Deskside AI Supercomputer - NVIDIA GB300 Grace™ Blackwell Ultra Desktop Superchip - W775-V10-L01

1/5

W775-V10-L01

Deskside AI Supercomputer - NVIDIA GB300 Grace^™ Blackwell Ultra Desktop Superchip

人工智慧平台高效能運算伺服器人工智慧訓練伺服器人工智慧推論伺服器

Closed-loop liquid cooling solution
1 x NVIDIA Blackwell Ultra GPU
1 x NVIDIA Grace-72 Core Neoverse V2 CPU
252 GB HBM3E GPU memory with 7.1 TB/s total bandwidth
496 GB LPDDR5X CPU memory with up to 396 GB/s bandwidth
2 x 400 Gb/s QSFP ports via NVIDIA ConnectX^®-8 SuperNIC^™
1 x 10 Gb/s LAN port via Marvell AQC113
2 x M.2 PCIe Gen6 x4
2 x M.2 PCIe Gen5 x4
1 x PCIe x16 slot with Gen5 x16 lanes for GPUs
2 x PCIe x16 slots with Gen5 x8 lanes
Single 1600 W ATX 80 PLUS Platinum power supply

Ordering Numbers: 6NW775V10MR000L01*

加入比較表

儲存

詢價清單

NVIDIA GB300 Grace™ Blackwell Ultra Desktop Superchip

To meet the demands of large-scale AI models, this platform features the NVIDIA GB300 Grace™ Blackwell Ultra Desktop Superchip and up to a massive 748 GB of coherent memory. This breakthrough architecture delivers an unprecedented amount of compute performance, enabling teams to develop and run large-scale AI training and inferencing workloads directly from the deskside.

Optimized for the NVIDIA AI software stack, cloud-native and AI-native developers, researchers, and data scientists can instantly prototype, fine-tune, and inference large AI models from the desktop and seamlessly deploy to the data center or cloud.

Up to 748 GB

Coherent Memory

Up to20 PFLOPS FP4

AI performance

Up to 9,000 tok/s

Peak aggregate throughput

W775-V10-L01 Block Diagram

W775-V10-L01 Product Overview

Technological Breakthroughs

NVIDIA Grace Blackwell Ultra Desktop Superchip

Powered by an NVIDIA Blackwell Ultra GPU and Grace CPU connected via NVLink-C2C, this platform delivers top-tier system communication and performance with next-gen CUDA and Tensor Cores.

Fifth-Generation Tensor Cores

Powered by latest NVIDIA Blackwell Tensor Cores, the system enables FP4 AI to boost model capacity and performance while maintaining high accuracy.

NVIDIA ConnectX-8 SuperNIC

The NVIDIA ConnectX^®-8 SuperNIC™ delivers up to 800 Gb/s network speed, supercharging AI factory workloads with ultra-fast, efficient connectivity.

NVIDIA DGX OS

This solution provides a stable, fully qualified OS environment with built-in configurations, drivers, and diagnostic tools, enabling seamless scaling across local, cloud, and accelerated data center infrastructure

NVIDIA NVLink-C2C Interconnect

NVIDIA NVLink-C2C extends NVLink technology to a chip-to-chip interconnect, enabling high-bandwidth, coherent data transfers between the CPU and GPU.

Large Coherent Memory for AI

As AI models grow in scale and complexity, the NVIDIA Grace Blackwell Ultra uses its C2C interconnect to create a large coherent memory pool, bypassing traditional bottlenecks to train and run massive models efficiently.

NVIDIA AI Software Stack

A full-stack AI solution for fine-tuning, inference, and data science, NVIDIA’s AI software stack lets you develop locally and seamlessly deploy to the cloud or data center using unified tools, frameworks, and models.

Workload Optimized Power-Shifting

The system leverages AI-driven optimization to dynamically shift power based on active workloads, continuously maximizing performance and energy efficiency.

