Exxact TensorEX TS4-140031951

Product Overview Exxact TensorEX TS4-140031951

Model: Exxact TensorEX 5U HGX H100 Server (TS4-140031951)
Form Factor: 5U Rack-mountable Server
Use Case: High-performance computing (HPC), AI/ML training, deep learning, and large-scale inference workloads
Base Price: Starting from $253,882.20 USD

The Exxact TensorEX 5U HGX H100 Server, model TS4-140031951, is a high-performance computing platform designed to handle some of the most demanding workloads in modern computing.

With its 5U rack-mountable form factor, this server is built for data centers that require robust processing power, high memory bandwidth, and maximum GPU acceleration for AI and deep learning applications.

The system is optimized for machine learning training, large-scale inference, and scientific computing, making it an ideal choice for research institutions, enterprises focused on AI innovation, and technology labs.

It’s built around NVIDIA’s HGX platform and comes equipped with eight H100 SXM GPUs, providing unmatched parallel processing capabilities.

The server’s starting price of over $253,000 reflects its cutting-edge components and enterprise-class reliability. Exxact offers a web-based configurator, allowing customers to fully customize the server based on their processing, memory, storage, and networking needs.

This makes the TS4-140031951 not just a powerful system but also a flexible one that adapts to different computational environments. As AI workloads continue to evolve, this server provides a scalable foundation for businesses that want to stay ahead in performance and future-proof their infrastructure.

Hardware Specifications

CPU Options

• Supported Processors: Up to 2x 4th/5th Gen Intel Xeon Scalable processors (Socket LGA-4677)
• Chipset: Intel C741
• Available CPU Configurations:
  • Intel Xeon Gold Series:
    • 6444Y: 16 cores, 3.6 GHz, 45 MB L3 Cache
    • 6442Y: 24 cores, 2.6 GHz, 60 MB L3 Cache
    • 6448Y: 32 cores, 2.1 GHz, 60 MB L3 Cache
  • Intel Xeon Platinum Series:
    • 8462Y+: 32 cores, 2.8 GHz, 60 MB L3 Cache
    • 8460Y+: 40 cores, 2.0 GHz, 105 MB L3 Cache
    • 8468: 48 cores, 2.1 GHz, 105 MB L3 Cache
    • 8470: 52 cores, 2.0 GHz, 105 MB L3 Cache
    • 8480+: 56 cores, 2.0 GHz, 105 MB L3 Cache
    • 8490H: 60 cores, 1.9 GHz, 112.5 MB L3 Cache

The Exxact TS4-140031951 supports dual Intel Xeon Scalable processors, specifically from the 4th and 5th generation families, which are among Intel’s most advanced CPUs for data center and AI workloads.

These processors are designed for Socket LGA-4677 and utilize the Intel C741 chipset, enabling support for large memory footprints and high-speed I/O. The server allows a wide range of CPU options to suit varying workloads.

For example, the Intel Xeon Gold series includes processors like the 6444Y, which offers 16 cores at 3.6 GHz, suitable for applications that need higher clock speed rather than massive core counts.

At the higher end, the Platinum series features CPUs like the 8490H with 60 cores and a large 112.5 MB L3 cache, making it ideal for massively parallel operations.

The dual-CPU configuration supports simultaneous multithreading and advanced power management, which ensures efficient workload handling and energy savings.

These CPU options give users the flexibility to prioritize core count, clock speed, or cache size depending on the application whether it’s AI model training, large-scale simulation, or virtualization.

Combined with support for DDR5 memory and multiple GPUs, the CPU options on this server make it a versatile powerhouse for enterprise-grade performance.

GPU Configuration

• GPU Baseboard: NVIDIA HGX H100-8
• GPU Slots: 8x NVIDIA H100 SXM5 GPUs
• Memory per GPU: 80 GB HBM3
• NVLink Support: Available for pairs of identical GPUs

The centerpiece of the Exxact TS4-140031951 is its GPU subsystem, powered by the NVIDIA HGX H100 platform. This configuration includes eight NVIDIA H100 SXM5 GPUs, each offering 80 GB of HBM3 memory.

