ResourcesWorking references
The sources I check
before accepting the claim.
This is a working directory, not a general recommendation list. I include a source when it helps define a test, reproduce the work or interpret a result.
On this site
Practical references and the complete field-note archive.
Useful Linux commands
Bounded FIO, SMART and legacy SSH examples, with the safety checks that matter before they are run.
↗ ArchiveField notes
The complete archive in publication order, including investigations that do not yet belong to a larger series.
↗ StorageStorage field notes
A curated starting point for NVMe, NVMe-oF, SPDK and S3 work, including the operational boundaries around each test.
↗ GPU / Series 01AMD Strix Halo GPU & local AI
The ongoing investigation: establish the baseline, compare runtimes, check model fit, verify correctness and then run for long enough to expose stability problems.
↗ CodeGitHub
Public code and data used in the work, plus the implementation, release and source material behind it.
↗Storage & systems
Primary sources for the storage work.
NVM Express
I use the specifications to check namespace, format and NVMe-over-Fabrics behaviour rather than relying on a summary.
↗ FrameworkSPDK
The primary source for release-specific APIs, configuration and user-space storage behaviour.
↗ StorageCeph
Current Ceph documentation, checked against the version and behaviour of the system being discussed.
↗ HPC storageDAOS
A useful comparison when the workload and architecture move beyond conventional block or file storage.
↗ BenchmarkingFIO
The workload generator and reference used to define exactly what an I/O number represents.
↗ CommunityIO500
Community results and methodology for comparing high-performance storage without losing the test context.
↗Local AI
Start with the runtime and silicon documentation, then measure.
AMD ROCm docs
I check installation, compatibility and optimisation guidance against the exact ROCm release used in a test.
↗ Runtimellama.cpp
Build instructions, release history and source for the Vulkan and ROCm backends used in local inference tests.
↗ ModelsHugging Face Hub
Model cards, weights and metadata used to identify the exact model rather than just the family name.
↗ HardwareRyzen AI Max+ 395
AMD’s specifications and current driver route for the Strix Halo system used in the series.
↗