Equipped with the NVIDIA GB10 Grace Blackwell Superchip, DGX Spark brings compact AI performance for prototyping, fine-tuning, inference and building autonomous agents at the desktop.
NVIDIA DGX Spark delivers strong AI performance in a compact, power-efficient design, with the NVIDIA AI software stack and 128 GB of memory built in so developers can prototype, fine-tune and deploy new reasoning AI models.
NVIDIA NemoClaw in the NVIDIA Agent Toolkit helps build, evaluate and optimize safer autonomous agents directly from the desktop environment.
The 1 petaFLOP FP4 figure is theoretical, based on the use of sparsity.
The latest DGX OS update installs NemoClaw faster, adds current open models and speeds up inference, shortening agentic AI development cycles on DGX Spark.
NVIDIA Agent Toolkit with NVIDIA OpenShell provides open-source models and software for developers to build safer autonomous agents.
Grace Blackwell architecture delivering up to 1 petaFLOP of theoretical FP4 AI performance with sparsity.
Unified system memory lets you develop, test and run AI models of up to around 200 billion parameters.
High-performance interconnect lets two DGX Spark systems be paired for larger models.
DGX OS, frameworks, libraries, NVIDIA NIM, and the AI software ecosystem come preconfigured for developers.
Plans are available on several billing terms, from prototype to production and enterprise AI.
NVIDIA DGX Spark
NVIDIA DGX Spark
NVIDIA DGX Spark Cluster
NVIDIA DGX Spark Cluster
DGX Spark lets developers and businesses build AI agents, fine-tune models and test inference faster on a complete NVIDIA AI platform.
Brings Grace Blackwell performance into a compact, power-efficient form factor that fits a workspace.
Ready for research into reasoning AI, autonomous agents, OpenShell and NemoClaw.
Shorten the development cycle from idea and prototype through fine-tuning to AI workload deployment.
Multiple DGX Spark units can be connected for larger workloads and multi-model scenarios.
Take advantage of NVIDIA NIM, frameworks, libraries, drivers and an optimized DGX OS environment.
Advice on plan selection, AI environment, connectivity, runtime and an operating model that fits your business.
DGX Spark suits AI developers, researchers, and data scientists who need Grace Blackwell performance in a desktop-sized form factor.
Build, test, and validate AI models or agentic AI applications directly in a desktop environment.
Fine-tune models and experiment with RAG workflows, reasoning AI and model optimization pipelines.
Run, evaluate and optimize inference for large models before moving to cloud or production.
Accelerates data analysis, notebooks, feature engineering, and algorithm experimentation at your desk.
Suited to experimenting with Isaac, Metropolis, Holoscan and edge AI applications.
A set of cloud-native tools, libraries and frameworks that speeds up production AI development, with security, optimized performance and enterprise support.
DGX Spark ships with the NVIDIA AI software stack and supports the NVIDIA software ecosystem, so AI projects can start quickly.
Powerful capacity, optimized to accelerate AI/ML and high-performance workloads at any scale.
Training, inference, or fine-tuning - HiTechCloud offers a range of GPUs to match your needs, with transparent pricing and on-demand deployment environments.
HiTechCloud GPU instances combine NVLink/PCIe, InfiniBand (RDMA), and RAIL topology to optimize AI/HPC performance.
A network and GPU fabric designed so AI workloads scale reliably, easing bandwidth bottlenecks and holding performance at scale.
High-speed GPU-to-GPU connectivity within and between nodes, reducing bottlenecks during model training.
Low-latency connectivity optimized for distributed training and reduced processing load on the host.
A parallel network architecture delivering higher bandwidth, redundancy, and consistent performance at any scale.
From a single GPU to large clusters, HiTechCloud provisions resources ahead of demand and gets the most out of every instance.
Scale GPU resources automatically from a handful to thousands of GPUs, with forecasting and provisioning ahead of demand.
Access over SSH, TCP, and HTTP with built-in protection layers that keep enterprise data safe and access fully controlled.
Partition a single GPU into multiple independent instances to run AI workloads in parallel, improving utilization and lowering infrastructure cost.
From large-scale training to real-time inference - pay only when you actually run, on a GPU cloud platform built for every AI and high-performance workload.
From sign-up to a running GPU instance in under 5 minutes - no complex setup, no resource reservations, no charges while idle. Just deploy, run, and pay only for what you use.
