L2 & Last-Level Cache: Review Checklist

Review checklist

Review Checklist for L2 & Last-Level Cache centers on L2 hit ratio, eviction pressure, and inter-SM coherence traffic. The objective is to connect profiler evidence to root-cause mechanism and release-safe action.

• Workload scope and target KPI are explicitly documented.

• Profiler + counter evidence is reproducible with revision tags.

• Bottleneck classification is proved with mechanism-level traces.

• Mitigation includes owner, blast radius, and rollback criteria.

• End-to-end benchmark matrix confirms closure.

• Owners signed: memory system lead, GPU architect, NoC owner.

Signoff ownership

GPU OWNERSHIP LAYERS — L2 & Last-Level Cache

artifact area     owner
----------------  ----------------------------
architecture    memory system lead
RTL/microarch   GPU architect
software/tools  NoC owner

Rule: each metric needs a named owner before signoff.

GPU deep dive

Bandwidth wins come from coalescing and locality discipline, not peak-memory specs alone.

Concept diagram

MEMORY HIERARCHY

register -> shared/L1 -> L2/LLC -> HBM/GDDR
access pattern quality decides latency

Metric graph

BANDWIDTH UTILIZATION

requested BW  ███████████
effective BW  ████████
wasted BW     ███

Reports and artifacts

• L1/L2 hit-rate report

• HBM efficiency counters

• coalescing transaction log

• shared-memory bank audit

Mini case study

Stencil kernel sat at 43% of peak HBM due to uncoalesced loads; layout rewrite recovered 1.6x effective bandwidth.

Debug branches

• Check transactions per request at warp granularity

• Classify cache-thrash versus true DRAM saturation

• Audit shared-memory bank conflicts before algorithm rewrites

Senior review question

Ask: which metric and benchmark pairing proves this topic is truly closed in production context?

Key takeaways

• Always pair micro-kernel metrics with end-to-end workload impact.

• Lock toolchain, driver, and launch metadata before comparing performance results.

Common pitfalls

• Optimizing occupancy without checking memory-system saturation.

• Comparing profiler captures from different driver or compiler builds.

• Declaring wins without reproducible accuracy and performance gates.

Review checklist explanation

A checklist is not bureaucracy here; it is how GPU teams avoid confusing local wins with product wins. Every signoff item should protect against a known class of false confidence.

For L2 & Last-Level Cache, the minimum checklist is workload scope, L2 hit ratio, eviction pressure, and inter-SM coherence traffic, artifact evidence (L2 traffic breakdown, eviction reason chart, and bandwidth pressure summary), bottleneck classification, owner, rollback path, and full matrix validation.

If the change affects architecture or RTL, include correctness and PPA evidence. If it affects compiler/runtime policy, include compatibility and deployment evidence. If it affects physical design, include timing, IR, thermal, and observability evidence.