H200: Redefining Accelerated Computing for AI and HPC

The H200 is a state-of-the-art accelerated computing platform designed to enhance performance for both artificial intelligence (AI) and high-performance computing (HPC) workloads. Building upon the foundation of its predecessor, the H100, the H200 introduces significant advancements in memory capacity, bandwidth, and computational efficiency.

The H200 represents a significant leap in GPU technology, offering enhanced memory, superior computational performance, and improved energy efficiency. It is well-suited for a wide range of applications, from AI and machine learning to high-performance scientific computing, meeting the growing demands of modern computational workloads.

Key Features

  • Enhanced Memory Capacity and Bandwidth: The H200 is equipped with 141 gigabytes (GB) of HBM3e memory, delivering a memory bandwidth of 4.8 terabytes per second (TB/s). This represents nearly double the capacity and a 1.4x increase in memory bandwidth compared to the previous generation, facilitating the acceleration of large language models (LLMs) and complex scientific computations.
  • Advanced Computational Performance: With multi-precision Tensor Cores, the H200 achieves up to 3,958 teraflops (TFLOPS) of FP8 performance, making it highly suitable for deep learning and AI applications. It also offers 67 TFLOPS in FP64 precision, catering to the demands of HPC tasks.
  • Scalability and Flexibility: The platform supports configurations with up to eight GPUs interconnected via NVIDIA NVLink and NVSwitch, providing over 32 petaflops of FP8 deep learning compute and more than 1.1 terabytes of aggregate high-bandwidth memory. This scalability is essential for handling the most demanding AI and HPC applications.
  • Energy Efficiency: Designed with power efficiency in mind, the H200 offers a 50% improvement over its predecessor, delivering enhanced performance for AI workloads and HPC while maintaining similar power consumption levels.

Applications

  • Generative AI and Large Language Models: The H200’s substantial memory and computational capabilities make it ideal for training and deploying large-scale AI models, such as GPT-3 and Llama2, enhancing inference speeds and overall performance.
  • Scientific Research and Simulations: Its high double-precision performance and large memory capacity support complex simulations and computations in fields like climate modeling, molecular dynamics, and other HPC applications.
  • Data Analytics and Machine Learning: The H200 accelerates data processing tasks, enabling faster insights and more efficient handling of large datasets, which is crucial for modern data-driven applications.

Specifications

Processor / System Bus

2 x Socket

4th Intel® Xeon® Processor Scalable Family (Up to 350W)
5th Intel® Xeon® Processor Scalable Family (Up to 350W)

Memory

Total Slots:
32 (8 channel per CPU, 16 DIMM per CPU)
Capacity:
Maximum up to 4TB per CPU socket
Memory Type:
4th: DDR5 4400 RDIMM/ 3DS RDIMM (2DPC)
5th: DDR5 4400 RDIMM/ 3DS RDIMM (2DPC)
Memory Size:
64GB, 32GB, 16GB RDIMM
256GB, 128GB RDIMM 3DS

* Refer to support page for more information

Storage

Optional Kits:
-Broadcom MegaRAID 9560-16i
-Broadcom MegaRAID 9540-8i

Drive Bays

10 x 2.5” hot-swap drive bays (8 NVMe, 2 NVMe/SATA/SAS*)

[PCIe Switch directly]
Front: 8 NVMe
[CPU directly]
Rear: 2 NVMe(CPU2)/SATA(PCH)

H100 SKU
2 x M.2 Gen5 x4 link (CPU1) / 2 x M.2 Gen3 x2 link (PCH)

H200 SKU
2 x M.2 Gen3 x2 link (PCH)

Networking

2 x 10 Gigabit LAN ports (Intel X710-AT2 Controller)
1 x Management Por

Dimensions

885mm x 447mm x 306.65mm

Form Factor

7U

Weight

Net Weight: 113 Kg.
Gross Weight: 174 Kg.

Power Supply

4+2 Redundant 3000W 80 PLUS Titanium Power Supply