Latency & Performance — Everything Explained

A plain-language reference for the concepts and tweaks in the tuning guide.

Hz (Refresh Rate)
Hz (hertz) means “times per second.” A monitor’s refresh rate is how many times per second it redraws
the image on screen. A 60Hz monitor redraws 60 times a second, a 144Hz one 144 times, a 240Hz one
240 times. Higher Hz means the image updates more often, so motion looks smoother and the most
recent frame reaches your eyes sooner.
The gap between refreshes on a 60Hz panel is about 16.7 milliseconds; on 240Hz it drops to about
4.2ms. That shrinking gap is part of why high-refresh monitors feel more responsive: even with
everything else equal, a new frame can appear up to four times sooner.
Frames per second (FPS) is the matching idea on the GPU side: how many images the graphics card
produces per second. For the smoothest result you want your FPS at or above your monitor’s Hz,
otherwise the screen has refreshes with no new frame to show.

Latency
Latency is delay: the time between an action and its visible result. In gaming the number that matters is
end-to-end or “click-to-photon” latency, the total time from when you move the mouse or click until the
screen actually shows the result.
That total is the sum of many smaller delays stacked together: the mouse sensor reading and reporting
movement, the USB connection carrying that report to the PC, Windows processing the input, the game
engine reacting, the CPU and GPU rendering the next frame, and finally the monitor lighting up the
pixels.
Every tweak in the guide targets one link in that chain. None of them individually is huge, but they add
up, and lower total latency is what makes a system feel “instant” and gives you a real competitive edge
because you see and react to things slightly sooner than an unoptimized setup.

Polling Rate
Polling rate is how often your mouse or keyboard reports its state to the PC, measured in Hz like refresh
rate. A 1000Hz mouse reports 1000 times per second, once every millisecond. An 8000Hz mouse reports
every 0.125ms.
Higher polling means the PC learns about your movement sooner and more often, lowering input
latency and smoothing cursor tracking. The catch is that each report costs a little CPU work, so very high
rates like 8000Hz can raise CPU usage enough to cause frametime spikes on weaker systems, which is
why the guide caps the rate by how powerful your PC is.
Polling rate is not the same as the mouse’s sensor quality or DPI; it is purely about how frequently the
existing data is sent.
MSI Mode (Message Signaled Interrupts)
An interrupt is a signal a device sends to the CPU saying “I need attention now,” for example your mouse
telling the CPU it moved. There are two ways devices deliver these signals.
The old method, line-based interrupts, uses shared physical lines, so devices can end up waiting in a
queue or competing for the same line, adding delay. MSI, Message Signaled Interrupts, lets a device
send its interrupt as a direct message straight to the CPU, with less queuing and lower, more consistent
latency.
Most modern GPUs, SSDs, and network cards support MSI but Windows does not always enable it by
default. The MSI Utility in the guide flips devices into MSI mode and lets you set their priority. This
mainly tightens latency consistency rather than raising raw FPS, and a few devices (some USB
controllers) do not support it and will simply refuse or revert, which is normal and not a sign anything
broke.

Interrupt Affinity
Even with MSI mode on, interrupts still have to be handled by a CPU core. By default Windows can
bounce a device’s interrupts across whatever cores are free, which adds inconsistency and can land
them on the same core your game is using, making the two compete.
Interrupt affinity pins a specific device’s interrupt handling to a chosen core so it always lands in the
same place and does not fight with your game threads. The guide pairs this with isolating the game itself
away from those reserved cores, so for example the GPU’s interrupts get a dedicated core, the network
card another, and the game runs on the remaining cores undisturbed.
With hyperthreading on, each physical core appears as two logical cores that share the same hardware,
so to truly dedicate a core to a device you must reserve both of its logical halves. The payoff is lower and
far more consistent input and frame latency.

Hidden BIOS / NVRAM Tweaking

The BIOS (or UEFI) is the low-level firmware that configures your hardware before Windows even loads.
The settings you see in the normal BIOS menu are only a fraction of what exists. Manufacturers hide
hundreds of additional toggles, stored in NVRAM (non-volatile memory), that control power
management, memory behavior, PCIe link states, and dozens of latency-relevant features.
Tools can dump these hidden settings to a text file, let you change values, and write them back. The
reason this matters for latency is that most hidden settings exist for power saving: they let parts of the
CPU, RAM, and PCIe bus drop into low-power sleep states when idle. Those states save energy but cost
time, because the hardware has to wake back up before it can respond.
Disabling them keeps everything awake and ready, trading higher power draw and heat for lower,
steadier latency. This is also the riskiest stage, because a wrong value can prevent the PC from booting,
which is why knowing how to clear CMOS (reset the BIOS by jumper or removing the battery) before you
start is mandatory.
C-States, Power Management, and Why They Hurt Latency
C-states are CPU idle power-saving levels. C0 is fully active; higher numbers (C1, C3, C6, and so on)
progressively shut down more of the core to save power. The deeper the state, the more power saved,
but the longer it takes to wake up and start working again.
For background or office use this is great. For low-latency gaming it is a problem, because the instant
you need the CPU it may be asleep and has to spend time waking, adding a tiny but real delay and
causing inconsistency. The same logic applies to features like SpeedStep/EIST (which constantly lowers
and raises CPU frequency), ASPM (PCIe link power saving), and many others in the BIOS.
Disabling these keeps the hardware at full readiness. The cost is higher idle temperatures, more power
use, and louder fans, which is the fundamental tradeoff of the entire low-latency approach: you are
spending energy and heat to buy responsiveness.

