hardware
- The 3090 unlock is real. Where it lands is the catch
gpu_unlock lifted my RTX 3090's BF16 GEMM from 74.9 to 128 TFLOPS. My daily 27B in llama.cpp got 0.7% faster at short context. Here's where the rest went.
- The pin template was for a different plug
Moving a Bluetooth-paired Tuya energy plug to OpenBeken without opening it, and the borrowed pin template that looked right and wasn't.
- Installing Fedora on a computer that hadn't shipped yet
Fedora 45 was installed, Secure-Boot-tested and debugged in a container and QEMU before the A9 Mega first ran Linux. Then the hardware brought its own bugs.
- Salvaging a dying disk by partitioning around the wound
A drive SMART calls dead had all 58 bad blocks in the first 5% of the platter. Read the map, partition past the wound, reclaim 6.7 TiB.
- Stop chasing a sensor that doesn't exist
The LSI SAS2008 HBA cooking itself has no temperature sensor. No tool, driver, or root binary conjures a number that was never wired to the die.
- All four drives at once, and why that was the good news
Every HDD on the storage box vanished at once — a quadruple drive death, it looked like. It was two mundane faults stacked together, and the data was fine.
- FiiO X1 FTL reversed: the map was in the OOB bytes I never dumped
The FiiO X1's flash translation layer rebuilds its block map from the spare bytes of each block's first page. My NAND dump kept 2,048 bytes a page.
- FiiO X1 PCM5142 DSP blocks: the flat head I believed
Three FiiO X1 firmware blocks the DAC loads per stream looked like passthrough filters, until the decoder read all 2,708 bytes of each one.
- Carving a SQLite database from a FiiO X1's raw NAND
One SQLite header in a FiiO X1's 128 MB NAND dump: a malformed carve, sqlite3 .recover, six of my own tracks and a previous owner's ten favourites.
- FiiO X1 EQ table: seven register bytes my generator never read
Decoding the FiiO X1's PCM5142 EQ coefficient table: the format was fine, the register bytes were not, and my first proof of the layout was wrong too.