A network architect is designing a data center fabric to support distributed AI training. Which pair of characteristics is most typical of AI training workloads and therefore drives the network design?

Low bandwidth but extremely high reliability

High north-south bandwidth and jitter-tolerant flows

High east-west bandwidth and low, predictable latency

Bursty, short-lived flows with minimal buffering

A cloud provider is designing a network for a multi-tenant GPU-as-a-Service platform. Each tenant runs distributed training jobs that generate bursty, synchronized east-west traffic. The network team must ensure that one tenant's traffic does not degrade another tenant's performance. The fabric uses a spine-and-leaf topology with deep buffers. Which combination of features is most effective for this multi-tenant environment?

Enable priority flow control (PFC) on all fabric links and rely on the deep buffers to absorb bursts

Implement per-tenant virtual routing and forwarding (VRF) with QoS policies that prioritize control traffic over data traffic

Deploy a separate physical fabric for each tenant to guarantee performance isolation

Use a centralized SDN controller to schedule flows and enforce per-tenant bandwidth limits at the leaf switches

What is a key reason to segment AI workloads from general-purpose corporate traffic?

To reduce the risk of lateral movement if a general-purpose system is compromised

To simplify the routing table by using a single VLAN

To allow unrestricted access between all systems

To increase the available bandwidth for AI workloads

Which network optimization technique is specifically designed to handle the bursty, synchronized traffic patterns typical of AI training?

Disabling flow control to avoid head-of-line blocking

Using a single queue for all traffic to simplify scheduling

Increasing the maximum transmission unit (MTU) to 9000 bytes

Explicit congestion notification (ECN) and priority flow control (PFC)

A national research lab is designing a network for a new LLM training cluster. The cluster will run training jobs that last for weeks, and a single job uses all 512 GPUs. The lab has a strict budget and cannot afford a fully redundant fabric. The network team must choose a design that minimizes the impact of a single switch failure on the training job. Which approach is the most cost-effective?

Deploy a single spine-and-leaf fabric with no redundancy and accept that any switch failure will restart the training job

Deploy a two-tier spine-and-leaf fabric with N+1 redundancy for spine switches and rely on the training framework's checkpointing to recover from failures

Use a fat-tree topology with multiple equal-cost paths and rely on ECMP to reroute around a failed switch

Deploy a fully redundant three-tier fabric with dual spines and dual leaves to ensure no single point of failure

CiscoCertified Design Expert (CCDE)

Domain 3Objective 2

3.2 AI Network Design Use Cases (such as Machine Learning, Large Language Models, and Pattern Recognition) 400-007 Practice Questions (Page 1)

Part of the 3.0 Network Design domain, which accounts for 30% of the 400-007 exam. Cisco does not publish an official question count, but from its 120-minute exam (~50–80 total, ~15–24 in this domain), expect 8–12 from this objective — we provide 27 practice questions to prepare you well beyond it. (estimate)

27questions here

6free pages

8concepts

30%of the exam

Questions 1–5

1foundation · easy
A network architect is designing a data center fabric to support distributed AI training. Which pair of characteristics is most typical of AI training workloads and therefore drives the network design?
Select an answer first
2expert · hard
A cloud provider is designing a network for a multi-tenant GPU-as-a-Service platform. Each tenant runs distributed training jobs that generate bursty, synchronized east-west traffic. The network team must ensure that one tenant's traffic does not degrade another tenant's performance. The fabric uses a spine-and-leaf topology with deep buffers. Which combination of features is most effective for this multi-tenant environment?
Select an answer first
3foundation · easy
What is a key reason to segment AI workloads from general-purpose corporate traffic?
Select an answer first
4foundation · easy
Which network optimization technique is specifically designed to handle the bursty, synchronized traffic patterns typical of AI training?
Select an answer first
5expert · hard
A national research lab is designing a network for a new LLM training cluster. The cluster will run training jobs that last for weeks, and a single job uses all 512 GPUs. The lab has a strict budget and cannot afford a fully redundant fabric. The network team must choose a design that minimizes the impact of a single switch failure on the training job. Which approach is the most cost-effective?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

Next 5 questions →Back to 400-007

Free Basic Practice is a study aid with revealable answers — not a scored exam. Examers.io is independent and not affiliated with or endorsed by Cisco. “400-007” is a trademark of its owner, used for identification only.

3.2 AI Network Design Use Cases (such as Machine Learning, Large Language Models, and Pattern Recognition) 400-007 Practice Questions (Page 1)

Questions 1–5

A network architect is designing a data center fabric to support distributed AI training. Which pair of characteristics is most typical of AI training workloads and therefore drives the network design?

What is a key reason to segment AI workloads from general-purpose corporate traffic?

Which network optimization technique is specifically designed to handle the bursty, synchronized traffic patterns typical of AI training?

Keep studying — 3.0 Network Design