Computer Architecture · All levels

Dataflow Architecture Choices — Interview Drills

Interview Drills for Dataflow Architecture Choices (Accelerator Architectures).

Interview drills

Practice aloud for Accelerator Architectures → Dataflow Architecture Choices. Use METRIC → HYPOTHESIS → FIX → REGRESSION.

Explain Dataflow Architecture Choices to a hiring manager in 60 seconds.

diagram
[INT][ARCH][TOPIC]

Q: Explain Dataflow Architecture Choices to a hiring manager in 60 seconds.

A:
Design compute-data movement choreography (weight-stationary, output-stationary, streaming, tiled) for stable efficiency across workload diversity.

FOLLOW-UP TRAP: Tool list without mechanism.

What report proves Dataflow Architecture Choices is done?

diagram
[INT][ARCH][TOPIC]

Q: What report proves Dataflow Architecture Choices is done?

A:
Name the primary architecture metric report and acceptance criteria.

FOLLOW-UP TRAP: No metric — only 'looks good'.

What breaks if Dataflow Architecture Choices is done poorly?

diagram
[INT][ARCH][TOPIC]

Q: What breaks if Dataflow Architecture Choices is done poorly?

A:
Downstream performance, power, verification, software, or implementation failures.

FOLLOW-UP TRAP: Only mentions runtime, not silicon risk.

10+ year interview answer bar

At senior/principal level, the interviewer is testing ownership judgment more than vocabulary. Answer Dataflow Architecture Choices through failure mode, evidence, tradeoff, and release decision.

You inherit a late-stage Dataflow Architecture Choices failure one week before release. What do you do in the first hour?

diagram
[INT][ARCH][STAFF]

Q: You inherit a late-stage Dataflow Architecture Choices failure one week before release. What do you do in the first hour?

A:
Freeze the workload/model/RTL tag, name the failing metric (Accelerator Architectures closure dashboard), confirm counter setup, cluster the issue by structure or workload phase, assign the first experiment, and publish a validation/owner plan before changing architecture.

FOLLOW-UP TRAP: Jumping directly to a larger cache, wider pipe, or extra NoC link without preserving evidence.

When would you stop trying to improve Dataflow Architecture Choices and escalate?

diagram
[INT][ARCH][STAFF]

Q: When would you stop trying to improve Dataflow Architecture Choices and escalate?

A:
Escalate when the remaining risk crosses ownership boundaries, consumes shared margin, changes signed-off assumptions, or threatens Memory subsystem sizing and runtime policy complexity.. Bring exact report lines and options, not vague concern.

FOLLOW-UP TRAP: Escalating without data or continuing alone after a cross-team decision is needed.

Whiteboard diagram to draw

diagram
VISUAL MODEL — Accelerator Architectures / Dataflow Architecture Choices

        workload / trace
              │
              ▼
   metric symptom (IPC, MPKI, bandwidth, latency, stalls)
              │
              ▼
     likely microarchitectural mechanism
              │
      ┌───────┼────────┐
      ▼       ▼        ▼
  pipeline  memory    fabric/coherency
  stalls    misses    queues / ordering
      │       │        │
      └───────┼────────┘
              ▼
        bounded design change
              │
              ▼
   validation workload + PPA regression

Architecture deep dive

Accelerators win on locality and bandwidth contracts, not peak OPS alone.

Concept diagram

diagram
ACCELERATOR DATAFLOW

Host CPU ── commands ──► Queue / scheduler
   ▲                         │
   │ completion              ▼
Coherent memory ◄── DMA ── Local SRAM ──► Compute array
                         ▲       │
                         └ tiles ┘

Peak TOPS matters only when data reaches the array at the needed rate.

Metric graph

diagram
UTILIZATION BREAKDOWN

compute active   ██████████████████  58%
DMA wait         ██████████          31%
host sync        █████               15%
cache/coherency  ████                12%
idle bubbles     ███████             22%

Low utilization is usually a system integration problem.

Metrics and artifacts

  • accelerator utilization

  • DMA bandwidth

  • kernel launch overhead

  • coherency invalidation rate

Mini case study

NPU met TOPs target but end-to-end inference slow — DMA and weight fetch dominated. Architecture added on-chip SRAM tile and double-buffering.

Debug branches

  • If util low, check launch overhead and host sync first.

  • If BW high, examine weight layout and sparsity support.

Senior review question

Ask: what single metric would prove this concept is working or failing on your workload?

Key takeaways

  • Connect every architecture claim to a workload and measurable metric.

  • State verification and PPA impact before proposing design changes.

Common pitfalls

  • Feature-driven design without MPKI/IPC/bandwidth evidence.

  • Ignoring coherency and NoC traffic in cache and accelerator sizing.

Study notes

Re-read this topic with one concrete workload.