How to Measure DPC Latency (and Fix the Stutter It Causes)
Published
On this page
DPC latency is the time a Windows driver spends holding a CPU core inside a deferred procedure call, and when one driver holds it too long you get audio crackle, frame stutter, and input that feels like it hitched — even at a high average FPS. The good news is that it is measurable, and the measurement points directly at the driver responsible.
This matters because DPC latency is invisible to every tool most people already run. MSI Afterburner and RTSS show you frame rate and frame times — they will show you that a frame was late, but never why. Task Manager shows CPU percentage, which stays low while a single driver blocks a core for milliseconds at a time. You need a tool that watches the kernel.
What DPC latency actually is
When hardware raises an interrupt, Windows does the bare minimum inside the interrupt service routine and defers the rest of the work to a DPC (deferred procedure call) that runs at a high priority level. While a DPC runs, it cannot be preempted by your game’s threads on that core.
That is the whole problem. A DPC that finishes in 20 µs is invisible. A DPC that takes 3 ms has stalled a core for roughly the length of two frames at 144 Hz — and no frame-rate counter will tell you a driver was the reason.
Because the work is deferred rather than distributed, a single badly written driver can degrade an otherwise fast machine. This is why a high-end PC can stutter while a modest one does not.
What counts as good, bad, and broken
| Highest single-driver DPC | What it means | Symptoms you’d notice |
|---|---|---|
| Under 100 µs | Healthy | None |
| 100–500 µs | Normal for most systems | None in games; possible clicks in a DAW |
| 500–1000 µs | Marginal | Occasional audio artifacts, rare hitches |
| 1000–2000 µs (1–2 ms) | Bad | Audible crackle, visible frame hitches |
| Over 2000 µs (2 ms) | Broken | Consistent stutter, dropped audio, hitching under load |
Two rules make these numbers useful. First, judge the peak, not the average — averages hide exactly the spikes that cause the symptom. Second, measure under load, not at an idle desktop. Plenty of systems look clean until the Wi-Fi adapter is actually moving traffic or the GPU is drawing power.
How to measure it
Method 1: Tier1Suite’s scheduling-latency monitor
Tier1Suite includes a DPC and scheduling-latency monitor in its Tier1Monitor module. It samples continuously and reports current, mean, and peak figures, so you can watch the numbers move while you reproduce the problem — start a download, launch a game, plug in a USB device — and see which action correlates with the spike.
It is read-only and in-app: there is no in-game overlay and no driver of its own. That is deliberate, and it means running it carries no anti-cheat risk.
Pair it with the module’s latency benchmark, which measures real Sleep(1ms) accuracy. Together they separate two different problems that feel identical: a timer that is too coarse (fixed with timer resolution) and a driver that is blocking a core (fixed by finding the driver).
Method 2: LatencyMon
LatencyMon is the long-standing free third-party tool for this, and it does one thing Tier1Suite does not: it attributes latency to a specific named driver file. If your peak is high and you need to know whether it is ndis.sys, nvlddmkm.sys, or storport.sys, run LatencyMon for several minutes under load and read the drivers tab.
Method 3: xperf / Windows Performance Analyzer
The Microsoft-supported route, and the most precise. Install the Windows Performance Toolkit, then capture a trace:
xperf -on PROC_THREAD+LOADER+DPC+INTERRUPT -stackwalk ProfileDPCStop it with xperf -d trace.etl and open the result in Windows Performance Analyzer. This is overkill for most people, but it is the only method that gives you call stacks rather than just a driver name.
How to fix a bad driver once you’ve found it
Work in this order — the cheap fixes solve most cases.
- Update the driver from the vendor, not Windows Update. Windows Update often ships an older WHQL build. Get network and chipset drivers from the motherboard or NIC vendor directly.
- Roll back if the problem started after an update. Device Manager → the device → Driver tab → Roll Back Driver. New is not always better, particularly for NICs.
- Turn off adapter power saving for network cards. In Device Manager, open the adapter, and on the Power Management tab clear Allow the computer to turn off this device to save power. This is one of the most common single fixes on laptops.
- Disable hardware you do not use. Bluetooth, a second NIC, fingerprint readers, and vendor RGB controllers all run drivers. Disabling an unused device in Device Manager removes its DPCs entirely.
- Remove motherboard “utility” software. Vendor monitoring suites frequently install kernel drivers that poll sensors on a timer. These are a recurring source of periodic spikes.
- Check power plan and C-states last. A deep C-state can add wake latency. Switching to a performance power plan is worth testing, but treat it as a workaround — it masks a bad driver rather than fixing it.
What DPC latency will not fix
Being honest about the ceiling here matters, because “fix your DPC latency” gets repeated as a cure-all.
If your peak DPC is already under 100 µs, there is nothing to gain — chasing it further will not add frames or reduce input lag. DPC work is a stutter problem, not a throughput problem. If your average FPS is simply too low, the answer is graphics settings, resolution, or hardware, not driver latency.
Equally, a clean DPC profile does not rule out other causes of hitching: shader compilation, stutter from a full shader cache, background downloads, and thermal throttling all produce similar symptoms and need different fixes.
Related guides
Frequently asked questions
What is a good DPC latency in Windows?
Under about 100 microseconds at the highest single-driver figure is healthy, 100–500 µs is usually fine for gaming but marginal for real-time audio, and sustained spikes above 1000 µs (1 ms) reliably cause audio dropouts and frame stutter. Judge the peak, not the average — a driver that is quiet for 59 seconds and then blocks for 4 ms still ruins one second of every minute.
Does DPC latency affect FPS or just audio?
It affects both, but in different ways. A long DPC blocks the CPU core it is running on, so a frame that needed that core arrives late and you get a visible hitch rather than a lower average FPS. Audio is more sensitive because its buffers are small, which is why audio crackle is usually the first symptom people notice.
Which drivers most often cause high DPC latency?
Network drivers (especially Wi-Fi and Killer/Intel NICs), storage controllers, and older audio drivers are the most common culprits, followed by USB controllers and vendor 'utility' drivers from motherboard software. On laptops, power-management and chipset drivers are frequent offenders because they run at aggressive idle states.
Can I fix DPC latency without reinstalling Windows?
Almost always, yes. The usual fix is updating or rolling back the specific driver you identified, disabling the device if you do not need it, or turning off adapter power saving for a network card. A clean install only helps when the problem is a stack of leftover vendor utilities, and even then removing them individually is faster.
Is DPC latency the same as input lag?
No. Input lag is the total time from your physical action to the pixel changing on screen. DPC latency is one specific contributor: time the kernel spends in deferred procedure calls, which can delay input processing and frame delivery. Fixing a bad DPC offender removes stutter spikes, but it will not, on its own, replace a low-latency monitor or a high polling-rate mouse.