CPU Design · All levels
L2/L3 Hierarchy Design
Cache & Memory Hierarchy: Private and shared cache layers must balance locality, coherence traffic, and QoS isolation so one core's bursty stream does not collapse latency for neighboring cores.
What this topic teaches
L2/L3 Hierarchy Design turns CPU design theory into actionable review decisions. Private and shared cache layers must balance locality, coherence traffic, and QoS isolation so one core's bursty stream does not collapse latency for neighboring cores. The target is evidence-backed closure, not opinion-driven tuning.
Senior-engineer framing question
When LLC hit rate, inter-core interference index, and effective memory latency shifts, can you prove first failing stage, dominant mechanism, accountable owner, and release-safe mitigation?
CPU PIPELINE VIEW - L2/L3 Hierarchy Design
fetch -> decode -> rename -> dispatch -> execute -> retire
| | | | | |
icache uop flow map table queueing FU ports ROB commit
steady-state goal:
keep every stage supplied without bubbles or flush storms
Focus: connect metric movement to the first stage loss
Metric tracked: LLC hit rate, inter-core interference index, and effective memory latencyArchitecture visuals
Draw the mechanism before changing knobs. These visuals are optimized for design reviews and interview whiteboards.
Shared LLC pressure map
CPU CACHE + MEMORY HIERARCHY - L2/L3 Hierarchy Design
[ L1I ] [ L1D ]
32-64KB, ~4 cycles
\ /
[ L2 ]
512KB-2MB, ~12 cycles
|
[ L3 ]
shared LLC, 30-60 cycles
|
[ DDR/HBM memory ]
80-150ns effective
Optimization lens: show cross-core interference and eviction pressure in L2/L3Hierarchy contention root-cause tree
ROOT-CAUSE TREE - L2/L3 Hierarchy Design
LLC hit rate, inter-core interference index, and effective memory latency regressed
|
reproducible on fixed seed?
/ \
no yes
| |
env/tool drift first failing stage?
/ | \
front-end execute memory/system
| | |
fetch/decode port/ROB cache/TLB/NoC
Stop at first confirmed mechanism, then patch with owner accountability.Out-of-order control map
OOO CORE BLOCK DIAGRAM - L2/L3 Hierarchy Design
decode -> rename -> dispatch -> reservation stations -> execute units
| | |
free-list / map table wakeup-select writeback
\ | /
+-------- reorder buffer / retire ---------+
Focus: rename to retire dataflowMemory hierarchy map
CPU CACHE + MEMORY HIERARCHY - L2/L3 Hierarchy Design
[ L1I ] [ L1D ]
32-64KB, ~4 cycles
\ /
[ L2 ]
512KB-2MB, ~12 cycles
|
[ L3 ]
shared LLC, 30-60 cycles
|
[ DDR/HBM memory ]
80-150ns effective
Optimization lens: latency vs capacity tradeoffSpeculation lens
BRANCH PREDICTOR VIEW - L2/L3 Hierarchy Design
fetch PC -> BTB lookup -> direction predictor -> target select -> fetch redirect
| | |
BTB miss cost confidence RAS / indirect path
branch resolves in execute:
correct prediction -> pipeline keeps flowing
mispredict -> flush + restart + refill
Focus: minimize wrong-path workOwnership layers
CPU OWNERSHIP LAYERS - L2/L3 Hierarchy Design
artifact area owner
---------------- ----------------------------
architecture memory hierarchy architect
RTL/microarch NoC owner
software/tools SoC integration lead
Rule: every regressed metric must map to an explicit owner and closure artifact.Evidence required
Primary metric: LLC hit rate, inter-core interference index, and effective memory latency.
Primary artifact: LLC residency report, QoS contention matrix, and latency stack chart.
Owners to include: memory hierarchy architect, NoC owner, SoC integration lead.
One reproducible failing workload and one stable comparator run.
One run with fully locked environment metadata for causal comparison.
Compute-memory limit lens
CPU ROOFLINE - L2/L3 Hierarchy Design
performance
^
| compute roof
| /
| /
|--------------/---------------- memory roof
+----------------------------------------------> arithmetic intensity
memory-bound compute-bound
Interpretation: separate compute and memory limitsKey takeaways
Classify stage loss before proposing fixes.
Use artifacts to separate mechanism from symptoms.
Close with owner accountability and rollback criteria.
Common pitfalls
Using average IPC alone while ignoring tail behavior.
Comparing traces across mismatched binaries or thermal states.
Calling closure without workload-level validation.
CPU deep dive
Memory hierarchy closure needs cache, TLB, and prefetch policy to be tuned together for real latency tails.
Concept diagram
MEMORY + TRANSLATION STACK
L1I/L1D -> L2 -> LLC -> DRAM
| | |
ITLB/DTLB hierarchy + page walkersMetric graph
LATENCY TAIL CONTRIBUTORS
cache miss chains █████
translation misses ████
coherence interference ███Reports and artifacts
L1/L2/LLC latency stack
TLB walk profile
prefetch usefulness report
memory tail percentile dashboard
Mini case study
Prefetch aggressiveness improved average misses but worsened p99 latency by polluting LLC and stressing page walkers.
Debug branches
Tag misses by source: capacity, conflict, translation, or coherence
Track TLB shootdowns and page-size behavior with workload phases
Evaluate prefetch policy on tail latency, not just average CPI
Senior review question
Ask: which CPI/latency evidence proves this topic is truly closed beyond synthetic benchmarks?
Key takeaways
Always connect microarchitectural counter changes to product workload outcomes.
Lock binary, compiler, firmware, and thermal metadata before comparing CPU traces.
Common pitfalls
Treating average IPC as sufficient proof while ignoring latency tails and outliers.
Applying predictor or prefetch tweaks without first-failing-stage attribution.
Declaring closure without reproducible perf, correctness, and power gates.