Cache sizes are inferred from a memory-latency probe (where access time jumps = a cache boundary); clock is inferred from a dependent-op loop. Both are rough estimates affected by the JIT, boost clocks and other load — not vendor specs. The browser can't read the real model/clock/cache.
Memory bandwidth
Sequential read
—
Sequential write
—
Copy
—
Buffer
—
Single-threaded JavaScript over a buffer far larger than L3, so it measures real DRAM traffic — but through one core and the JS engine, so hardware peak (all channels, all cores) is higher. Copy counts bytes read + written, the STREAM convention. Comparable across machines; best of three passes.
SIMD compute
f32 single-core
—
f32 all cores
—
SIMD speedup
—
Integer SIMD
—
Hand-written WebAssembly kernels using 128-bit SIMD (f32x4 multiply-add chains, i32x4 add/xor/shift), pure register work — real GFLOPS, much closer to native than any JavaScript loop. Speedup is SIMD vs an identical scalar WASM kernel; the theoretical maximum is 4×. All-cores runs the same kernel on every logical core at once.
Threads & shared memory
Timer resolution
—
Atomics · contended
—
Atomics · independent
—
Shared read · all cores
—
Read scaling
—
True shared-memory multithreading via SharedArrayBuffer: every core hammering ONE counter (cache-line contention — the cost of core-to-core traffic), each core on its own counter (the uncontended ceiling), and all cores reading one shared buffer at once (aggregate DRAM bandwidth, vs the single-core card above). Needs cross-origin isolation, which also sharpens the browser clock — the resolution readout shows what your measurements are working with.
Video encode
Codec
—
1080p encode
—
Acceleration
—
Output
—
WebCodecs driving the machine's video-encode engine (the silicon streamers rely on — NVENC, QuickSync, AMF): 120 generated 1080p frames through the best supported codec, hardware-accelerated where the browser allows it. Well above your recording framerate = headroom to stream; near or below it = the encoder, not the game, will be the bottleneck.