These GPUs are connected via NVIDIA NVLink, which facilitates high-speed, low-latency interconnectivity among GPUs ideal for large AI models that must span multiple GPUs in parallel.

The H100 GPU is NVIDIA’s flagship for AI training and inference, featuring powerful tensor cores, support for Transformer Engine, and performance that surpasses previous generations by a wide margin.

The SXM5 form factor allows for greater thermal efficiency and higher performance per watt compared to PCIe alternatives. With 8 GPUs connected through NVLink, the system achieves extremely fast internal communication critical for tasks like training deep learning models with billions of parameters.

Moreover, this configuration ensures scalability for high-density GPU computing, supporting frameworks like TensorFlow, PyTorch, and distributed computing platforms such as NVIDIA’s NCCL.

Whether running natural language models, large-scale recommendation systems, or high-resolution simulations, the GPU configuration in this server is designed to deliver elite performance across a broad range of use cases.

Memory

• Memory Type: DDR5 ECC Registered
• Speed: 4800 MT/s
• Total Slots: 32 DIMM slots
• Available Configurations:
  • 32 GB (1×32 GB)
  • 64 GB (1×64 GB)
  • 128 GB (1×128 GB)
  • 256 GB (1×256 GB)
• Maximum Supported Memory: Up to 1,024 GB

Memory is another critical component in high-performance computing, and the TS4-140031951 excels here with support for DDR5 Registered ECC memory.

The server features 32 DIMM slots, allowing configurations of up to 1 TB of RAM. DDR5 offers a significant speed improvement over DDR4, operating at 4800 MT/s, which enhances data transfer rates and overall system responsiveness.

Registered ECC (Error-Correcting Code) ensures that memory errors are detected and corrected, improving system stability and uptime especially important in enterprise and scientific computing environments.

Users can configure the memory in increments of 32 GB, 64 GB, 128 GB, or 256 GB modules. This modularity provides flexibility, whether the server is intended for memory-bound simulations, large dataset loading for AI training, or in-memory databases.

Applications like genomics, real-time analytics, and 3D rendering will benefit significantly from the high-capacity, high-bandwidth RAM. In AI-specific workloads, sufficient memory ensures that the data pipeline feeding the GPUs does not become a bottleneck.

This combination of high speed, scalability, and ECC reliability makes the memory architecture of the TS4-140031951 exceptionally well-suited for mission-critical deployments.

Storage Options

• NVMe U.2 SSDs (PCIe 4.0 x4):
  • 960 GB
  • 1.92 TB
  • 3.84 TB
  • 7.68 TB
  • 15.36 TB
  • 30.72 TB
• SATA SSDs:
  • 480 GB
  • 960 GB
  • 1.92 TB
  • 3.84 TB
  • 7.68 TB

The Exxact TS4-140031951 offers versatile storage options to accommodate both high-speed data processing and large capacity storage needs. Users can choose between NVMe U.2 SSDs and traditional SATA SSDs.

NVMe drives support PCIe Gen4 x4, delivering ultra-fast read/write speeds essential for training AI models, loading large datasets, or running high-performance applications.

Available capacities range from 960 GB to an impressive 30.72 TB per drive, giving users the ability to scale storage based on their needs.

SATA SSDs, while not as fast as NVMe, provide cost-effective and reliable options for less I/O-intensive tasks or long-term storage. These drives are offered in sizes from 480 GB to 7.68 TB, suitable for operating system installation, software deployment, and backup storage.

The system’s flexible storage architecture allows for mixed configurations combining high-speed NVMe drives for active data and high-capacity SATA drives for archival.

This tiered approach to storage makes the server adaptable to both compute-heavy workloads and large-scale data retention.

For use cases such as AI model checkpointing, video processing, or genomic sequencing, the ability to read/write at high speeds while maintaining ample capacity is essential and this server handles both with ease.

