Understanding Memory Speed and Cache

When you talk about memory performance, you need to tell two numbers apart. One is throughput, which is how much data can be moved at once, and the other is latency, which is how long you wait until the data you need arrives. Work that handles large files or images as a whole is strongly affected by throughput, while work that follows scattered data is strongly affected by latency. Web Bench measures both to show in what way your memory is strong or weak. Throughput is in GB/s so higher is better, and latency is in ns so lower is better. Understanding this distinction explains why memory of the same capacity can feel different. If the memory results are low, first check whether other tabs or extensions in the browser are using resources.

The throughput tests read, write and copy 16 MB of contiguous memory in order. The quick test runs read and copy, and the standard test runs all three. The latency test follows data using dependent pointer chasing, where the next location depends on the result of the previous read, and repeats this while changing the size of the region it accesses from 16 KB to 256 MB. When the region is small, it is served from the fast cache inside the CPU and latency is short, and when the region grows beyond the cache, it has to reach slower memory so latency increases. Comparing the results by size lets you estimate where the boundaries of the cache hierarchy lie. This value does not replace the capacity or standard on a hardware spec sheet. A cache is a mechanism that keeps frequently used data close by to reduce waiting, and whether the working set fits in the cache often decides the speed.

In daily life, memory performance is felt in a browser with many tabs open, in editing large photos and videos, and in scene changes in games. However, situations where memory capacity runs out and spills to disk are not part of this measurement, and real-world experience is also affected by the CPU and storage. To get stable results, close other apps while testing and repeat several times under the same conditions to check the variation. On a laptop, the power mode can affect how memory and CPU behave, so note your settings. Rather than declaring a winner over small differences in numbers, it is more reasonable to interpret the items with large gaps, and every value is a reference that varies by environment. Rather than recording a single peak speed, use values measured in the same environment as your usual setup as the baseline so that it is easier to compare changes later. The memory temporarily reserved after the test is cleaned up by the browser, so you do not need to clear it yourself.