Monitor Latency (Input Lag, Response Time, Overdrive)

A monitor adds its own delay in two distinct ways that people often confuse. Input lag is how long the
monitor takes to process an incoming frame before it starts displaying it; processing features like
dynamic contrast, motion smoothing, and image “enhancements” all add to this, which is why a Game
or Instant mode that strips them lowers lag.
Response time, separately, is how fast a pixel can physically change color from one shade to another;
slow response causes smearing or ghosting behind moving objects. Overdrive is the monitor’s tool to
speed up pixel response by pushing the pixels harder. Set too low, you get ghosting; set too high, you
get overshoot, also called inverse ghosting, where a bright halo trails moving objects.
The sweet spot is usually a middle overdrive setting, which is why the guide says medium rather than
maximum. Lowering monitor latency is a large and often overlooked part of total click-to-photon time,
frequently bigger than many of the software tweaks combined.

VRR, V-Sync, and Frame Pacing

When the GPU finishes a frame at a different moment than the monitor refreshes, you can get screen
tearing, where the top and bottom of the screen show two different frames. V-Sync fixes tearing by
forcing the GPU to wait for the monitor, but that waiting adds latency, sometimes a lot.
Variable Refresh Rate (VRR), sold as G-Sync or FreeSync, fixes tearing the smart way: instead of the GPU
waiting for the monitor, the monitor adjusts its own refresh timing to match the GPU. This removes
tearing with almost no latency penalty, as long as your FPS stays within the monitor’s VRR range, which
is why the guide suggests capping FPS slightly below your refresh rate when using VRR.
For absolute lowest latency some competitive players turn everything off and run uncapped with
tearing, accepting the visual artifact in exchange for the freshest possible frame. NVIDIA Reflex is a
newer in-game feature that reduces the render queue directly and is the single most effective latency
tool when a game supports it.
Stutters, Micro-Stutters, and Frame Times
Average FPS hides a lot. What you actually feel is frame time, the time each individual frame takes to
render, measured in milliseconds. At a steady 100 FPS every frame takes about 10ms, and the result
feels smooth. The problem is consistency.
A stutter is a noticeable hitch where one or more frames take dramatically longer than the rest, for
example a sudden 50ms frame in a stream of 10ms ones, felt as a visible jerk or freeze. Micro-stutters
are smaller, rapid variations in frame time, say frames alternating between 8ms and 16ms, which the
eye perceives as a subtle unevenness or “not quite smooth” feeling even when the average FPS looks
high.
This is why the guide measures the 1% and 0.1% lows, not just the average: those numbers represent
your worst frames, and they are what stutters live in. Causes include background processes stealing CPU
time, power-saving states waking too slowly, interrupts landing on busy cores, memory instability,
thermal throttling, and shader compilation.
Most of the tuning in the guide is ultimately aimed at flattening frame time variance, because consistent
frames feel far better than a high average riddled with spikes.

DPC Latency
DPC stands for Deferred Procedure Call, the mechanism Windows uses to handle the bulk of a device
driver’s work shortly after an interrupt. If one driver hogs this process and runs too long, it delays
everything else, producing audio crackling, dropped frames, and stutters.
LatencyMon, the tool in the guide, measures DPC latency and shows which drivers are the worst
offenders. A clean system shows low, steady DPC times with no red warnings; a problem driver (often
network, audio, or storage drivers, or buggy RGB and peripheral software) shows up as spikes.
Reducing DPC latency, by updating or removing bad drivers and trimming background software, directly
removes a common and frustrating source of stutters that no amount of FPS will fix.

Delay vs Latency vs Lag

These words get used loosely but mean slightly different things. Latency is the precise, measurable delay
in a system, usually a fixed or near-fixed amount like the time for a frame to render. Lag is the felt
experience of high latency, the subjective “this is sluggish” sensation. Delay is the general umbrella term
for any gap between cause and effect.
In networking specifically, latency often refers to ping, the round-trip time for data to travel to a server
and back, which is separate from your local input and display latency. A perfectly tuned local system can
still feel laggy online if your network ping is high, and vice versa, which is why the guide separates local
hardware tuning from network card tuning.

The Core Tradeoff (Read This Last)

Almost every tweak in the guide spends one resource to gain another. Disabling power saving spends
electricity and heat to gain responsiveness. Higher polling and disabled interrupt moderation spend CPU
cycles to gain lower input latency. Overclocking spends power, heat, and stability margin to gain raw
speed. Disabling safety and reporting features spends your ability to detect problems to gain a sliver of
performance.
None of this is free, and the right amount depends on your hardware, your cooling, and how much you
actually feel the difference. This is why the guide is built around benchmarking before and after every
change: the only way to know whether a tweak is worth its cost is to measure it on your own machine.
A change that helps one system can do nothing, or hurt, on another. Optimize with numbers, keep what
works, and revert what does not.