold up to 8TB of RAM. The perfect choice for data-heavy applications of AI, ML and HPC that require a lot of parallel computing.

Superior Cooling

Stay cool even when you are pushing your system to the limit with 6 hot-swap fans,featuring high-static pressure and counter-rotating fans. Designed for high density computing, the OCTOSERVER guarantees the best possible airflow and system stability for any workload.

Cost Efficient

Achieve more with less through efficient power usage, toolless design of the components and high performance computing. Get the best value for your investment on AI and HPC infrastructure that provides best in class pricing.

Fast & Flexible

Grow without a hitch with snap in modules for storage, flexible networking, and remote IPMI control. The 4U-E10G-ROMEX2 was created for quick deployment, future growth, and easy integration into your data center.

Unleash the Power of RTX 5090 for AI, HPC and Many Other Applications!

The OCTOSERVER 4U-E10G-ROMEX2 is an enterprise-class, high-performance GPU server that is meant to provide an extreme computational capability for a wide range of applications including AI, deep learning, HPC and 3D rendering by taking full advantage of high power GPUs like the NVIDIA RTX 5090. The OCTOSERVER 4U-E10G-ROMEX2 has been built with a 4U chassis, which is important that cold air can circulate properly and hot air is removed effectively from around high power components. The OCTOSERVER 4U-E10G-ROMEX2 comes with dual 10GbE networking, PCIe Gen 4 expansion, and support for high-speed NVMe and optional SATA storage, which provides unprecedented levels of bandwidth and scalability for data intensive computing. To ensure it can easily be integrated into any data center, it includes redundant power supplies, enterprise security features, and remote management. Whether for the training of AI models, large scale simulations or any kind of creative production, this powerhouse server is built to push the computational boundaries.

Unmatched EPYC Power for the Most Demanding AI and HPC Workloads!

Powering the OCTOSERVER 4U-E10G-ROMEX2 is the AMD EPYC™ 7003 series, which is a processor family that is well known for its high performance and energy efficiency in enterprise-class computing systems. These CPUs are intended to work optimally with high-end dual slot NVIDIA RTX 4090 and RTX 5090 GPUs, and are packed with industry leading core density, high memory bandwidth and PCIe Gen 4 compatibility – which makes them suitable for use in AI workloads, deep learning, and large data processing. Customers in search of a good value for money and at the same time powerful system for their data center requirements will find the stability and scalability of the EPYC 7003 family to be a good match, thus enabling the best application performance with DDR4 memory and security features. The OCTOSERVER 4U-E10G-ROMEX2, powered by AMD EPYC, is ready to solve the most complex computational tasks with high efficiency, reliability and raw power – whether it is for intensive AI models or tough simulations.

Industry Leading Cooling Performance

To achieve this, the OCTOSERVER 4U-E10G-ROMEX2 has been developed with world-class cooling capacity to ensure that thermal management is always at its best, especially during high computational loads. This server is designed to support high power draw GPU’s like the NVIDIA RTX 4090, RTX 5090, H100 and H200 GPUs and AMD EPYC™ processors. It has a high efficiency airflow system that pushes cold air where it is needed and pulls hot air out of the system quickly, to prevent thermal throttling. The cooling system is also enhanced by high-performance counter rotating fans, optimal heat sink arrangement, and intelligent airflow streams to maintain peak performance at all times. This enterprise-class cooling design not only enhances the service life of the hardware but also enhances compute performance, thus making the OCTOSERVER 4U-E10G-ROMEX2 the right system for AI training, HPC, and data center applications that demand reliability and performance.

Built To Perform

Accelerate performance for HPC and AI workloads

4U High Density GPU Server

The Octoserver 4U-E10G-ROMEX2 is a dual-socket AMD EPYC™ 7003 server that is designed for enterprise AI infrastructure and provides outstanding performance along with the highest level of power GPUs, interconnects, and bandwidth. It can hold up to 8 dual slot active or passive GPUs in a scalable arrangement and has options for either NVIDIA NVLink® Bridge or AMD Infinity Fabric™ Link to boost the bandwidth and performance. This server is designed for AI and HPC workloads and constructed to solve the most complex computing challenges with ease.

Eight dual-slot GPU in 4U

Versatile Computing Accelerator Support. OCTOSERVER servers are rigorously tested and optimized for all kinds of workload and made to work seamlessly with high-performance computing accelerators. These servers, which are designed for 3D rendering, deep learning and HPC, equip content creators, data scientists, researchers and engineers with the tools they need to solve complex problems they once couldn’t even dream about. Our 4U chassis is ready for even the most powerful GPU like RTX 4090, RTX 5090, H200 and RTX A6000 PRO. We also support the entire PCIe AMD Instinct range of GPUs.

