NVIDIA DGX™ Systems - NVIDIA -Macnica

A milestone in AI infrastructure

NVIDIA DGX™ B300

The NVIDIA DGX™ B300 helps you innovate and optimize your business. The latest addition to NVIDIA's legendary DGX system and the foundation of the NVIDIA DGX SuperPOD™, the DGX B300 is powered by the groundbreaking NVIDIA B300 Tensor Core GPU, accelerating the use of AI. Designed to maximize AI throughput, it provides enterprises with a highly sophisticated, systemized, and scalable platform to help them achieve breakthroughs in natural language processing, recommendation systems, data analytics, and more. Available on-premise or with a variety of access and deployment options, the NVIDIA DGX B300 delivers the performance enterprises need to solve their biggest challenges with AI.

Detailed information on the NVIDIA DGX B300 is available.

Reference material explaining the NVIDIA DGX B300 from a network perspective,

We have prepared a book entitled "A Detailed Analysis of the NVIDIA DGX™ B300 from a Network Perspective: The Role of NVIDIA ConnectX®-8 in Maximizing the Use of GPUs." Please make use of it.

You can view the full text without entering any personal information.

Click here for the document

NVIDIA DGX™ B200

The NVIDIA DGX™ B200 is the latest generation of AI platform designed to accelerate enterprise AI innovation and strengthen competitiveness. Based on NVIDIA's innovative Blackwell architecture, the DGX B200 is a core component of the NVIDIA DGX SuperPOD™ and is optimized for large-scale generative AI and deep learning workloads. It delivers unparalleled AI computing performance, enabling enterprises to achieve unprecedented results in advanced use cases such as natural language processing, image recognition, and recommendations. The NVIDIA DGX B200 is suitable for on-premises deployment, and its high-density and efficient design allows you to quickly scale out AI projects and solve the complex challenges your organization faces with the power of AI.

NVIDIA DGX™ H200

The NVIDIA DGX™ H200 helps you innovate and optimize your business. The latest addition to NVIDIA's legendary DGX system and the foundation of the NVIDIA DGX SuperPOD™, the DGX H200 is powered by the groundbreaking NVIDIA H200 Tensor Core GPU and accelerates the use of AI. Designed to maximize AI throughput, it provides enterprises with a highly sophisticated, systemized, and scalable platform to help them achieve breakthroughs in natural language processing, recommendation systems, data analytics, and more. Available on-premise or with a variety of access and deployment options, the NVIDIA DGX H200 delivers the performance enterprises need to solve their biggest challenges with AI.

Product spec

	NVIDIA DGX™ B300	NVIDIA DGX™ B200	NVIDIA DGX™ H200
GPUs	8x NVIDIA Blackwell Ultra GPUs	8x NVIDIA Blackwell GPUs	8x NVIDIA H200 Tensor Core GPUs
TFLOPS	144 PFLOPS FP4 inference 72 PFLOPS FP8 training	72 petaFLOPS FP8 training and 144 petaFLOPS FP4 inference	32 petaFLOPS FP8
GPU memory	2.3TB	1,440GB	1,128GB
storage	Storage OS: 2x 1.9TB NVMe M.2 Internal storage: 8x 3.84TB NVMe E1.S	OS: 2x 1.9TB NVMe M.2 Internal storage: 8x 3.84TB NVMe U.2	OS: 2x 1.92TB NVMe M.2
network	8x OSFP ports serving 8x NVIDIA ConnectX-8 VPI >Up to 800Gb/s NVIDIA InfiniBand/Ethernet 2x dual-port QSFP112 NVIDIA BlueField-3 DPU >Up to 400Gb/s NVIDIA InfiniBand/Ethernet	4x OSFP ports serving 8x single-port NVIDIA ConnectX-7 VPI >Up to 400Gb/s NVIDIA InfiniBand/Ethernet 2x dual-port QSFP112 NVIDIA BlueField-3 DPU >Up to 400Gb/s InfiniBand/Ethernet	4x OSFP ports serving 8x single-port NVIDIA ConnectX-7 VPI > Up to 400Gb/s InfiniBand/Ethernet 2x dual-port QSFP112 NVIDIA ConnectX-7 VPI > Up to 400Gb/s InfiniBand/Ethernet

