أجهزة احترافيةمزيد من الأداء

جديد · Teklif ile tedarik

NVIDIA DGX Station GB300 (partner models)

العلامة التجارية: NVIDIA رمز المنتج: AI-NVIDIA-DGX-STATION-GB300

NVIDIA DGX Station is a platform that brings the data-centre-class GB300 Grace Blackwell Ultra superchip to a desktop workstation. It is designed for research teams that want to fine-tune large language models locally, run inference and carry out data science work without needing a server room. The system is offered as partners' own models; the suitable model and configuration are chosen together at the quotation stage.

  • Garanti ve servis koşulları teklif aşamasında üretici ve tedarik kanalına göre netleşir.
  • يتم تحديد شروط التوصيل والتركيب في مرحلة عرض السعر
  • لا يتم تخزين بيانات بطاقتك على موقعنا

NVIDIA DGX Station is a platform that brings the data-centre-class GB300 Grace Blackwell Ultra superchip to a desktop workstation. It is designed for research teams that want to fine-tune large language models locally, run inference and carry out data science work without needing a server room. The system is offered as partners' own models; the suitable model and configuration are chosen together at the quotation stage.

NVIDIA DGX Station GB300 (partner models) overview

DGX Station is NVIDIA's top-tier AI platform in a desktop form factor. It uses the GB300 superchip, which combines a Grace CPU and a Blackwell Ultra GPU in a single chassis, and presents CPU and GPU memory as a single coherent pool.

Hardware architecture

The 72-core NVIDIA Grace CPU is connected to the Blackwell Ultra GPU via NVLink-C2C. HBM3e memory is on the GPU side and LPDDR5X memory on the CPU side. For networking there is an NVIDIA ConnectX-8 SuperNIC with two QSFP112 ports (up to 800 Gb/s); two DGX Stations can work together over this connection.

Use cases in AI workloads

Local large language model fine-tuning, RAG and agent development, in-house testing of models that work with confidential data, computer vision and scientific computing prototypes.

Business advantages

Using the same NVIDIA software stack from model development to production makes it easier to move work prepared on the desktop to DGX servers and the cloud.

Scalability and infrastructure requirements

Runs in a desktop chassis, no rack required; due to high power consumption, a dedicated electrical circuit and a well-ventilated environment are recommended. Power and dimension figures vary by partner model.

Who is the NVIDIA DGX Station GB300 (partner models) for?

Suitable for R&D units that do not want to send their data to the cloud, university labs and AI teams that want to mature their models on the desktop before moving to the data centre.

Key features

  • GB300 Grace Blackwell Ultra Desktop Superchip
  • 72-core NVIDIA Grace CPU
  • HBM3e GPU memory + LPDDR5X CPU memory (coherent memory)
  • NVIDIA ConnectX-8 SuperNIC, 800 Gb/sn'ye kadar
  • Two systems can work together
  • NVIDIA DGX OS and NVIDIA AI software stack

Technical Specifications

MarkaNVIDIA
ModelDGX Station (GB300 Grace Blackwell Ultra Desktop Superchip)
Product typeDesktop AI workstation
PlatformNVIDIA GB300 Grace Blackwell Ultra Desktop Superchip
CPUNVIDIA Grace, 72 cores (Arm Neoverse V2)
GPU1 × NVIDIA Blackwell Ultra
GPU memoryHBM3e — capacity conflicting across sources (288 GB / 252 GB); finalised by model at the quotation stage
CPU memory496 GB'a kadar LPDDR5X
CPU–GPU interconnectNVLink-C2C
NetworkNVIDIA ConnectX-8 SuperNIC, 2 × QSFP112, 800 Gb/sn'ye kadar
Sales channelNVIDIA partners (ASUS, Dell, HP, MSI, Supermicro, etc.)

Procurement and corporate quotation process

For NVIDIA DGX Station GB300 (partner models), the partner model, memory and storage are selected together according to your needs. Price, delivery schedule and warranty terms are provided in the corporate quotation for the selected model. The technical specifications and availability of this product are confirmed with the manufacturer before quoting. Stating your intended use and delivery country in the quote form speeds up the process.

منتجات مشابهة