AZ-900 Describe Azure architecture and services Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Compute-optimized VMs (Fsv2-series)

  Why it's wrong here

  Fsv2-series compute-optimized VMs are engineered around a high CPU-to-memory ratio, typically offering only 2 GiB of RAM per vCPU. This makes them powerful for batch processing, web serves, and other compute-bound tasks, but their limited memory capacity and bandwidth cannot accommodate large in-memory workloads like SAP HANA or giant databases. Choosing these for a memory-intensive workload would result in exhausting RAM and relying on slow disk paging, degrading performance severely.

• ✓

  Memory-optimized VMs (M-series, E-series)

  Why this is correct

  Memory-optimized M-series and E-series VMs are specifically built to support very large memory footprints, offering high memory-to-CPU ratios and RAM capacities ranging from hundreds of GiB to multiple terabytes. These families are certified for SAP HANA, in-memory analytics, and large relational databases because they provide the RAM capacity and memory bandwidth to keep entire datasets resident in memory. This design ensures the SQL Pool or SAP workloads can run without constant disk reads, delivering the low-latency query performance expected from a memory-intensive deployment.

• ✗

  Storage-optimized VMs (Lsv2-series)

  Why it's wrong here

  Lsv2-series storage-optimized VMs focus on high disk I/O throughput and low latency, featuring direct-attached NVMe storage for scratch and cache-heavy applications. Their local ephemeral disks and CPU/memory layout are selected for storage throughput, not for maximizing RAM capacity or memory bandwidth. While they may handle some database workloads, their memory size is generally insufficient and the design intent is to eliminate storage bottlenecks rather than host enormous in-memory data sets, making them inappropriate for a primarily memory-intensive scenario.

• ✗

  GPU VMs (NC-series)

  Why it's wrong here

  NC-series GPU VMs are designed for massively parallel compute tasks such as AI training, deep learning, and graphics rendering. Their powerful NVIDIA GPUs accelerate matrix math and visualization, yet the underlying CPU and RAM configuration is not tuned for memory density or high memory-to-core ratios. Large in-memory databases do not benefit from GPU parallelism in the same way, so this family lacks the massive memory footprint and memory bandwidth that such workloads demand.