Start with peak memory, build-cache size, and storage throughput for a single task, then consider how many pipelines must run at once. OakVM offers two fixed configurations, and every order maps to one dedicated physical machine—not shared compute carved from a single node.
Both tiers include the full macOS graphical interface and command-line environment, and both are dedicated Apple Silicon physical nodes. Don’t choose by chip name alone—check total usage across peak memory, derived data, build caches, archives, and temporary files.
M4 · 16GB RAM · 256GB SSD
M4 Pro · 64GB RAM · 2TB SSD
| Comparison | OakVM M4 | OakVM M4 Pro |
|---|---|---|
| Chip | M4 | M4 Pro |
| Memory | 16GB | 64GB |
| Built-in SSD | 256GB | 2TB |
| Per day | $20.3 | $59.1 |
| Per week | $54.9 | $159.6 |
| Per month | $101.7 | $295.5 |
| Per quarter | $276.6 | $803.8 |
| Node locations | Singapore, Tokyo, Seoul, Hong Kong, US East, US West | Singapore, Tokyo, Seoul, Hong Kong, US East, US West |
A five-person team may run one serialized release task—or tests, archives, and model validation at the same time. Identify the heaviest task first, choose the machine accordingly, and scale concurrency by adding dedicated physical nodes.
With stable project dependencies, daily work typically involves pulling code, installing dependencies, running tests, creating archives, and exporting artifacts, with one primary pipeline running at a time.
Multiple build, test, or data-processing tasks may consume memory and disk bandwidth simultaneously, or the project may need to retain multiple dependency caches, symbol files, archives, and experiment data.
When tasks don’t need shared memory, assigning them to multiple dedicated nodes usually makes project, cache, and retry isolation easier. A higher-spec single node does not mean unlimited concurrency.
OakVM scales in units of one dedicated Apple Silicon physical node. Each node keeps its own system environment, project directory, cache, and task labels; the scheduler simply sends jobs to nodes with matching labels.
Pending tasks organized by project, branch, priority, and required Xcode environment.
Keep signing assets, project directories, and release toolchains fixed to reduce environment switching.
1 ORDER = 1 DEDICATED NODEUse an independent runner label for test jobs so they do not compete with release tasks on the same node.
Independent cache and task queueIsolate models, dependencies, and experiment output so high-memory tasks don’t consume everyday build capacity.
Choose configuration by task typeIf two tasks must share data in the same process or the same high-speed local storage, choose a single configuration that accommodates their combined peak usage. If tasks are independent, adding nodes improves concurrency, access isolation, and failure diagnosis.
Use the table as an initial guide to regional direction for remote operations and code transfer. The figures are medians from 20 ICMP requests to each node, sent over common commercial broadband connections from each access point.
| Access point \ Node | Singapore | Tokyo | Seoul | Hong Kong | US East | US West |
|---|---|---|---|---|---|---|
| Singapore access point | 22 ms | 71 ms | 83 ms | 39 ms | 238 ms | 174 ms |
| Tokyo access point | 69 ms | 34 ms | 41 ms | 52 ms | 181 ms | 112 ms |
| Seoul access point | 81 ms | 40 ms | 33 ms | 46 ms | 194 ms | 128 ms |
| Hong Kong access point | 37 ms | 54 ms | 48 ms | 24 ms | 218 ms | 151 ms |
| US East access point | 236 ms | 179 ms | 191 ms | 216 ms | 29 ms | 76 ms |
| US West access point | 171 ms | 109 ms | 126 ms | 149 ms | 74 ms | 28 ms |
If your team frequently uses a graphical interface, prioritize round-trip latency from the main users’ access points to each node. For CI/CD-driven tasks, also consider the code repository, dependency sources, artifact storage, and team handoff time zones.
These figures are network medians from fixed sample conditions. They do not represent Xcode compile speed or a continuous network metric. Actual results vary with local carriers, cross-border routing, Wi-Fi, VPNs, congestion, and task timing.
Start by filtering for low-latency regions from the access point closest to your main workplace, then factor in team handoff time zones. All six nodes run normally 365 days a year; actual availability is shown in real time in the console.
Cost planning comes down to two questions: how long the node will run continuously, and whether the environment will need to be rebuilt repeatedly during that time. The examples below use current listed prices, without discount assumptions or blended-term calculations.
For incident reproduction, version archiving, dependency validation, or short experiments with a clear start and end time. If the task fits within one day, check the daily cost directly.
For several continuous days of testing, fixes, archiving, and distribution, while keeping the same environment through release. No need to prepare dependencies and caches from scratch every day.
For teams that trigger builds, tests, and archives throughout the workweek, or need to retain runner labels, dependency caches, and project directories over time.
For teams that have confirmed their pipelines will run across multiple release cycles and want fixed toolchains, cache policies, and node responsibilities.
Measure how long the node actually needs to retain its environment, caches, and task queue—not just the duration of individual builds.
Include dependency downloads, cache warming, toolchain checks, and human handoffs instead of comparing lease prices alone.
When two independent pipelines must run simultaneously, budget for two dedicated nodes rather than assuming one node can handle unlimited tasks.
Add storage separately to the order: +1TB SSD is $2.9/day, $7.8/week, $14.5/month, or $39.4/quarter; +2TB SSD is $5.8/day, $15.6/week, $29/month, or $78.8/quarter. Thunderbolt 5 daisy chaining costs $1.7/day, $4.5/week, $8.3/month, or $22.6/quarter per node.
Sample one of your heaviest real tasks rather than an empty project or a result taken just after clearing caches. Record the five items below in your purchasing notes to create a reviewable baseline for future node or lease changes.
Record memory usage at the peak of archiving, parallel testing, or model tasks. If 16GB is near its limit or tasks are frequently interrupted by memory pressure, evaluate the Pro tier with M4 Pro and 64GB memory.
Add up project directories, dependencies, derived data, archives, symbol files, logs, and temporary files. The base tier includes a 256GB SSD; the Pro tier includes a 2TB SSD.
Distinguish parallel threads from pipeline concurrency. When multiple independent projects run simultaneously, consider adding dedicated nodes and isolate tasks with runner labels instead of crowding every queue onto one machine.
Choose among Singapore, Tokyo, Seoul, Hong Kong, US East, and US West based on primary access locations, code-transfer paths, and team handoff times.
If build caches, historical archives, or experiment data cannot be moved out regularly, select +1TB SSD or +2TB SSD in advance. Don’t treat free system-disk space as a long-term artifact-management strategy.
When a single task’s peak fits within 16GB, 256GB SSD is enough for the active project, and you mainly run one standard iOS build pipeline.
When large workspaces, parallel tasks, AI experiments, or high storage throughput need 64GB of memory and 2TB SSD for greater capacity headroom.
When tasks are independent but must progress simultaneously, add nodes to scale concurrency and assign runner labels by project, environment, or priority.
OakVM offers two dedicated Apple Silicon physical-node configurations, both non-virtualized, with billing in USD. Choose nodes in Singapore, Tokyo, Seoul, Hong Kong, US East, or US West; actual availability is shown in real time in the console.