Rent a dedicated Mac mini M4 by the day and run your experiment with 16GB of unified memory—without buying a machine first.
Currently available: OsxNode M4-16 with M4 / 16GB unified memory / 256GB SSD. Choose a prepaid daily, weekly, monthly, or quarterly term. Starting at $21.6/day; the machine shuts down at the end of the term unless renewed, with no further charges.
Add up the memory needed for model weights, context, and other processes first. An SSD add-on gives you more storage, not more unified memory.
One machine is assigned to one customer for the full rental term.
Daily, prepaid
$21.6 / day
A good fit for a one-day test to see whether your model loads and runs.
The base price covers the full machine for the selected term; add-ons cost extra. Check the console for live availability and your final order total.
Choose OsxNode M4-16These are four base prices for the same machine, not four hardware tiers. Switching terms does not change the M4, 16GB unified memory, or 256GB SSD. To keep a task running longer, renew before the term ends and check the updated expiry date in the console.
| Billing term | Price | Suggested experiment length | Check before paying |
|---|---|---|---|
| Daily | $21.6/day | 1–2 days: initial environment checks, a single inference run, or a short build. | Allow time for environment setup, downloading weights, and exporting results. |
| Weekly | $58.2/week | 3–7 days: compare the weekly rate for extended tuning or experiments that need another look after the weekend. | List the tasks you need to rerun so you don't estimate based on inference time alone. |
| Monthly | $107.8/month | More than 3 weeks: compare the monthly rate if you want to keep the same experiment environment. | A longer rental does not increase the memory limit. Check that your model can run first. |
| Quarterly | $293.2/quarter | For a full-quarter project that needs continuous access to the same physical machine. | Choose a term that matches your project; there's no need to prepay for a full quarter for a short task. |
All four terms are prepaid. The machine shuts down at expiry unless renewed, with no additional charge. Shutdown does not mean your results have been backed up: export models, datasets, experiment logs, and any configuration you need to keep before the term ends. Don't use an expired machine as long-term storage.
The base configuration includes a 256GB SSD. Multiple model versions, datasets, and build caches can all take up disk space. Choose an SSD add-on for your workload, then point model caches and dataset directories to the mounted add-on after delivery. Check the mount path and available space before downloading weights.
Store model weights, datasets, and experiment output. Ideal if you want to avoid downloading the same files repeatedly.
Adds disk space only; model loading is still limited by 16GB of unified memory.
For experiments that need multiple sets of weights, dataset versions, and intermediate checkpoints. Estimate your total storage needs before choosing.
An add-on drive is not an off-term backup service; export your files before expiry.
Priced per machine. Choose this only if your experiment needs an inter-machine connection, and check that your application supports the relevant multi-machine workflow.
Linking machines does not combine their unified memory or guarantee automatic acceleration in your model software.
Each add-on has its own term-based price and is billed separately from the base machine. Check the selected term, quantity, and total at checkout. Available options are confirmed by the console.
Parameter count is only one factor. Quantization, context length, KV cache, batch size, and other running tools all affect peak memory use. These are planning recommendations, not benchmark results for a specific model or a guarantee that it will run.
| Workload | Suggested machine | Suggested add-ons | What to expect |
|---|---|---|---|
| Light builds | OsxNode M4-16 | Start with the base 256GB SSD; consider +1TB SSD if you have many caches and build artifacts. | Run a full build and record disk and memory use. Don't load a model at the same time when measuring peak build requirements. |
| Everyday development | OsxNode M4-16 | Consider +1TB SSD if dependencies, project copies, and caches exceed the base storage budget. | Your IDE, simulator, and inference process share 16GB of unified memory. The more tasks you run in parallel, the less memory remains for the model. |
| LLM inference | OsxNode M4-16 | Consider +1TB SSD for multiple quantized weight sets; assess +2TB SSD if you have large datasets as well. | Start by testing a 3B or 7B 4-bit model with a short context and small batch size. Loading model weights over approximately 12GB is not recommended; 8-bit models with 13B or more parameters are not recommended on this machine. |
| Small-scale LoRA fine-tuning | OsxNode M4-16 | Choose +1TB SSD or +2TB SSD based on the combined size of your datasets, checkpoints, and exports. | Start by testing a smaller model with short sequences and a small batch size. Even with a 4-bit base model, training can use more peak memory than inference. Weight size alone cannot tell you whether training will work. |
To decide whether to continue, start with a short input and confirm that the model loads and generates output. Then check Memory Pressure in Activity Monitor. If it turns yellow, sustained swapping occurs, or the system terminates a process, shorten the context, reduce the batch size, or close other workloads. If it still won't run, switch to a smaller model rather than adding SSD storage.
OsxNode M4-16 is listed in four physical locations: Singapore, Tokyo, Seoul, and Hong Kong. Location determines where the machine is hosted; it does not change this model's unified memory or base SSD. The table shows the model lineup, not real-time availability.
| Model | Singapore | Tokyo | Seoul | Hong Kong |
|---|---|---|---|---|
| OsxNode M4-16 | Available | Available | Available | Available |
Check the console for real-time availability. Remote desktop performance depends on your local network, connection route, and image quality settings. For long-running inference, you can manage tasks over SSH instead of relying on desktop responsiveness as a measure of compute performance.
Every step from selection to connection should have a result you can verify. Same-day provisioning does not mean the machine is ready to connect the moment you pay. If payment or delivery needs a manual check, open a console ticket with your order number—don't pay again.
Select OsxNode M4-16, your rental term, location, and add-ons. Result: the order lists the configuration and amount due. Check each item before paying.
Choose an available payment gateway returned at checkout and complete payment. Result: the order shows a confirmed payment status. A payment receipt does not mean the machine has been delivered.
Sign in to the console to view the location, connection details, and rental term. Result: access credentials appear for the instance; use the delivery details to confirm the expiry date.
From the signed-in console, open the remote desktop and change the initial password on first login. Result: once you're in macOS, install your own tools and test your first task.
Only the two payment methods below are supported, and all payments are settled in US dollars (USD). This page lists supported methods; the order's available gateways are determined by the response from the backend. Never send payment to an address or recipient you found independently based on text on this page.
Follow the network, recipient, and amount shown at checkout. Confirm that the network is TRC20 before paying.
For billing questions, provide your order number and transaction hash to help us check the payment. Never share your private key or recovery phrase with support staff.
A transfer screenshot is not proof that the machine has been provisioned. Wait for payment confirmation.
Credit card payments are processed by Stripe. Check the USD amount at checkout, then complete any verification required by the payment page.
If your payment has not been confirmed, check the order status first. If you still have questions, open a ticket with your order number and the time of the issue.
You don't need to provide your full card number, security code, or payment verification code when contacting support.
A cloud Mac gives you dedicated use of a physical machine for a set period; it does not change the hardware's resource limits. Review these three constraints before budgeting for your experiment.
The current lineup includes one model with 16GB of unified memory. If your workload requires more, a longer rental or an SSD add-on won't solve the problem.
Check the combined requirements for model weights, runtime caches, and system processes first.
This machine uses M4 compute resources and does not include a separate accelerator card. Check compatibility first if your software requires another hardware platform or a specific driver.
Support for specific models in tools such as MLX depends on the software and its version.
You're responsible for preparing and installing models, datasets, and experiment dependencies. You must also verify software licensing and ensure your data use complies with applicable requirements.
Allow for downloads, environment checks, backups, and cleanup when planning your rental time.
OsxNode M4-16, with 16GB unified memory and a 256GB SSD, from $21.6/day. Choose a term and location, then check your add-ons and total before paying. The machine shuts down at expiry unless renewed.