The benchmark that matters
Compiling the Linux kernel is a heavy, repetitive job. It runs dozens of cores in parallel, saturates memory bandwidth, and keeps every part of the processor busy. It's not a synthetic microbenchmark — it's the exact kind of task developers run every day. That's why chipmakers and system builders use it as a baseline for general-purpose server performance.
For a server buyer, compile speed maps directly to money. Shorter build times mean faster continuous integration pipelines, fewer machines needed for the job, and less time waiting on software builds. It's a workload that shows up in real costs, not just on a spec sheet.
Why Hot Chips is the right stage
Hot Chips is a conference where chip designers share architecture details, not marketing fluff. The audience is engineers and fellow architects, and claims get challenged in the room. Presenting a benchmark there means Nvidia is willing to put its result in front of people who will poke at it.
The choice of opponent matters too. AMD's EPYC 9655P is the current-generation server chip aimed squarely at cloud and enterprise buyers. Going directly against that model makes the comparison unambiguous. It's not a vague slide claim — it's a direct test against the chip AMD sells to the same customers Nvidia wants.
The EPYC 9655P's position
The 9655P is one of AMD's strongest server parts. It packs a large core count and a long list