A complete management platform for deploying and operating AI agents securely at scale on enterprise-grade infrastructure.
Explore AgentBaseA unified platform for training, fine-tuning, and deploying AI models at any scale.
Explore AI PlatformSupports fast search, real-time analytics, log and large-scale event data, and vector databases for RAG.
Explore Vector DatabaseManaged Kubernetes for container orchestration, AI services, and GPU cloud workloads.
Explore KubernetesDeploy systems and applications closer to your users, reducing latency and meeting local regulatory requirements.
BKK-01
HCM-01 · HCM-02 · HCM-03
HAN-01 · HAN-02
Large enterprises and fast-growing startups choose HiTechCloud for secure, high-performance AI cloud solutions that let them innovate and scale.
The HiTechCloud team advises on GPU architecture, networking, security, and an operating model matched to your real workloads.
Quick facts before choosing the NVIDIA DGX Spark at HiTechCloud.
DGX Spark suits developers, researchers, data scientists and businesses that want to work on agentic AI, fine-tuning, inference, RAG and data science in a desktop environment.
With 128 GB of unified memory per system, DGX Spark is suited to experimenting with and running large AI models. Connecting two systems lets the workload scale to larger models.
NemoClaw is an open-source reference stack in the NVIDIA Agent Toolkit that supports building, evaluating and optimizing safer AI agents with security and privacy guardrails.
The plans on this page are available on 3-, 6-, 12-, 24-, 36- and 48-month terms.
AI Starter suits getting started and prototyping. AI Research suits teams that need help setting up the AI environment, higher bandwidth and more frequent use.
Choose AI Production or AI Enterprise when you need multiple DGX Spark systems, larger resources, multi-model workloads, HA and 24/7 priority support.
Yes. HiTechCloud advises on DGX OS, the NVIDIA AI software stack, containers, AI frameworks, NVIDIA NIM, networking and workload-specific operating models.
HiTechCloud helps you select the DGX Spark plan, interconnect architecture, NVIDIA AI software stack and operating roadmap that fit your business.
Start deploying quickly with HiTechCloud
Announcements, in-depth analysis and events
Start building with HiTechCloud and shape a complete AI and cloud strategy
Log in to explore an ecosystem of more than 30 products and the tools that come with them
Get started
Talk to our sales team and find the AI and cloud solution that fits your business
Contact salesCông ty Cổ Phần Giải Pháp Công Nghệ và Phần Mềm Phổ Tuệ
Address: 128 Đường Bình Mỹ, Bình Mỹ Commune, Ho Chi Minh City, Vietnam
Tax code: 0318222903, first issued by the Ho Chi Minh City Department of Planning and Investment on 20/12/2023, amended for the second time on 15/10/2024
Responsible for content: Nguyễn Thanh An
Hotline: Phone: 0865.920.041
Email: Email: info@photuesoftware.com
Email Abuse: abuse@photuesoftware.com
Certificate: Telecommunications service registration certificate No. 18/GCN-SKHCN, issued by the Ho Chi Minh City Department of Science and Technology on 10/03/2026.
Căn hộ OT03, Tòa nhà The Landmark 81, 720A Đ. Điện Biên Phủ, Vinhomes Tân Cảng, Phường Thạnh Mỹ Tây, Tp. Hồ Chí Minh
5F, Ginza Sky Building, 3-13-4 Ginza, Chuo-ku, Tokyo, Japan
Email: contact.jp@hitechcloud.asia
6F, Yanghwa Tower, 5 Teheran-ro 20-gil, Yeoksam-dong, Gangnam District, Seoul, South Korea
5F, Sugimoto Building, 1-4-7 Tokui-cho, Chuo-ku, Osaka, Japan
Email: contact.jp@hitechcloud.asia
7F, ONE FUKUOKA BLDG, 1-11-1 Tenjin, Chuo-ku, Fukuoka, Japan
Email: contact.jp@hitechcloud.asia
81-83 Campbell Street Surry Hills, NSW 2010, Australia
Email: sydney@hitechcloud.asia
150 E Brokaw Rd, San Jose, CA 95112, United States
Email: contact.us@hitechcloud.asia
36 Robinson Road, #20-01, City House, Singapore 068877
Email: contact.sgp@hitechcloud.asia
Mezzanine Floor, Jumeirah Living Marina Gate 3, Dubai Marina, Dubai, PO Box 121828
