Chaitex
Chaitex
Back to Servers
YH002 Mezzanine Server 768GB 1

YH002 Mezzanine Server 768GB

The YH002 Mezzanine Server is a cutting-edge AI computing platform built on the world's first RISC-V cloud AI architecture.

It is specifically engineered to handle massive Large Language Models, with deep optimization for the DeepSeek series.

The server features eight high-performance YH002 Mezzanine modules (OAM form factor), each equipped with 96GB of memory, totaling 768GB of high-speed VRAM.

Its TPU-like architecture utilizes specialized matrix engines to deliver exceptional efficiency in FP8 and INT4 precision tasks.

With a high-speed interconnect system comparable to NVLink, the YH002 achieves near-linear scalability and provides up to a 52% performance boost in real-world AI workloads compared to traditional GPU architectures.

CPUs
2x Intel 8558(48 ядер,2.1ГГц)
RAM
1,5ТБ (24x 64ГБ)
Accelerators
8x YH002 Module 96GB
Storage
2x 960G SATA SSD + 4x 3.84T NVME SSD
Pre-order
Product status

CPU & Memory

CPU
2x Intel 8558(48 ядер,2.1ГГц)
RAM
1,5ТБ (24x 64ГБ)
RAM Type
-

Storage Configuration

SSD Model
-
Total Storage
2x 960G SATA SSD + 4x 3.84T NVME SSD
Disk Interface
-

GPU Configuration

GPU Vendor
YH
GPU Model
YH002 Module 96GB
GPU Count
8
Total VRAM
768ГБ HBM3e
Interconnect
YHLink 1200ГБ/c

Network & Virtualization

Network Interfaces
DH E310 (2*25Гбит/с) + 8x DH E316 (2*200Гбит/с) или MLX CX7(2*200Гбит/с)
Virtualization (MIG)
-
CUDA Compatibility
Не зависит от совместимости с CUDA, нативно поддерживает основные фреймворки

Power & Thermal

Power Supply
-
TDP
8 600 - 11 200
Cooling Type
Пассивное

Server Specifications

Technical ID
YH002 Mezzanine Server
Platform
YH
Form Factor
8U
Warranty
3
Primary Tasks
Обучение LLM,распределённое обучение, высоконагруженный inference, построение масштабируемых AI-кластеров
Price
On request