1. Service Scope
TcpNode offers daily rentals of complete NVIDIA DGX Spark systems. Each device is dedicated to a single customer, with no hardware resources shared with others. The service is intended for CUDA development and debugging, ARM64 container deployment, local inference, and fine-tuning experiments.
These terms apply to public orders, machine delivery after payment, and order and support-ticket activity in the console. General account rules are in Section 2. DGX Spark specifications, system details, and expiration requirements are in Sections 3–9. The scope and separate ordering rules for the other product line are in Section 10.
Machine availability is shown in the console in real time. Service descriptions do not guarantee delivery times, availability percentages, or model performance. Suitability for a particular workload depends on its architecture, software dependencies, and memory requirements.
2. Accounts & Credentials
The registration email address must be accurate, active, and able to receive messages. It is used for account verification, linking orders, and delivery notices. Keep access to the email account, account password, verification codes, and machine credentials secure. Do not expose them in public code repositories, screenshots, or shared documents.
The console provides access to delivered machines, order history, billing records, and support tickets. If account details are incorrect, the email address is inaccessible, or activity appears unauthorized, contact TcpNode through a console ticket or support@tcpnode.com and identify the relevant order and what needs to be investigated. Do not send passwords or verification codes directly by email.
Details on how account, order, and support-ticket information is handled are in the TcpNode Privacy Policy. Submit only the information needed to resolve the issue. Remove secrets, access tokens, and other sensitive content from logs or attachments that contain business data.
3. DGX Orders & Payment
Order DGX Spark through its dedicated order page; it is not part of the configuration flow for other products. Before submitting an order, check the rental duration, fixed specifications, amount due, and payment options, and confirm each of these three points: ARM64 cannot run x86 programs or Windows; the only available OS is NVIDIA DGX OS; data is erased at expiration.
All orders are billed in US dollars (USD). Available payment options are those shown on the configuration page before payment is confirmed. Payment is confirmed by the actual payment receipt and order status, not by clicking the payment button or leaving the payment page.
Payment and delivery status are provided through the order process. After delivery, sign in to the console to view the machine and credentials. Check the TcpNode DGX Spark overview for compatibility details, or visit the order page to confirm the rental.
4. DGX Billing
DGX Spark rentals cost $16.7 per day. Rental periods must be whole numbers from 1 to 365 days. The amount due is calculated as “amount due = number of days × daily rate.” There are no discounts, setup fees, or contracts; monthly, quarterly, semiannual, and annual billing periods are not available.
The rental duration must be confirmed before checkout. The amount due is calculated from the number of days and the listed daily rate. DGX and other products are ordered and billed separately; rental periods and billing formulas cannot be combined or transferred between orders.
Data is erased when the rental expires. Allow enough time for environment setup, model downloads, workload execution, and exporting results. Transfer any code, weights, outputs, and logs that need to be retained before expiration.
5. DGX Specifications
The system uses NVIDIA GB10 Grace Blackwell, with a 20-core Arm CPU (10×Cortex-X925 + 10×Cortex-A725), 128GB LPDDR5x unified memory, and 4TB NVMe storage. CPU and GPU share the unified memory; it is not a separate 128GB of GPU memory.
Network service is fixed at 1 Gbps with unlimited traffic and includes one static IPv4 address. The data center is in Tokyo, Japan. DGX Spark hardware and network specifications are fixed; bandwidth changes, additional IP addresses, extra storage, and memory upgrades are not available.
TcpNode is responsible for machine hardware, power, and network connectivity. Customers are responsible for software configuration, business data, data redundancy, and backups. The 4TB NVMe capacity does not include backup services. Store important files separately in a customer-managed backup location.
6. DGX System Limits
DGX Spark runs NVIDIA DGX OS, based on Ubuntu 24.04, with CUDA and the NVIDIA AI software stack. The system uses ARM64 architecture; other distributions are not available, and models are not preinstalled.
ARM64 systems cannot run x86 programs or Windows. Check target architecture and software compatibility before installing dependencies, compiling programs, or deploying containers. Software available only as x86 packages or images should not be treated as confirmed compatible workloads.
Root access is available for installing other tools. Customers are responsible for software they install, configuration changes, and model files. Before changing the toolchain, record the existing environment and keep restorable copies of code and configuration. Reinstallation must be requested through a support ticket.
7. DGX Access & Delivery
DGX OS is accessed over SSH. After payment and delivery, sign-in credentials are available in machine details in the TcpNode Console ; use the connection address and account information provided for the specific order.
