How JBlanked Compares ZimaBoard 2 and Raspberry Pi 5 for a Home Lab

Eva Wong is the Technical Writer and resident tinkerer at ZimaSpace. A lifelong geek with a passion for homelabs and open-source software, she specializes in translating complex technical concepts into accessible, hands-on guides. Eva believes that self-hosting should be fun, not intimidating. Through her tutorials, she empowers the community to demystify hardware setups, from building their first NAS to mastering Docker containers.

Thank you to JBlanked for sharing a detailed look at his everyday home lab. His comparison gives a useful answer: ZimaBoard 2 suits a small server built around x86 applications, SATA storage, and wired networking; Raspberry Pi 5 remains a strong fit for projects that need its GPIO and camera ecosystem. Neither choice removes the need to plan memory capacity, backups, or software compatibility.

The review follows an 8GB ZimaBoard 2 running ZimaOS, Docker applications, DNS filtering, file services, and an AI-related workflow. Crucially, an RTX 5070 Ti handles the demanding model computation in that workflow. Watch JBlanked’s original comparison, embedded below from the price-and-firmware discussion, then use the article to separate the creator’s experience from the checks your own deployment needs.

Start with the job your home lab needs to do

JBlanked’s starting point is a familiar one: a Raspberry Pi that began as an inexpensive place to update scripts and test firmware gradually became a machine expected to keep services available. Once a board carries daily applications, the question changes. The most useful comparison is the complete working system, including its storage, network connections, software architecture, and recovery process.

A sensor project and a small NAS can both be called a home lab, but they put different demands on the hardware. A camera ribbon connector may determine the first project’s platform. A supported application image and a direct connection to two SATA drives may matter more for the second. Start by listing dependencies that would be difficult to replace, rather than ranking the boards by memory capacity alone.

Your main workload What to check first Why it changes the choice
Sensors, robotics, or a Pi camera GPIO, camera connectors, and accessory support The peripheral ecosystem can be more important than general server connectivity.
Docker services and small databases Image architecture, persistent storage, and combined RAM use An installable application still needs a maintainable data path.
A two-drive file server Drive connections, power, permissions, and independent backups Drive attachment is only one part of reliable storage.
GPU-assisted local AI GPU memory, supported backend, connection, and power The small host and the inference accelerator have different responsibilities.

For a concrete example of a continuously available application, see LADTech’s Minecraft server with Crafty and Playit.gg. That build turns the hardware question into a testable service: can a client join, can friends reach it remotely, and can its world be restored after a problem?

What changes with ZimaBoard 2’s built-in connections?

The review uses the 832 configuration, with 8GB RAM and 32GB eMMC. The other ZimaBoard 2 configuration provides 16GB RAM and 64GB eMMC. The current ZimaBoard 2 specifications list an Intel N150 processor, two SATA connections, two 2.5GbE ports, USB, Mini DisplayPort, and PCIe expansion. Those connections explain much of JBlanked’s interest in using the board as a server.

JBlanked shows the PCIe connector on the side of ZimaBoard 2
The hardware walkthrough highlights the board’s PCIe connection. Expansion still depends on the selected device, driver, physical clearance, and power arrangement. Video frame at 3:05, courtesy of JBlanked.

Native SATA connections reduce the number of adapters between the machine and its drives. Dual Ethernet provides two physical network paths. PCIe creates an expansion option. None of those features configures itself: a second Ethernet cable does not establish isolation, and an expansion connector does not guarantee that a particular card, driver, or power arrangement will work.

The Raspberry Pi 5 also has expansion options. Its official hardware specifications include a PCIe interface, the 40-pin GPIO header, and camera/display connectivity. An NVMe arrangement needs the appropriate adapter and physical layout. Compare the complete assembly you will actually build, including enclosure clearance and cooling, rather than assuming one board is permanently limited to its default boot medium.

For a server that rarely needs a monitor, management and recovery access matter more than the display connector’s convenience. Keep a way to reach the console when a network setting fails. If the machine will also provide DNS or routing, record which services depend on it before scheduling a reboot.