Product Overview

4U-E10G-ROMEX2

Wide Software Ecosystem Support

Our systems provide seamless integration with major software ecosystems, ensuring optimal compatibility, reliability, and performance in enterprise deployments. Our hardware is rigorously tested with Red Hat, Windows Server, VMware, SUSE, Ubuntu, and Citrix, allowing organizations to run mission-critical applications without compatibility concerns. Whether for virtualized environments, cloud infrastructures, or high-performance computing, OCTOSERVER guarantees flawless interoperability with these trusted software platforms.

Highly Optimized Power Delivery

To achieve the stability and reliability of the enterprise workloads, OCTOSERVER is designed with high efficiency power supply systems. Our power architecture is incorporates 80 PLUS Platinum and Titanium certified power supplies that guarantee maximum energy efficiency.
Our advanced PCB-based power delivery system is engineered for maximum efficiency, reliability, and ease of maintenance. This design reduces cable clutter, improving airflow for better cooling and system stability, with thick copper 4oz and 8oz PCB’s. From streamlining power transmission, reducing resistance and voltage drops, it enhances efficiency and ensures a more reliable power supply for high-performance computing workloads, such as AI and HPC. Furthermore, it simplifies system maintenance, including faster and more efficient system upgrades, troubleshooting, and repairs, with the reduced reliance on traditional power cables and less potential points of failure.

Easy ServicQeability & Optimized Design

Maximum Signal Integrity (SI)

 

Our GPU backplanes are designed with great care and are fully optimized to meet official PCI-SIG standards for maximum compatibility and performance. Our PCB design has the shortest possible trace lengths to minimize signal degradation and provide better signal integrity. This improves the GPU communication reduces the latency and maximize the efficiency and reliability.

Optimized Power Delivery

 

Our GPU backplanes are designed with much care and are fully optimized to meet official PCI-SIG standards for best compatibility and performance. Our PCB design has the shortest possible trace lengths to minimize signal degradation and to gain better signal integrity. This enhances the GPU communication, reduces the latency and thus improve the efficiency and reliability.

Hot-Swap Drive Modules

 

Our hot-swap U.2 and SATA drive modules offer effortless storage expansion and dependability. Designed for enterprise and high-performance computing, these modules allow for effortless and instantaneous drive replacement without system downtime. With a solid connection and high-speed data transfer, they guarantee the best performance and simplicity of repair for critical applications.

 

Industry Leading Cooling Performance

OCTOSERVER was developed with the best thermal management solutions that guarantee the highest level of cooling for enterprise-level load. We have an advanced airflow design and a modular fan architecture that avoids overheating of the system, increases its lifespan and delivers optimal performance even under heavy loading. With hot-swapable cooling components, our cooling system is designed for continuous operation, low-maintenance and higher energy efficiency. Optimized Cooling Components for Peak Efficiency

Optional Rear Fan Module (40x28MM)

The compact form factor of the Dual 40mm Fan Module comes with a high-speed airflow which is ideal for cooling densely packed server components. These fan modules were created for use in systems that are configured with an optional 10 GPU configuration.

Counter Rotating Dual Motor Central Fans (120x76MM)

High density server configurations require a high-static pressure design to deliver powerful cooling efficiency and uniform airflow with Counter Rotating Dual Motor Central Fans (120x76MM). These fans are designed for top reliability and keep the cooling performance consistent, whether it’s under the most demanding workload or not and are also fully hot-swappable.

Rear Exhaust Hot Swap Fan Module (120x38MM)

Rear Exhaust Hot Swap Fan Modules provide toolless cooling continuity with replacement without the need to power down the system. These fan modules at the rear enhance the systems static airflow pressure and guarantee that even the most powerful of GPUs are operating at optimum temperatures.    With the central fans, they form a highly efficient ‘push pull’ fan configuration.

Options for Flexible Chassis Expansion for Maximum Flexibility

 

Boost your OCTOSERVER’s performance with easily attachable storage and connectivity modules well suited for enterprise workloads that require scaling. Our modular design ensures that the integration of the extra HDD storage and additional PCIe devices through the use of optional PCIe switch modules is seamless, resulting in high data processing, virtualization and computing capacity. For your high-speed SSD arrays or high capacity HDDs needs, OCTOSERVER is the solution for flexibility, efficiency and reliability.

Option #1: PCIe 4.0 Switch + 8x 2.5 U.2 NVMe Expansion

The 52 lane PCIe Switch + 8× 2.5″ HDD Expansion module offers the option to expand the chassis with additional U.2 NVMe hot-swap drives and up to 4x Half-Height, Half-Length PCIe add-in-Cards. 

Option #2: 8x 3.5 HDD Expansion

HDD Expansion module is for those organizations that require a big amount of storage and have a primary concern with reliability and efficiency. This module is compatible with high-capacity enterprise HDDs and is ideal for data centers, archival storage, and other applications that require large amounts of data storage. This expansion is built for simple scalability and serviceability so that your OCTOSERVER is prepared for future workload demands.

