RAM Test
Measure how fast your machine moves data through memory. The test streams hundreds of megabytes through four passes and turns the sequential bandwidth into a GB/s score you can compare between devices and browsers.
256 MB is the default — big enough to spill out of the CPU cache and into RAM. 1 GB can fail on phones and low-memory machines.
Close heavy tabs first, and compare machines in the same browser — the engine is part of what gets measured.
64 MB can sit in the CPU cache and flatter the result. From 256 MB up, the data comes from real memory.
Hopping between unpredictable addresses defeats the cache and prefetcher, so the random pass lands far below the sequential ones — that's latency, not a fault.
Scores move with the browser and background load, not just the hardware. Same browser, same buffer size, quiet machine — then the comparison means something.
Memory speed is the quiet bottleneck: a machine with a fast processor but slow or single-channel RAM stutters when you switch between heavy tabs, and no CPU meter shows why. A bandwidth number makes it visible — compare an old laptop with its replacement, or a machine before and after a second memory stick. Nothing is installed or uploaded: the buffers live in the browser tab and free the instant you close it.
When you press Start, a Web Worker allocates two large typed arrays off the main thread — so the tab stays responsive — and warms them up so compilation and first-touch costs don't skew the result. Four passes then run, each repeated five times and averaged. Sequential write and sequential read are the classic bandwidth pair, and their average is your score. Random access hops between pseudo-random addresses, defeating the cache and prefetcher, so it measures latency and lands far below the sequential numbers. Memory copy uses the engine's optimized block move — usually the fastest of the four.
How to read your score
| Score | Rating | What it typically means |
|---|---|---|
| 12+ GB/s | Elite | Recent desktop or Apple-silicon Mac with fast multi-channel memory |
| 9 – 12 GB/s | Exceptional | Modern performance laptop or a DDR5 desktop |
| 6.5 – 9 GB/s | Excellent | Healthy recent hardware — a dual-channel DDR4 desktop or a good laptop |
| 4 – 6.5 GB/s | Good | Mainstream laptops and older desktops — comfortable for daily work |
| 2 – 4 GB/s | Fair | Budget, compact, or aging hardware, often running single-channel |
| Under 2 GB/s | Poor | Very old hardware — or something is off: close heavy tabs, plug in, retest |
Choosing a buffer size
| Buffer | What it tests | Trade-off |
|---|---|---|
| 64 MB | Small enough to sit partly in the CPU cache, so the numbers run high | Quickest pass — treat it as a cache test, not a RAM test |
| 256 MB | Spills out of every consumer cache into actual RAM | The default — balanced, and the size to compare devices on |
| 512 MB | Deep into RAM territory, far past any L3 cache | Steadier averages for a slightly longer run |
| 1 GB | Stresses allocation as well as bandwidth | Can fail outright on phones and low-memory machines |
What a browser test can — and cannot — measure
This test measures the bandwidth JavaScript can actually reach — what matters for how fast web apps feel, not the sticker spec. A DDR5-6000 kit won't post its rated tens of GB/s here: one thread, one browser engine, with garbage collection in the way, so scores land well below native tools like AIDA64 and are best compared with other browser scores. The sandbox can't see memory timings, channels, or ECC status (paste a kit's numbers into the RAM Latency Calculator instead), and a failing stick needs a boot-level tool like MemTest86, not a web page. Memory is also only one suspect — load the processor with the CPU Stress Test and the graphics card with the GPU Stress Test to see which is holding things back.