How JBlanked uses ZimaOS and Docker every day

JBlanked describes a mix of AdGuard, MetaTrader-related work, APIs, llama.cpp, and local file access. His account of moving a troublesome workload from another machine is useful as a user story, but it is not a general uptime guarantee or a controlled application benchmark. An application’s version, runtime, resource limits, and surrounding services can change the result.

ZimaOS application dashboard shown during JBlanked’s daily-use review
The ZimaOS dashboard illustrates the creator’s application-focused workflow. Check the image architecture, persistent data, and resource use of each service you plan to run. Video frame at 5:50, courtesy of JBlanked.

The x86 platform can simplify deployment when the required software is distributed for that architecture. However, the review’s broad claim about running every Docker image needs qualification. A container image targets an operating system and architecture. x86 hardware does not automatically run an Arm-only image, and a Linux container host does not make a Windows application a native Linux service.

Docker’s multi-platform image documentation explains how an image can contain variants for different platforms. A native Arm build can run without emulation on a compatible Arm host. Conversely, an amd64-only dependency may make an x86 machine the simpler option. Check the actual application, rather than treating all software on either architecture as equivalent.

A useful application migration check

  1. Identify the application image, version, and supported architecture.
  2. Record its configuration, credentials, ports, and persistent data directories.
  3. Restore a copy of its data into a test deployment before changing the working service.
  4. Verify an ordinary user action and confirm that newly written data survives a restart.
  5. Measure the whole stack under normal use before adding another application.

DNS filtering adds another dependency: clients need to use the selected resolver. Installing AdGuard does not prove that every device sends its queries there, and domain filtering does not remove every kind of advertisement. The separate Pi-hole and Tailscale walkthrough illustrates the same need to verify the client’s actual DNS path, using a different filtering application.

Does this solve the RAM-upgrade question?

The video’s title raises Raspberry Pi’s restrictions on RAM-chip replacements. The primary Raspberry Pi forum discussion about a Compute Module 5 RAM upgrade provides a narrower, useful context. An engineer explains that devices are tied to their original RAM capacity and additional memory attributes, citing untested modified boards and their support burden. This concerns board-level chip replacement, rather than an ordinary removable-memory upgrade.

JBlanked discusses operating-system choice during his ZimaBoard 2 comparison
JBlanked discusses operating-system choice in the firmware and ecosystem section. The ability to install a different OS does not change memory capacity: ZimaBoard 2 also uses soldered RAM. Video frame at 8:50, courtesy of JBlanked.

ZimaBoard 2 also has soldered RAM. The official hardware comparison identifies its 8GB and 16GB LPDDR5 configurations as fixed onboard memory. Choosing it for SATA, networking, or software compatibility should not be presented as gaining a socketed RAM upgrade path.

Keep three questions separate: can you replace the operating system, can you attach more storage, and can you replace the memory module? A platform may offer the first two while its RAM remains fixed. That distinction prevents an attractive expansion story from becoming a misleading promise about the component you actually need to grow.

Plan memory for the combined workload. The operating system, application processes, database caches, background jobs, and any local inference all share resources. Leave headroom for ordinary peaks, then observe whether a slow period is caused by memory pressure, storage activity, or CPU saturation. Buying more RAM cannot fix a missing driver or an unsupported application image.

Keep the AI host separate from the AI accelerator

JBlanked explicitly says the board does not perform the demanding 27B-model work by itself. He describes an RTX 5070 Ti handling the heavy computation while ZimaBoard 2 provides an always-available front end for requests and application traffic. That division of responsibilities is the most useful architectural lesson in the AI section.

Part of the system Role in the described workflow What to verify yourself
Small server Runs supporting applications and routes requests. Service availability, RAM use, persistent configuration, and access controls.
GPU system Provides acceleration for the demanding model workload. Available VRAM, runtime support, power, and thermal behavior.
Inference runtime Loads the selected model and serves requests. Model format, quantization, context, backend, and concurrency.
Storage Holds weights, application data, and logs. Capacity, actual load time, retention, and backups.