Networking

• Ethernet:
  • 2x 10GBase-T Ethernet ports
  • 1x 1GbE Dedicated Management Port (IPMI)
• InfiniBand (Optional):
  • NVIDIA Mellanox ConnectX-6 VPI HDR InfiniBand
    • Single Port: 200 Gbps, QSFP56
    • Dual Port: 200 Gbps, QSFP56

Efficient data transfer is essential in AI and HPC environments, and the TS4-140031951 meets this need with multiple networking options.

By default, it includes dual 10GBase-T Ethernet ports for high-speed network connectivity, allowing rapid data ingress and egress to match the pace of the internal compute and storage operations.

Additionally, there’s a dedicated 1GbE port for IPMI (Intelligent Platform Management Interface), which provides out-of-band management for monitoring and remote control vital for enterprise IT administrators who need to manage servers remotely.

For users requiring ultra-high bandwidth and low latency, especially in clustered environments or supercomputing applications, optional InfiniBand cards are available. These include NVIDIA Mellanox ConnectX-6 VPI adapters that deliver 200 Gbps through QSFP56 ports.

InfiniBand is particularly useful for workloads distributed across multiple servers or GPU nodes, as it enables fast inter-node communication, reducing training time for massive AI models or facilitating real-time processing across systems.

With its robust standard networking and optional high-performance interconnects, the server is well-equipped to integrate into any data center architecture whether standalone, part of a local HPC cluster, or in a cloud-hybrid configuration.

Power Supply

• Power Units: 6x 2200W (4+2) Redundant Power Supplies
• Total Power Consumption: Approximately 6,377 Watts
• Thermal Output: Approximately 21,760 BTU/Hr

Powering a high-density system like the TS4-140031951 requires robust, reliable energy delivery. This server includes six 2200W power supplies configured in a 4+2 redundant setup.

This means four units actively power the system while two are in standby as hot-spares ensuring continuous operation even if one or two units fail.

Redundant power supplies are a staple of enterprise-grade servers, providing essential uptime for critical workloads.

The system’s total power draw is approximately 6,377 Watts under full load, reflecting the intensive demands of 8 GPUs, dual CPUs, and high-performance memory and storage.

In terms of heat generation, this setup can produce up to 21,760 BTU/hr, which necessitates a well-ventilated rack and sufficient cooling infrastructure.

The server chassis and internal layout are engineered for optimal airflow, supporting both front-to-back and rack-level cooling designs.

For organizations planning to deploy this server, understanding the electrical and thermal requirements is crucial.

Data centers must ensure not only adequate power availability but also backup UPS systems and thermal management solutions to maintain server health and performance over the long term.

Physical Dimensions

• Depth: 37.2 inches

Physically, the Exxact TS4-140031951 is a 5U server chassis, meaning it occupies five units of vertical space in a standard 19-inch server rack.

Its depth is approximately 37.2 inches, which is common for high-density servers that house multiple GPUs and large power supplies.

These dimensions indicate that the server is built for professional data centers, not shallow racks or constrained environments.

Due to its size and weight especially when fully populated with 8 GPUs, 2 CPUs, multiple storage drives, and 6 PSUs proper rack infrastructure is necessary.

Rails and support trays should be used to manage installation and servicing. The deeper chassis allows better airflow and cable routing, ensuring components remain within thermal thresholds even under full workload.

When planning deployment, administrators must ensure enough front and rear clearance for cable management and airflow.

Because of its form factor and component density, the server is best suited for high-performance data centers, research labs, or enterprise environments with advanced cooling systems.

The 5U size offers a balance between density and serviceability, allowing for powerful components without excessive maintenance complexity.

Software & Operating System

• Operating Systems:

  • Rocky Linux 9
  • Ubuntu 22.04

• Deep Learning Software Stacks (Optional):

  • Exxact EMLI – Conda Deep Learning Stack
  • Exxact EMLI – Container Deep Learning Stack
  • Exxact EMLI – DIY Deep Learning Stack
• Custom Software Support: Available upon request (e.g., AMBER, Relion, Schrodinger, NVIDIA AI Enterprise)

Warranty & Support

• Standard Warranty: 3-Year Warranty on Parts and Labor
• Support: Perpetual Email and Telephone Support