1% lows explained: the frame-pacing masterclass (why 300 fps still stutters)

Averages lie: learn to read percentiles

300 fps average means frames arrive every 3.3ms on average. But if one frame in a hundred takes 30ms, you get a visible hitch ten times a minute while the average barely moves. That's why we read percentiles: the 1% low is the frame rate your worst 1% of frames delivers — effectively, how bad your bad moments are.

A healthy competitive rig keeps 1% lows above two-thirds of the average (300 avg / 200+ lows). A 300/80 split plays worse than a locked 144. When you benchmark, log frame times and look at the graph's shape: flat line good, spikes bad, sawtooth suspicious.

• 1% low ≥ ~66% of average = healthy pacing
• Judge every tweak by its effect on lows, not the average
• A locked lower cap often feels better than an unstable higher rate

The five species of stutter and their fingerprints

Shader-compilation stutter: sharp spikes the first time effects appear, disappearing on a second run of the same area — the game is compiling GPU programs mid-match. Asset-streaming stutter: spikes tied to traversal, new zones, or fast camera rotation with VRAM near full. Scheduler stutter: random spikes uncorrelated with gameplay — background processes stealing the main thread.

Thermal/power stutter: a clean graph that degrades into periodic sawtooth after 10–20 minutes as clocks cycle against a limit. Memory stutter: broad, constant noise on systems with XMP/EXPO disabled or single-channel RAM.

• First-run spikes that vanish on replay → shader compilation
• Spikes on traversal/rotation → asset streaming / VRAM pressure
• Random spikes at any load → background processes
• Clean-then-sawtooth after minutes → thermals or power limit
• Constant broad noise → memory configuration

The diagnosis workflow (15 minutes, free tools)

Record a 5-minute frame-time log with CapFrameX or PresentMon in a repeatable scenario, with hardware telemetry alongside: GPU clock and power, per-core CPU load, VRAM allocation, temperatures. Reproduce the stutter and match its timestamp against the telemetry — what moved when the spike hit? A clock drop names thermals, VRAM spillover names streaming, a background process spike names the scheduler.

Then do a second run immediately without restarting the game — anything that improved was shader compilation warming up, not your imagination.

The fix ladder — in this order, one rung at a time

Fixes are ordered by effort-to-impact. One: enable the game's shader pre-compilation and let it finish, update GPU drivers, fresh boot. Two: cap the frame rate under your refresh — pacing loves headroom. Three: quiet the OS — per-frame overlays inject into the render loop and are the most common scheduler offender we find, followed by launcher stacks and Game DVR.

Four: memory — enable XMP/EXPO (the single most-skipped free upgrade in the PCs we audit) and confirm dual-channel. Five: drop textures one notch if VRAM allocation sits above 90%. Six: thermals — dust, fan curves, and an undervolt that holds clocks flat instead of cycling. Only after all six rungs do you get to blame the hardware.

• 1. Shader pre-comp + driver update + fresh boot
• 2. Frame cap under refresh
• 3. Overlays off, background apps culled
• 4. XMP/EXPO on, dual-channel confirmed
• 5. Textures down one notch if VRAM > 90%
• 6. Thermal ceiling: cooling + undervolt

When it actually is your hardware

If the graph shows one CPU core pinned at 100% while the GPU coasts, no setting fixes a CPU bottleneck in a CPU-bound title — that's an upgrade conversation. Same for 8GB of VRAM at 4K textures, or a single stick of RAM in a laptop. The good news: the six rungs above rescue the large majority of stutter cases we see without touching hardware — and they're exactly what our FPS Boost and system packs automate, fully reversibly.