Customers are responsible for protecting connection credentials and any keys created on the system. Do not make access public. After changing SSH settings, firewall rules, or account permissions, check how those changes affect future connections.
For first-time connection instructions, see the TcpNode Post-Payment Connection Guide. If a connection fails, note when it happened, the access method used, and the error message, then submit a support ticket for investigation.
8. Support & Scope
Report hardware faults, hard reboots, and OS reinstalls through a console support ticket. DGX Spark does not have BMC/IPMI, and customers cannot directly access this type of hardware management interface.
Include the order details, symptoms, conditions, configuration changes already made, and relevant error logs in support tickets. For a hard reboot or reinstall, specify the requested action and first transfer any data that needs to be kept. These actions may interrupt running workloads or change system contents.
TcpNode investigates hardware and network issues. Customers are responsible for software and models they install, business services, and data backups. Public contact options are submitting a ticket in the TcpNode Console and email at support@tcpnode.com. No fixed response time is specified.
9. Use & Expiration
The service must not be used to abuse networks, send spam, attack other systems, or access systems without authorization. Customers are responsible for managing software and externally exposed services on the machine, and for preventing misuse caused by exposed credentials, misconfiguration, or runaway processes. Service may be suspended for violations of these use restrictions.
Data is erased when DGX Spark rental expires. Transfer and verify any model weights, code, experiment records, container configurations, and business results that need to be retained before expiration. The machine cannot be used to store data after the rental ends.
TcpNode is responsible for hardware, power, and network connectivity. Customers are responsible for system configuration, applications, data, and backups. Responsibility for impacts caused by circumstances beyond reasonable control or customer actions depends on the actual cause. These terms do not specify refund percentages or compensation formulas.
10. Separate Terms for AMD Dedicated Servers
TcpNode also offers AMD EPYC 7543P dedicated servers on a monthly rental basis, with one machine dedicated to a single customer. Each system has a 32-core, 64-thread CPU; a choice of 128GB or 256GB DDR4-3200 ECC memory; 2×1TB NVMe drives directly attached; 250 Mbps with unlimited traffic; and one dedicated IPv4 address. Hardware RAID is not included. Customers are responsible for software RAID, ZFS mirroring, or other data redundancy arrangements, as well as independent backups. Memory cannot be upgraded on the same machine after provisioning; changing the memory tier requires a separate order.
Order a dedicated server through the TcpNode Dedicated Server Configuration page . Monthly prices by system and optional add-ons are listed on the TcpNode AMD Monthly Rental Rates. Add-on storage, bandwidth, extra IPv4 addresses, and Windows licenses are included in the monthly total. The amount due is calculated as “amount due = monthly total × number of months”: monthly is 1 month, quarterly is 3 months, semiannual is 6 months, and annual is 12 months. There are no discounts, setup fees, or contracts. Subscriptions renew automatically for the selected billing period and can be canceled in subscription management in the TcpNode Console . Billing is in US dollars; available payment options are those listed on the configuration page.
AMD servers offer 14 free operating systems: AlmaLinux 8/9/10, CentOS Stream 9/10, Debian 11/12/13, Ubuntu Server 22.04/24.04/26.04 LTS, FreeBSD 13, and Proxmox VE 8/9. Windows Server Standard and Datacenter are available in versions 2019/2022/2025, with licenses billed monthly. System languages are English, Traditional Chinese, and Simplified Chinese. Customers manage software, virtualization configuration, and business data on the system. Preconfigured data redundancy and business backups are not included.
After payment and delivery, IP addresses and login credentials are available in the console. Connect to Linux, FreeBSD, and Proxmox VE over SSH, and to Windows over RDP. If multiple IPv4 addresses are assigned, SSH and RDP use the first address in the list; customers configure the others within the system. Report faults, request hard reboots, and request reinstalls through a console support ticket. The account rules in Section 2 and the restrictions on abuse in Section 9 also apply. TcpNode is responsible for hardware and network connectivity; customers are responsible for systems and data, including backups.
11. Governing Law & Disputes
These terms are governed by the laws of the jurisdiction where the platform operator is based. Questions about interpreting these terms, fulfilling orders, or assigning responsibility may be submitted with relevant order records and details through a console support ticket or to support@tcpnode.com for discussion between the parties.
Disputes that cannot be resolved through discussion will be submitted to a court with jurisdiction in that jurisdiction. Requests concerning account information, order records, or data will be reviewed in light of the scope of these terms and the privacy policy.