Skip to content
← CUDA, Triton & GPU Performance / 19
hardNewNVIDIAOpenAIAnthropic

A kernel PR claims 12x over eager PyTorch on the SwiGLU activation and passes the tests. How do you decide whether it is real?

The bytes say the best possible fusion here is 1.67x, so 12x means something else is being measured. The answer finds it in the order: a kernel that never writes its output, a test that runs the reference first, and an allocator that hands back the answer.

Updated Oct 2026 · Worked answers on inference mechanisms, with calculations, implementation details, and follow-up questions.

The bytes say the best possible fusion here is 1.67x, so 12x means something else is being measured. The answer finds it in the order: a kernel that never writes its output, a test that runs the reference first, and an allocator that hands back the answer.

10 answers per topic instead of 5, plus saved progress and bookmarks · costs nothingor unlock all 213 remaining answers · $25

Every worked answer and derivation in all 12 topics, every concept page, the full practice-test bank, and every code drill’s edge-case tests and reference solution.

6 months · One payment · No auto-renewal

UP NEXT ON YOUR JOURNEY
Popular right nowWhy is batch-1 decode on a 70B model bandwidth-bound, and what is its latency floor on two H100s?Popular right nowDecode attention intensity is fixed at the query-to-KV head ratio. Derive that, then tell me what MLA does to it.Popular right nowAccount for every gigabyte on a GPU running vLLM: weights, KV pool, activations, CUDA graphs and NCCL buffers.
DISCUSSION · 0

Nobody has commented on this answer. Add a correction or another way to work it.