Compute Engine gives you a lot of flexibility when it comes to picking the hardware that runs your VMs. The Professional Cloud Architect exam expects you to know the major machine type families, when each one fits, and what custom configurations are good for. I want to walk through that here so you can make confident sizing calls on the exam and in real architecture work.
Compute Engine machine types fall into four broad categories. Each one targets a different workload profile.
Standard (general purpose). These are balanced machines with a moderate ratio of CPU to memory. They handle a wide range of workloads where you do not need to lean heavily in any one direction. The N1, N2, and E2 series sit in this category. If a workload is mixed and you do not have a strong reason to specialize, a standard machine is usually the right starting point.
High memory. These machines carry more RAM per vCPU than the standard families. They are built for memory-heavy work, such as in-memory databases, large caching layers, and analytics jobs that need to load big datasets at once. The M2 and M3 series fall here. If you see exam scenarios where data has to fit in memory or where the bottleneck is RAM rather than compute, high-memory machines are the answer.
High CPU. These machines flip the ratio in the other direction. You get more vCPUs relative to memory, which suits CPU-bound tasks like batch processing, encryption, scientific computing, and heavy data transformations. The C2 and C2D series are the typical choices. When the workload spends most of its time on the processor and does not need much RAM, this is where you go.
Shared core. These are the lightweight, low-cost options. The vCPU is shared across instances rather than dedicated, which keeps the price low at the cost of consistent performance. E2-micro and E2-small are the common picks. They suit development environments, small websites, low-traffic services, and anywhere CPU performance does not need to be steady.
The Professional Cloud Architect exam will not ask you to memorize every machine series. What it will do is describe a workload and expect you to map it to the right family. The pattern is straightforward once you have it.
If the scenario emphasizes large datasets in memory, in-memory caching, or memory-bound analytics, pick high memory. If the scenario emphasizes CPU-intensive compute, simulation, encryption, or batch transformation, pick high CPU. If the workload is described as balanced, mixed, or general purpose, pick standard. If the workload is a small dev server, a low-traffic site, or a cost-sensitive non-production system, shared core is the right call.
Cost trade-offs also matter. Shared core machines are the cheapest but give up performance consistency. The specialized families cost more than standard machines because you are paying for the CPU-to-memory ratio they provide. Standard machines are the safe middle ground when the question is ambiguous.
On top of the predefined families, Compute Engine lets you build custom machine types. You set the number of vCPUs and the amount of memory directly, instead of accepting whatever ratio the preset gives you.
This is useful when none of the presets line up well with what your workload actually needs. A common case is a workload that wants slightly more memory than a standard machine offers but nowhere near as much as a high-memory one. Picking the next preset up wastes money on resources you will not use. A custom machine lets you size to the actual requirement and stop paying for headroom you do not need.
The trade-off is that custom configurations can carry a small premium per vCPU and per GB of memory compared to the equivalent preset. So custom machines win on cost when they let you avoid significant over-provisioning, and they lose when the workload happens to fit a preset cleanly. The architectural call is whether the savings from right-sizing beat the per-unit premium.
The exam is unlikely to ask you to build a specific custom configuration. What it can ask is whether a custom machine type is the right answer for a given scenario. If the workload has unusual resource requirements that no preset matches well, custom is the right choice. If the workload fits neatly into one of the families, stick with the preset.
The mental model I use for the Professional Cloud Architect exam is to start with the workload profile, map to the matching family, and only reach for custom configurations when no preset is a clean fit. Standard for balanced work, high memory for memory-bound work, high CPU for compute-bound work, shared core for lightweight or cost-sensitive work, and custom when nothing else lines up. That is enough to handle the machine type questions you will see.
My Professional Cloud Architect course covers Compute Engine machine types and custom configuration alongside the rest of the compute material.
