Raspberry Pi 5 bootloader has blocked RAM-swapped boards since September 2024

For roughly two years, Raspberry Pi 5 boards with unofficially upgraded memory have failed to boot on current firmware, and almost nobody knew why. According to a dev.to writeup, bootloader EEPROM releases dated 23 September 2024 and later decline to initialise SDRAM unless it matches the memory configuration the board was provisioned with at the factory. The restriction only surfaced publicly on 21 September 2026, when Jeff Geerling documented it and noted that the update responsible shipped in late 2024.

What the lockout does

The Pi 5 is sold in 1 GB, 2 GB, 4 GB, 8 GB and 16 GB variants, all LPDDR4X-4267. Enthusiasts had taken to replacing the memory package on cheaper boards to gain headroom for containers, self-hosted services and local AI models. On any bootloader from 23 September 2024 onward, a board with a swapped chip fails SDRAM initialisation and never boots; the visible symptom is the green ACT LED blinking eight times in a repeating loop, the board's error code for a memory failure. Because the check runs below the operating system, no Linux-side configuration change can work around it.

A change the changelog never mentioned

The release notes give no hint of the restriction. The changelog entry for the 2024-09-23 EEPROM lists only SDRAM performance tuning, a new SDRAM_BANKLOW parameter and manufacturing test updates. The clearest primary evidence is rpi-eeprom issue 761, filed on 15 October 2025 by a user whose Pi 5, sold as an 8 GB unit, would not boot on any release from 2024-09-23 onward but worked fine on 2024-09-10. Raspberry Pi engineers requested the board's OTP dump, concluded from the provisioning data that it had been modified after manufacture, and said such a board is stuck on a pre-2712D0 firmware release running old test SDRAM code. A company engineer also confirmed the policy on the Raspberry Pi forums. Worth noting here: this is Raspberry Pi Ltd, the trading company, not the Raspberry Pi Foundation.

Why the check exists

The dev.to post lays out several motives that are not mutually exclusive. Unofficial swaps can cause instability that lands on Raspberry Pi's own support desk. Memory chips are binned and validated against specific timing envelopes. Higher-RAM SKUs are a meaningful part of the business model. And the bootloader genuinely needs to know which part is fitted: the changelog includes per-variant tuning such as improved SDRAM refresh timings for the 16 GB model, which is why even a same-capacity chip pulled from another Pi 5 can be rejected — the check is not purely about capacity. Raspberry Pi's stated concern is counterfeiting: the company's position is that resellers were soldering larger chips of unknown origin onto cheap boards and selling them as factory 8 GB units.

What modded-board owners should do

A RAM-swapped Pi 5 that still boots is necessarily running the 2024-09-10 bootloader or older, meaning it is two years behind on fixes, and no release channel offers both current firmware and tolerance for the mod. The rpi-eeprom-update command shows the installed and staged EEPROM versions; the release channel lives in the FIRMWARE_RELEASE_STATUS variable in /etc/default/rpi-eeprom-update, where default and latest are the current options and older names such as stable and beta are historical. Downgrades are also getting harder: the 2026-09-23 bootloader added a check that blocks EEPROM writes below the minimum manufacture version recorded in OTP unless bootloader_allow_mfg_downgrade=1 is set in config.txt. For anyone who actually needs the memory, the realistic paths are the official 16 GB board, NVMe-backed swap on an M.2 HAT, or moving the workload to a small x86 machine.

Why it matters

This is a firmware-enforced hardware restriction on a platform whose community was built on exactly this kind of tinkering, and it was applied silently — the changelog still does not record it. Anyone running a modified Pi 5 is sitting on stale firmware with a boot failure scheduled for their next update, and secondhand buyers have no easy way to tell whether a board advertised as 8 GB is a modified unit that will refuse future releases. It is also a broader reminder that on embedded platforms, changelogs are not a complete record of behaviour changes; sometimes the real documentation is an issue tracker and an OTP dump.