About TcpNode

Dedicated servers, one machine per customer

For CUDA and AI engineers, as well as operations teams that need dedicated hardware. TcpNode provides entire machines for rent, never shared with other customers.

Rent a TcpNode DGX Spark Configure an AMD server

The two product lines are ordered separately. Machine details and service requests are handled through the console after delivery.

Handled by our engineering team

One support path, from hardware issues to support tickets

TcpNode is run by a small engineering team. The same team handles hardware, networking, and support tickets, focusing on machine delivery, infrastructure operations, and troubleshooting. No claims are made based on customer counts, uptime, or unverified performance figures.

For existing orders, communication takes place through console support tickets. Include the relevant order, what happened, the conditions under which it occurred, and any troubleshooting already completed. This helps link the report to the right machine. Login passwords, keys, and business data should never be shared on public pages.

TcpNode handles hardware and networking issues; the customer manages application compatibility, dependency versions, and in-system configuration. Requests for a hard reboot or OS reinstall should also be submitted through a support ticket, keeping machine administration separate from day-to-day workload configuration.

Two separate product lines

Choose a machine for development or general workloads

TcpNode offers distinct products based on architecture, operating system, and billing model. Each product line has its own ordering flow; configurations and quotes are not combined across products.

CUDA and AI development

NVIDIA DGX Spark

NVIDIA GB10 Grace Blackwell with 128GB LPDDR5x unified memory shared by the CPU and GPU. Designed for CUDA development, debugging, inference, and fine-tuning experiments on ARM64.

  • 20-core Arm CPU, 4TB NVMe.
  • 1 Gbps unmetered traffic, 1 static IPv4 address.
  • NVIDIA DGX OS only; access via SSH.
  • Rented per day with fixed specifications; models are not preinstalled.

Data is erased at the end of the rental period. Code, weights, and logs must be transferred before expiry. x86 software and Windows environments are not substitutes for checking ARM64 compatibility.

View TcpNode DGX specifications
Operations and business workloads

AMD EPYC 7543P dedicated server

32 cores and 64 threads, with a choice of 128GB or 256GB DDR4-3200 ECC memory. Suitable for virtualization hosts, databases, middleware, and Windows workloads.

  • 2×1TB NVMe direct-attached storage, no hardware RAID.
  • Standard 250 Mbps unmetered traffic, 1 dedicated IPv4 address.
  • 14 free operating systems, plus paid Windows Server.
  • Billed by the selected monthly rental term; options are confirmed on the configuration page.

Linux, FreeBSD, and Proxmox VE are accessed via SSH; Windows is accessed via RDP. Memory cannot be upgraded in place after provisioning; a different configuration requires a separate order.

View the TcpNode AMD lineup
Service delivery

Infrastructure and in-system operations are managed separately

The infrastructure behind machine delivery

This diagram illustrates service responsibilities. It is not a photograph of the facility or a live monitoring view.

Infrastructure diagram showing dedicated power and network equipment, with system configuration and business data managed by the customer
Power and networkInfrastructure managed by TcpNode
Dedicated machineOne machine per customer
System and dataConfigured, maintained, and backed up by the customer
Located in Tokyo, Japan.
Who handles what

TcpNode manages infrastructure; customers protect their own data

A dedicated machine does not remove the need for system administration. Clear ownership of hardware, software, and data helps determine the right response when something goes wrong.

What TcpNode handles

  • Hardware: Investigates hardware issues using the order details and reported symptoms.
  • Power and networking: Manages power and network access for the machine's infrastructure.
  • Support requests: Handles hard reboot and OS reinstall requests, as well as infrastructure faults.

Support tickets should reference the relevant order. Before any operation involving data, confirm that a restorable backup is available.

What the customer handles

  • System and software: Manages system configuration, dependencies, access permissions, and the application runtime.
  • Data and recovery: Plans and verifies backups and recovery for code, models, databases, and logs.
  • Redundancy: Configures software RAID, ZFS mirroring, and off-site backups as workload requirements dictate.

The AMD server's two NVMe drives are direct-attached and do not use hardware RAID. Mirroring is not a substitute for independent backups.

DGX Spark has no BMC or IPMI. Hard reboots and OS reinstalls are handled through console support tickets. Data is erased at the end of the rental period; backups and transfers must be completed before then.

Pricing

Published prices, with no setup fees

All TcpNode prices are settled in USD. There are no setup fees, contracts, or discounts. Machine and add-on costs are confirmed in their respective ordering flows; billing terms do not change the listed prices.

AMD pricing by term

The monthly total consists of the machine and selected options. Amount due = monthly total × number of months. Monthly, quarterly, semiannual, and annual terms are billed at 1, 3, 6, and 12 times the monthly total, respectively; the term does not change the listed monthly price.

DGX pricing by day

Amount due = number of days × daily rate. Rental periods range from 1–365 days. The fixed hardware does not offer AMD product-line add-ons such as extra drives, bandwidth, IPv4 addresses, or Windows.

TcpNode prices are settled in USD. Available payment methods are listed on the configuration page before payment. The console shows the current availability of each option and keeps a record of payments.

Before ordering

Built around one dedicated machine, not a cluster project

  • No multi-machine clusters

    TcpNode rents individual machines. Cluster networking across multiple machines, cross-machine scheduling, and distributed software deployment are not part of the service.

  • Models are not preinstalled

    DGX Spark comes with NVIDIA DGX OS and the CUDA software stack. Customers download model weights and check model licensing, dependencies, and ARM64 compatibility.

  • Separate billing for each product

    Each order follows the billing unit for its product. Charges for different products are not converted or combined.

  • No SLA compensation commitment

    Published information does not include uptime percentages or compensation formulas. Hardware and network issues are handled through order-related support tickets; customers are responsible for business continuity planning and backups.

Contact and support

For order-related issues, start with a console support ticket

Sign in to the TcpNode console to submit a ticket with the order details, issue, and relevant troubleshooting information. For account access or general enquiries, email support@tcpnode.com. Do not include passwords or private keys in email.

Report a machine issue to TcpNode

Hardware, networking, hard reboot, and OS reinstall requests are all handled through support tickets.

Open the TcpNode console