← Skill graph
GPU execution model
Part of LLM inference cost and latency optimisation
Needs first
Helps first
Nothing recommended beyond what is required.
Unlocks
Where it is assessed
14 assessed items, 0 done.
- AIE-207 M1.4 Measure the hierarchy yourself · tier 1
- AIE-207 M1.5 Attention's IO cost, from your own numbers · tier 2
- AIE-207 M1.6 Retention: the memory hierarchy and its costs · tier 1
- AIE-207 M2.2 Four operations, predicted then measured · tier 1
- AIE-207 M2.3 A roofline for the hardware you actually have · tier 2
- AIE-207 M2.5 Retention: arithmetic intensity and the roofline · tier 1
- AIE-207 M3.3 Attribute every millisecond of a training step · tier 1
- AIE-207 M3.4 One change, predicted before it is made · tier 2
- AIE-207 M3.5 Retention: profiling and attribution · tier 1
- AIE-207 M5.1 The systems view of your own inference path · tier 2
- AIE-207 M5.2 Publish the performance chapter · tier 3
- AIE-207 M5.3 Where the systems view changes an architecture choice · tier 2
- AIE-207 M5.4 Oral defense · tier 4
- AIE-207 M5.6 Retention: systems reasoning applied to architecture · tier 1