Case study

Click here for DGX series introduction examples

Other GPU Servers

We also carry Supermicro GPU servers. Please contact us for more details.

item	NVIDIA HGX H100/H200 Server	NVIDIA H100 Server (Scalable ①)	NVIDIA H100 Server (Scalable ②)	Water-cooled Server
SYS	SYS-821GE-TNHR	SYS-741GE-TNRT	SYS-521GE-TNRT	SYS-421GE-TNHR2-LCC
CPU	Dual Socket E (LGA-4677)5th Gen Intel® Xeon® / 4th Gen Intel® Xeon® Scalable processors	Dual Socket E (LGA-4677)5th Gen Intel® Xeon® / 4th Gen Intel® Xeon® Scalable processors	Dual Socket E (LGA-4677)5th Gen Intel® Xeon® / 4th Gen Intel® Xeon® Scalable processors	Dual Socket E (LGA-4677)5th Gen Intel® Xeon® / 4th Gen Intel® Xeon® Scalable processorsSupports Intel Xeon CPU Max Series with high bandwidth memory (HBM)
Memory	Slot Count: 32 DIMM slotsMax Memory (1DPC): Up to 4TB 5600MT/s ECC DDR5 RDIMMMax Memory (2DPC): Up to 8TB 4400MT/s ECC DDR5 RDIMM	Slot Count: 16 DIMM slotsMax Memory (1DPC): Up to 2TB 5600MT/s ECC DDR5 RDIMMMax Memory (2DPC): Up to 4TB 4400MT/s ECC DDR5 RDIMM	Slot Count: 32 DIMM slotsMax Memory (1DPC): Up to 4TB 5600MT/s ECC DDR5 RDIMMMax Memory (2DPC): Up to 8TB 4400MT/s ECC DDR5 RDIMM	Slot Count: 32 DIMM slotsMax Memory (1DPC): 4TB 5600MT/s ECC DDR5 RDIMMMax Memory (2DPC): 8TB 4400MT/s ECC DDR5 RDIMM
form	8U Rackmount type	Tower Rackmount type	5U Rackmount type	4U Rackmount type
GPU Model	NVIDIA SXM: HGX H100 8-GPU (80GB), HGX H200 8-GPU (141GB)	NVIDIA PCIe: H100 NVL, H100,	NVIDIA PCIe: H100 NVL, H100	NVIDIA SXM: HGX H100 8-GPU (80GB), HGX H200 8-GPU (141GB)
Cooling Method	Air Cooling	Air Cooling	Air Cooling	Water Cooling
Image

Detailed information on building a GPU cluster is available.

Basic reference materials for building GPU clusters,
"An Introduction to GPU Cluster Construction: Understanding Multi-Node GPU System Architecture for Generative AI/LLM"

We have them available, so please make use of them.
You can also view the full text here without entering any personal information.

Click here for the document

Quote / Inquiry

Getting started with NVFP4 inference on NVIDIA DGX™ B200

As large-scale language models (LLMs) become larger and applications using them become more complex, reducing the amount of calculations and shortening inference times is an important issue.

The latest NVIDIA Blackwell generation GPUs, including the NVIDIA® DGX™ B200, now support 4-bit floating-point (FP4) data, which is expected to reduce the amount of calculation and shorten inference times.

The following article, spanning five parts, explains how to quantize LLM into a format called NVFP4, which is newly supported by Blackwell-generation GPUs, and then deploy it to NVIDIA® NIM™, NVIDIA's inference Microservices, for inference, as well as how to perform inference using the NVIDIA TensorRT™-LLM Python API.

Chapter 4: Benchmarking NVFP4 and FP8 Chapter 5: Deploying Llama-3.1-405B-Instruct
We will provide this information to those who fill out the form at the end of Episode 3. You will be redirected to a simple form entry screen, and after completing the form, we will send you a URL by email.

AI TRY NOW PROGRAM

This is a support program that allows you to verify the latest AI solutions on an NVIDIA environment before introducing them into your company.

You can verify the benefits of implementing software products provided by NVIDIA, including NVIDIA AI Enterprise and NVIDIA Omniverse, as well as the AI learning environments and tools that we can provide, and the feasibility of applications before purchasing.

Click here for details