Efficient Remote Management Console

For proper server and cluster management the OCTOSERVER Management Console that is installed by users can be used. It means that IT staff can monitor and control the server health in real-time using this browser based, graphical interface.

It provides:

      • Compatibility with standard IPMI specifications that enables to integrate services easily into a single platform through an open interface.
      • Included are features like HTML5 iKVM with remote media.
      • Integration of SAS/SATA/NVMe devices and RAID controller firmware to monitor and control the Broadcom® MegaRAID adapters from the console.

Maximum Compute Density, Reliability and Cooling Performance In 4U Chassis

Extreme GPU Compute Density

Because of OCTOSERVER’s GPU-optimized design, it can support up to ten* active or passive dual-slot GPUs for the best performance and scalability in AI, HPC and deep learning applications. Our design has been engineered for optimal airflow, power efficiency and stability, and it makes it easy to integrate high-performance GPUs and to provide exceptional computing power for the most demanding applications.
*10 GPU support is only possible with a upgradeable PCIe switch GPU backplane, or if 8 GPU are configured for PCIe 4.0 X16 link and 2 GPU configured for PCIe 4.0 X8 link)

High-Speed Memory & Storage

We also provide up to 8TB of DDR4 RAM and several Hot-Swap NVMe SSDs for instant data access and better I/O performance.

Enterprise-Grade Reliability

A strong chassis, hot-swappable cooling fans, optimized thick copper power delivery components and fully redundant power supplies for the most reliable and efficient enterprise grade GPU server solution.

4U-E10G-ROMEX2 Full Load Stress Test (9x RTX 4090)

www.octoserver.com

Octoserver 4U-E10G-ROMEX2

Datasheet

Processor / System Bus

2 x Socket SP3
AMD EPYC™ 7002 / 7003 series processors*

Memory

Total Slots:
32(16 channel per CPU, 12 DIMM per CPU)
Capacity:
Maximum up to 4TB per CPU socket
Memory Type:
DDR4 3200  RDIMM
Memory Size:
128GB, 64GB, 32GB RDIMM , RDIMM/LRDIMM-3DS: up to 256GB

Expansion Slots

Up to 10 slots

Rear:
9 x PCIe x16 for dual-slot GPU cards (Gen4 x16 link, FH/FL)
1 x PCIe x16 for NIC/ Add In Card  (Gen4 x8 link, FH/HL)

Front:
2x U.2 SSD , Hot Swap, PCIe x4 (Gen4 x4 link)

—–

OPTIONAL Front Module (PCIe 52 Lanes Switch Based):
4 x PCIe x8 (Gen4 x8 link, FH/HL)

Storage

1 x M.2 Socket, supports up to 22110 type (PCIe Gen4 x4 link)

2x U.2 SSD , Hot Swap, PCIe x4 (Gen4 x4 link)

—–

(OPTIONAL*) 8 x 2.5″ SATA Front Hot-swap Storage Bays

Networking

2 x 10GbE LAN ports (RJ45, X550-T2)
1 x Management Port, RJ45

Multi-GPU Support

Supports all common dual slot PCIe add in cards

Front I/O Ports

1 x VGA
4 x USB3
2 x RJ-45 LAN, 10G
1 x RJ-45 Management LAN Port

Rear I/O Ports

2 x RJ-45 LAN 10G
1 x RJ-45 Management LAN port

Switch/LED

Front :
1 x Power switch/LED
1 x Location ID switch/LED
1 x Message LED
1 x Reset Switch

Rear :
1 x Power switch/LED
1 x Reset switch/ LED
1 x Location ID switch/LED

OS Support

Supports most common OS

Management Solution

ASPEED AST2500: IPMI2.0 with iKVM and vMedia support

Regulatory Compliance

CE

Dimensions

940mm x 440mm x 175mm

Form Factor

4U

Weight

Net Weight: 42 Kg
Gross Weight: 45.00 Kg

Power Supply

3+1 redundant 1600W 80 PLUS Platinum power supply
80-240 Vac, 10A (x4) 50/60Hz
—
Upgradeable to 3+1 redundant 2000W 80 PLUS Titanium

Environment

Operating temperature: 10°C ~ 35°C
Non-operating temperature: -40°C ~ 60°C
Non-operating humidity: 20% ~ 95% (Non-condensing)

White-Label & OEM
Solutions

OCTOSERVER can provide an entire white label product, which enables companies, data centers, and service providers to private label and use high-quality computing infrastructure with their own brand. Whether it is AI, HPC or cloud hosting you are building we can provide a full white-label branding solution for our B2B clients.

 

As an OEM, OCTOSERVER specializes in designing and manufacturing custom server solutions tailored to the unique needs of our business clients. With fast turnaround times and extensive customization options, we provide flexible, high-performance systems that meet specific requirements for various industries. Our in-house design and manufacturing capabilities ensure superior quality and reliability, enabling us to deliver solutions that align perfectly with our clients’ business goals.

 

* Example 4U White-Label GPU server that was built for Flux.