Output-Stationary Deep Dive: Interview Drills

Interview drills

Interview Drills for Output-Stationary Deep Dive is anchored on Partial-sum spill frequency and accumulator energy per output element at target precision modes.. Convert measurements into mechanism-backed decisions with clear owner accountability.

PROMPT
You observe regression in Partial-sum spill frequency and accumulator energy per output element at target precision modes. for Output-Stationary Deep Dive. Explain root cause and release decision.

STRONG ANSWER
1. Defines workload and first failing mechanism.
2. Explains mechanism: Output-stationary mapping holds partial sums close to compute until reduction completion, minimizing repeated writeback and reload of intermediate values. Activations and weights traverse the array while each PE accumulates into local state, then emits final outputs when tiles close. This can materially improve efficiency for reduction-heavy operators, but only if accumulator sizing, saturation policy, and update timing align with workload statistics. If local accumulation capacity is undersized, spill traffic to shared buffers grows quickly and can dominate both latency and power.
3. Requests proving artifact: Accumulator sizing and spill-risk analysis with precision policy and writeback thresholds.
4. Proposes bounded fix + owner + rollback-safe validation.

WEAK ANSWER
Gives generic optimization ideas without mechanism proof or ownership.

AI accelerator deep dive

Dataflow choices are durable architecture decisions that shape memory and scheduling cost.

Concept diagram

DATAFLOW DECISION

input/output/weight stationary
  -> locality pattern
  -> movement cost
  -> throughput and power

Metric graph

DATAFLOW COST MIX

activation traffic   ███████
weight traffic       █████
partial-sum traffic  ██████

Metrics and artifacts to collect

• reuse factor map

• buffer pressure profile

• NoC traffic mix

• shape sensitivity analysis

Mini case study

A dataflow that won for convolution lost on attention-heavy batches due to activation movement pressure.

Debug branches

• Segment by model family

• Compare reuse vs movement

• Re-check mapping assumptions under batch variance

Senior review question

Ask: which first-principles bottleneck class explains the symptom, and what artifact proves it reproducibly?

Key takeaways

• Tie every accelerator claim to a reproducible workload slice and one primary metric trend.

• Prefer bounded fixes with clear owner and rollback boundary over broad tuning bundles.

Common pitfalls

• Optimizing synthetic kernels without production-shape validation.

• Reading average latency while ignoring p95 and p99 behavior.

• Declaring sparse or precision wins without fallback and quality evidence.

Interview answer expansion

A strong answer on Output-Stationary Deep Dive names the workload symptom, explains mechanism (Output-stationary mapping holds partial sums close to compute until reduction completion, minimizing repeated writeback and reload of intermediate values. Activations and weights traverse the array while each PE accumulates into local state, then emits final outputs when tiles close. This can materially improve efficiency for reduction-heavy operators, but only if accumulator sizing, saturation policy, and update timing align with workload statistics. If local accumulation capacity is undersized, spill traffic to shared buffers grows quickly and can dominate both latency and power.), and proposes one measurable validation plan.

Then it identifies owner and fallback action if the proposed fix under-delivers.

The goal is practical engineering reasoning, not keyword listing.