The llama.cpp project supports multiple compute backends, so its name alone does not reveal where a workload runs. The video’s reported throughput improvement should remain an observation about the creator’s configuration. Without matched model settings, connection details, software versions, and measurements, it does not establish a general percentage advantage for either host architecture.

Likewise, local deployment still has finite capacity. A request can queue behind another request, a large context can consume additional memory, and the accelerator can become unavailable. If the front end remains on while the GPU system is off, test what the application does: wait, reject the request, or use another explicitly configured path. Use the model hardware requirements directory to investigate the particular model, rather than treating the board’s presence as proof that the model fits.

Plan storage and compare the whole working system

File streaming and photo access form another part of JBlanked’s routine. A practical layout can separate boot storage from application data and larger media, but every directory still needs a clear destination. Before adding a container, know where its database, uploads, and logs will be written and what happens if the data drive is temporarily unavailable.

JBlanked handles ZimaBoard 2 and accessories during the connectivity discussion
The physical setup is part of the storage decision. Account for drive connections, cabling, cooling, and recovery copies alongside the board itself. Video frame at 10:15, courtesy of JBlanked.

eMMC, microSD, and SSDs can all fail, and changing the storage medium does not provide power-loss protection or an independent backup. Native drive connections simplify assembly; recoverability comes from saved configuration, usable copies of important data, and a restore procedure that has actually been tested. A failed update and a failed drive are different events, so keep a recovery plan for each.

The video’s prices and coupon belong to its recording context. Its examples also compare different memory capacities and different bundles. For a purchase decision, rebuild the comparison using the same requirements and current checkout information. Include the power supply, cooling, enclosure, storage adapter, drive, cables, shipping, and tax as applicable. Reuse existing parts only when they meet the new system’s requirements.

Check the setup before making it a household dependency

  • Run the applications you actually need together and observe their resource use.
  • Verify that a client can still find the server after a normal restart.
  • Confirm that application data is on the intended persistent storage.
  • Restore a small test dataset from the backup destination.
  • Retain console access and a record of the working network settings.

The decision is strongest when it is specific. Keep Raspberry Pi’s peripheral advantages when the project needs them. Choose the server-oriented connections when they remove real adapters or compatibility problems from your own workload. Treat both as systems to maintain, with a workload that can be explained and tested.

FAQ

Can I upgrade ZimaBoard 2 from 8GB to 16GB later?

Its LPDDR5 memory is soldered. Choose the required configuration when purchasing; SATA and PCIe expansion do not create a removable RAM slot.

Does JBlanked run a 27B model on the Intel N150 alone?

No. He explicitly attributes the demanding model computation to an RTX 5070 Ti. The small server provides supporting and front-end services in the described workflow.

Does an Arm Docker image always need emulation?

No. A supported Arm image can run natively on a compatible Arm host. Inspect the image’s platforms and dependencies before deciding which architecture makes deployment easier.

Should a working Raspberry Pi project be migrated immediately?

Only a concrete requirement justifies the work. If its peripherals, performance, and maintenance already fit the job, the video does not establish a need to replace it.

Do two Ethernet ports automatically create separate networks?

No. Isolation depends on the software configuration and the actual traffic path. Two physical connections alone do not establish routing or firewall policy.

Does an SSD remove the need for backups?

No. Keep independent copies of important data and test a restore. An easier drive connection does not protect against deletion, corruption, or hardware failure.

Conclusion

JBlanked’s comparison is most useful as a record of how his lab changed: a development board became a collection of services he wanted available every day. ZimaBoard 2’s x86 platform, SATA connections, and dual Ethernet fit that direction. Raspberry Pi 5’s GPIO and camera ecosystem continue to serve a different set of practical requirements.

Choose the platform around the workload, count the complete system, and keep the limits visible. ZimaBoard 2 has fixed RAM, and the demanding AI work in this story depends on separate GPU acceleration. Watch JBlanked’s full review for his experience, then validate your own software, storage, and recovery path before moving a daily service.

Zima Campaign Hub

More to Read

Get More Builds Like This

Stay in the Loop

Get updates from Zima - new products, exclusive deals, and real builds from the community.

Stay in the Loop preferences

We respect your inbox. Unsubscribe anytime.