Hardware: Components, Performance, and Buying Decisions
Hardware is the physical layer of computing: processors, memory, storage, input/output devices, and the buses and controllers that connect them. The DSST exam tests whether you can map a specification to a real capability, compare technologies on the right metric, and diagnose a bottleneck instead of guessing. Questions are usually scenarios, such as a machine that is slow, a device that will not fit, or a budget that must be spent where it helps most.
Processor and the memory hierarchy
- The CPU fetches, decodes, and executes instructions; throughput depends on clock speed, cores, cache, and instruction set, not clock speed alone.
- Hierarchy from fastest and smallest to slowest and largest: registers, L1/L2/L3 cache, RAM, SSD/HDD, tape or cloud archive.
- RAM is volatile working storage; ROM and firmware are non-volatile and hold boot code such as BIOS/UEFI.
- A 32-bit address bus reaches 2^32 bytes, about 4 GiB; wider addresses remove that ceiling.
Storage and input/output
- HDDs are judged by capacity, seek time, and rotational latency; SSDs by IOPS and latency, with no moving parts.
- Interfaces matter: SATA, NVMe/PCIe, USB, and Thunderbolt differ in bandwidth, so a fast drive on a slow bus is wasted.
- Peripherals need drivers; interrupts and DMA let I/O proceed without constant CPU polling.
Worked example
A workstation with 8 GB of RAM slows dramatically when a large spreadsheet is open, and the drive light stays on. The user proposes buying a faster CPU. Reasoning: the activity indicates swapping, meaning the working set exceeds physical RAM, so the disk is the bottleneck. Adding RAM removes the paging, and a faster CPU would mostly idle waiting on I/O. If load times remain poor after RAM is adequate, an SSD on an NVMe interface addresses storage latency instead.
Traps that cost marks
- Confusing GB (decimal, 10^9 bytes) with GiB (binary, 2^30 bytes); drive vendors typically advertise decimal.
- Treating cache as a substitute for RAM, or comparing MHz across different architectures as if it predicted performance.
- Assuming more cores always helps; single-threaded work does not scale with core count.
- Forgetting that maximum RAM is capped by the motherboard chipset and operating system, not just by free slots.
- Mixing up input, output, and storage devices when classifying a device in a scenario.