[zerof1l]

The article gives no mention of what exactly was done to achieve the speedup and whether or not the kernel is still able to perform the same function as before.

I’m doubtful this is a meaningful result. Kernel contains a lot of legacy code and generalizations to support different hardware etc.; removing that would result in a speedup. Next are all the mitigations for hardware vulnerabilities and attacks. If removed would give a nice speedup as well at the cost of security. And then finally, just specializing the Kernel in whatever the benchmark is measuring, making it useless as a general piece of software would also make it fast.

[miloignis]

The article is talking about "Kernel" as in a low level piece of code to compute math, in this case extended attention for running LLMs on a GPU or accelerator, not as in the Linux Kernel.