Rent for the length of your experiment

Rent a Mac mini M4
in Tokyo

For users in East Asia, run model experiments on a dedicated physical Mac. Get 16GB unified memory and a 256GB SSD, with rentals from one day—no need to buy a machine for a single experiment.

Rent OsxNode M4-16 View all plans

From $21.6/day · Provisioned the day payment is confirmed · Stops when your term ends unless renewed

Only one configuration is currently available. Dedicated access does not mean unlimited memory: check the combined usage of model weights, context cache and other processes before loading a model.

Specs & rental terms

Check the memory, then choose a term

All four billing terms use the same physical machine; they are not four performance tiers. Choose a prepaid term based on how long you expect to run your experiment. The machine stops automatically at the end of your term unless you renew, so there are no further charges.

OsxNode M4-16

One tenant gets exclusive use of one Mac mini M4. It is not a virtual machine, and its unified memory and GPU cores are not shared with other tenants.

Only configuration available
Chip
M4
Unified memory
16GB
Built-in SSD
256GB

Choose a billing term

Switch terms to see the prepaid price. Confirm the final term, node and add-ons on the order page.

$21.6 / day

Daily prepaid
$21.6/day
Weekly prepaid
$58.2/week
Monthly prepaid
$107.8/month
Quarterly prepaid
$293.2/quarter

Choose a term that fits your experiment—not a full month by default

For a one-off environment check or inference run, start with a daily term. If you expect to run continuously for several days, compare the weekly and daily rates; for longer projects, compare the monthly and quarterly rates. Changing your term does not add unified memory or change the chip or built-in storage.

The 16GB is unified memory for the entire machine, not memory reserved exclusively for model weights. Start with a short context and small batches to check memory pressure. 8-bit models with 13B parameters or more are not recommended on this configuration. No higher-memory configuration is currently available, and model weights are not installed for you.

Choose OsxNode M4-16 Check add-on prices

Check available disk space before downloading models and datasets. An additional SSD adds storage; it does not increase unified memory. Check the console for the current availability status.

Network reference

Tokyo node for East Asia

When choosing a node, consider the network route between you and the node, and whether your work requires sustained desktop interaction. The range below is a reference round-trip latency, not a live test from your network or a guaranteed average.

Reference round-trip latency

30~60ms

Reference range for East Asia. Actual latency depends on your network.

Latency affects interaction, not inference speed

Dragging, typing and screen refresh in a remote desktop are affected by round-trip latency, jitter and your local network. When a model is running on the machine, computation happens on the physical node; access latency cannot tell you its tokens per second.

If you need to work with a graphical interface repeatedly, test how quickly the screen responds. For log monitoring or long-running jobs, you can use SSH with tmux to keep your session alive. Dataset upload time also depends on available bandwidth and the number of files, so the latency range alone cannot predict how long a transfer will take.

Open the browser-based remote desktop from the console after signing in. Losing the browser connection does not stop the machine, but your job may be affected when the rental term expires. Check how much time remains before starting a long-running job.

From order to ready

Four steps to provisioning and your first connection

Setup and initial checks take as long as your own process requires. Payment confirmation and provisioning status are shown in the console. Service is provisioned the day payment is confirmed; network confirmation time is not a fixed-minute guarantee.

  1. Check the configuration and term

    On the order page, confirm the model, billing term, destination node and any add-ons. Estimate the size of your model weights and output files first, then decide whether you need an additional SSD.

    Estimated time: Allow around 3–5 minutes to check your choices if you already know what your experiment needs. This is not platform processing time.

  2. Prepay in USD

    Payment is accepted only via USDT-TRC20 or Visa / Mastercard / Amex through Stripe. All payments are settled in USD. Available payment gateways are determined by the response from the backend.

    Estimated time: Depends on payment verification or transaction confirmation. Until payment is confirmed, a screenshot of a charge does not replace the order status.

  3. Get your credentials in the console

    Provisioning takes place the day payment is confirmed. The node and connection credentials will appear in the console when delivery is complete. Orders that need a manual review are subject to the status shown for that order; instant unattended delivery is not guaranteed.

    Estimated time: Provisioning is completed that day. No fixed schedule for manual reviews is provided. Keep your order number and check progress through a console support ticket.

  4. Connect and run a small test

    After signing in to the console, open the browser-based remote desktop, change the initial password, and check disk space and memory pressure. Run a small test before downloading the full dataset and starting a long-running job.

    Estimated time: Allow around 5–10 minutes for your own checks once your credentials are ready. Model downloads and environment setup take additional time.

Include environment setup, data uploads and result exports in your rental term. Back up your code, weights and experiment results before the term ends; do not rely on the machine to store data after expiry.

Choosing a node

Before you order, check these three network and location details

Use these answers to assess whether a node suits your experiment. Reference latency and node location are not guarantees that a task will finish successfully.

I'm in East Asia. Will my latency definitely be 30~60ms?

No. The reference range is not a per-user guarantee. Actual round-trip latency to the Tokyo node depends on your local connection, your provider's routing, network congestion and wireless conditions. Remote desktop performance is also affected by jitter. On your first connection, test typing and screen refresh before deciding how much interactive work to do remotely.

If I choose Tokyo, will every download source be closer?

No. The node's location tells you where the physical machine is; model download sources, package repositories and dataset services each have their own network routes. Test with an actual download rather than using the reference access latency for East Asia to estimate speeds from third-party sources.

Can I move my experiment to another node without changes once it has started?

Do not plan your workflow around live migration. To switch nodes, first confirm the new order's configuration and provisioning arrangements. Export your code, environment manifest, model files and intermediate results, then restore and verify the job on the destination machine. Running processes, memory state and graphical sessions should not be treated as data that can simply be moved.

Set a clear rental term for this experiment

Start by checking the 16GB unified memory and how many days you expect to run. Confirm the node, term and current availability on the order page. Provisioning takes place the day payment is confirmed.

Rent OsxNode M4-16 Compare terms & add-ons