Putting Raspberry Pi Compute Module onto a blade for a low-cost and energy-efficient alternative to a rack-based server. In the latest issue of The MagPi (pg72) Lucy Hattersley weighs up the benefits.
Compute Blade (£54/$65) is a fascinating product developed by Ivan Kuleshov at Uptime Lab that puts a Compute Module in a rack-mountable carrier board to create a high-density server for industrial or home use.
Blade servers are the current type of format popular amongst server systems. A traditional rack-mounted server houses a single computer inside a wide chassis (like a tray). Thanks to its small footprint, you can fit up to 20 Compute Blade devices into a rack space.
Compute Blade puts all of the components (storage, RAM, and computer) on a single PCB. In the case of Compute Blade, it uses a Raspberry Pi Compute Module 4 as the main computer.
You can also dedicate each blade to a specific task: web serving, file serving, email support, virtualisation, and so on. So taking one task offline for maintenance leaves the rest up and running.
There are security advantages to separating services via blades, as it reduces the surface area for a hacker to gain access to the system. To this end, Compute Blade can feature a TPM 2.0 chip (Trusted Platform Module) to store cryptographic keys and encrypt your data.
Above all though, blade servers are more energy-efficient than traditional servers. Energy costs are always important for server infrastructure, and this has only become more of an issue recently with higher energy costs. The Arm chips found on Raspberry Pi computers are more energy-efficient than their rivals.
There’s a lot here that’s interesting and we’re delighted that Ivan Kuleshov raised over a million euros on Kickstarter to bring the project to life. Ivan sent us two Compute Blade Dev boards, a fan, and a four-board enclosure for testing.
Check it Blade
Compute Blade is a single PCB that can house a Raspberry Pi Compute Module. This rack-mountable PCB measures 42.5 mm × 255 mm × 17.5 mm with Raspberry Pi Compute Module 4 connectors and PoE (Power over Ethernet), so you can power and transfer data from a single Ethernet cable.
We used a TP-Link TL-PoE150S for setup testing, and fitted the two Compute Blade boards to the main Raspberry Pi rack server. Alongside Ethernet sit a USB-C and USB-A port (with a switch to move between the two), an HDMI port, UART, TPM 2.0, microSD card port, M.2 M-key (for an NVMe SSD up to 22110), and a fan unit connector. The Dev unit we tested features an anodized heatsink secured with T7 hex screws. (Stress testing remained below 65°C).
There are LEDs for SSD, Power, and Activity (these can be disabled in software) and a programmable Edge Button on the front panel. This button acts as a pull tab and can be clicked while the blade is mounted in a chassis. You could use it to perform a restart, for example, or custom script (it is connected to GPIO 20).
Three models are available: Basic, TPM, and Dev (which we are testing here). The Compute Blade Documentation website has more information on the varying specifications. The Kickstarter campaign offered 4-node cluster cases, 10-inch rack clusters, and 19-inch rack clusters which you can 3D print for your rack.
How sharp is it?
It’s a wonderfully designed piece of kit, and we believe it’d be a worthy addition to any server space. The question for us is what would a non-server administrator do with a rack of Compute Blades? It would make a great cluster computer for testing and experimentation, running your own small server enables you to deploy personal cloud services (much more privacy), or it could make a great Pi-hole ad-blocker addition to your network. You could also deploy a Raspberry Pi in the home on your network for any available coding or scripting tasks you want. We set ours up as a custom GPT chatbot that we can log into any time without depending on the OpenAI website.
Compute Blade more than doubled its Kickstarter funding, and the end result is pretty well-built from our testing, and demand is certainly out there. If you are looking to build an Arm-based server, this is our recommendation and we think it’s a neat product to have on your home network.
Compute Blade is a wonderful piece of design that combines the low-cost availability of Compute Module with storage, PoE networking, and a blade format. Perfect for building your Arm cluster or server solution.
Input/Output: Gigabit Ethernet, USB-A, USB-C, HDMI, UART, Compute blade headers, microSD card slot, hardware switchable Wi-Fi, BT, and EEPROM write-protection
Support: Raspberry Pi Compute Module 4, NVMe SSD up to 22100, TPM 2.0
Power: Power over Ethernet IEEE 802.3at (PoE+) up to 30 W (normal operation 2–8 W)