產品規格

Dimensions (WxHxD, mm)

Pedestal
245 x 500.4 x 531

Superchip

NVIDIA GB300 Grace^™ Blackwell Ultra Desktop Superchip:
- 1 x NVIDIA Grace^™ CPU
- 1 x NVIDIA Blackwell Ultra GPU
- Connected with NVIDIA NVLink^™-C2C

Memory

NVIDIA Grace^™ CPU:
- 496 GB of LPDDR5X memory with ECC
- 396 GB/s memory bandwidth

NVIDIA Blackwell Ultra GPU:
- 252 GB HBM3E
- 7.1 TB/s memory bandwidth

LAN

2 x 400 Gb/s QSFP (1 x NVIDIA ConnectX^®-8 SuperNIC^™)
1 x 10 Gb/s LAN (1 x Marvell AQC113)
1 x 10/100/1000 Mb/s Management LAN

Video

Integrated in onboard BMC
- 1 x Mini-DP (1920 x 1080 @ 60 Hz)

Optional GPUs:
- NVIDIA RTX PRO 6000 Blackwell Workstation Edition
- NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition
- NVIDIA RTX PRO 4000 Blackwell SFF Edition
- NVIDIA RTX PRO 2000 Blackwell

Audio

Realtek^® ALC4080 Audio Codec
Front: 2 x Audio jacks (Audio out/Mic)
Rear: 3 x Audio jacks (Audio in/Audio out/Mic)

Storage

Internal M.2:
2 x M.2 (2280), PCIe Gen5 x4, from CPU, with software RAID 1 support
2 x M.2 (2280), PCIe Gen6 x4, from ConnectX-8

PCIe Expansion Slots

Slot_1: PCIe x16 (Gen5 x16), for GPUs
Slot_2: PCIe x16 (Gen5 x8)
Slot_3: PCIe x16 (Gen5 x8)

1 x M.2 (2230), E-key (Wi-Fi), PCIe Gen2 x1, from ConnectX-8

Front I/O

1 x USB 3.2 Gen2 port (Type-C) (5 V Output)
2 x USB 3.2 Gen1 ports (Type-A)
2 x Audio jacks (Audio out/Mic)
1 x Power button with LED
1 x Reset button with storage activity LED

Rear I/O

4 x USB 3.2 Gen2 ports (Type-A)
1 x Micro USB port (USB-to-UART)
1 x Mini-DP
2 x QSFP ports
1 x RJ45 port
1 x MLAN port
2 x Wi-Fi antenna ports
3 x Audio jacks (Audio in/Audio out/Mic)

Power Supply^[#1]

Single 1600 W ATX 80 PLUS Platinum power supply^[1]

AC Input:
- 100-240V~, 15-8A, 47-63Hz
- 115-240V~, 15-8A, 47-63Hz

DC Output:
- Max 1300W/ 100-240V~ & Max 1600W/ 115-240V~
+12V1-12V6/ 50A
+5V/ 25A
+3.3V/ 25A
+5Vsb/ 3.5A
-12V/ 0.3A

^[1]The system power supply requires C19 power cord.

^[Note]115-240 V AC input is recommended for optimized user experience.

^[Note]System requires a 20-amp circuit to plug-in.

OS Compatibility

Ubuntu with NVIDIA AI Developer Tools

System Fans

3 x 120x120x25mm

Operating Properties

Operating temperature: 10°C to 35°C
Operating humidity: 8% to 80% (non-condensing)
Non-operating temperature: -40°C to 60°C
Non-operating humidity: 20% to 95% (non-condensing)

^[Note]To ensure system stability and prevent condensation, when the relative humidity exceeds 50%, the coolant inlet temperature must be higher than the dry-bulb temperature and it should not exceed 45°C.

Weight

Net Weight: 29.2 kg
Gross Weight: 34.4 kg

Packaging Dimensions

732 x 400 x 775 mm

Packaging Content

1 x W775-V10-L01
1 x Superchip liquid cooling kit

Part Numbers

- Barebone package: 6NW775V10MR000L01*

[#1] 伺服器產品未附帶電源線。客戶需自行選擇適當的電源線 (可選配件)，或聯繫業務協助客製化。

[#2] 本網站所提供之產品規格、圖片及其他資訊僅供參考，如有差異應以實際產品為準。技嘉科技保留隨時修改內容的權利，並不對因使用上述資料而導致的任何損失承擔責任。

[#3] 本產品所標示之效能數據為晶片廠商或相關介面標準制定組織所公佈的最大理論值，實際效能可能因系統配置及使用環境而有所不同。

[#4] 本文所使用的所有商標及企業識別標誌均為其合法所有者的財產。