A new GPU driver lands, the release notes promise a fix for the exact game that’s been stuttering, and you click Install. Two hours later you’re staring at a black screen or a benchmark score that dropped 15 percent. The smart move isn’t to avoid updates. It’s to install them like you expect to undo them.
Here’s the whole methodology: build a rollback point before you touch anything, test the driver against a baseline instead of a feeling, and only then decide whether it stays. It takes about twenty minutes and saves you a reinstall.
Before you install anything: make the undo trivial
Windows keeps the previous driver version in its driver store, so the rollback button in Device Manager usually works. “Usually” is doing a lot of work in that sentence. Create a restore point anyway. NinjaOne’s guide on driver rollbacks lists it as the first preventive measure for a reason: if the new driver is bad enough to destabilize the system, you want a recovery path that doesn’t depend on Windows booting cleanly.
Open Device Manager (Win + X, then Device Manager works fine), expand Display adapters, double-click your GPU, and go to the Driver tab. Write down the current version number. That’s your baseline, and you’ll want it when you’re deciding whether the new build is actually better.
Then take a screenshot of the same tab. If you’re on an OEM machine (MSI, ASUS, Dell, HP, Lenovo, Razer), also note whether you’re running a vendor-specific driver package. OEM BIOSes and driver packages sometimes carry custom hardware IDs that reject the generic NVIDIA or AMD installer with a “not compatible” error, and MSI, Dell, and Lenovo explicitly warn about this. Doing a rollback test and then discovering you can’t reinstall the OEM build is a bad afternoon.
Capture a performance baseline
You can’t evaluate a driver without a number to compare against. Run whatever you have: a built-in game benchmark, 3DMark, or a scripted route through a demanding scene with a frame-time overlay. Note the average FPS and, more importantly, the 1 percent lows. Frame-time consistency is where bad drivers show up first.
Keep this baseline short. One or two runs is enough to catch a real regression. A 2 FPS difference between runs is noise, not data.
Install with the right expectations
Grab the new driver from the manufacturer’s own site or its companion app, never from a driver aggregator. NVIDIA’s Driver Updates page, AMD Software: Adrenalin Edition, or Intel Graphics Software all work and will tell you if a newer version exists. During the install, expect the display to flicker a few times. That’s normal. A restart afterward isn’t strictly required but doesn’t hurt.
Here’s the part most people skip: decide how you’re installing before you click anything.
- Standard update: your normal path. New files install on top of the old ones.
- Clean install (checkbox in the NVIDIA installer): wipes your settings and shader cache but still uses the same registry entries.
- Full clean install with DDU: boots to Safe Mode, strips every trace of the previous driver, then installs fresh.
For a simple test of a new version, a standard update is fine. DDU is for when a normal update leaves problems behind, or when you’ve been on the same GPU for a year and the driver store is full of orphaned packages and stale shader caches that Windows never removes on its own. Defaulting to DDU for every monthly update is overkill. Save it for when the debris is the actual problem.
One thing to close first
Quit every overlay, monitoring, and GPU utility before you reboot into Safe Mode if you’re going the DDU route. MSI Afterburner, RTSS, GPU-Z, Discord’s overlay, whatever’s running. Even if Safe Mode won’t load them, some leave scheduled tasks or startup entries that interfere with the fresh install. This is the step people forget, and it’s the step that causes a “the driver didn’t stick” post on a forum two hours later.
How to actually test the driver
Now you have a new driver and a baseline. Run the same benchmark, same settings, same scene. Compare.
Three outcomes matter:
- Better or equal performance, no glitches. Keep it. Done.
- Performance is close but you see artifacts, flickers, or random crashes. The driver is the suspect. Visual glitches and BSODs are two of the classic symptoms of a bad driver install, and a graphics card showing visual artifacts is one of NinjaOne’s named rollback scenarios.
- Performance dropped noticeably. Same conclusion: roll it back.
Give it a real session, not a synthetic run. Play for an hour. The failure modes that matter (a crash on a specific map, a texture that doesn’t load, a stutter after an alt-tab) don’t show up in a 60-second benchmark. If you switched from AMD to NVIDIA or the other way around, expect the old driver’s leftovers to cause trouble sooner rather than later, and treat DDU as part of the install rather than an emergency fix.
Check the Event Viewer, not just the game
NinjaOne’s rollback verification steps include checking Event Viewer for related error messages. It’s good advice and almost nobody does it. Open it, look at the System and Application logs around the time you were testing. A Display or nvlddmkm/amdkmdag error sitting in the log tells you the driver is failing even when the game looks fine. That’s a rollback signal, not a “give it another day” signal.
Rolling back when the test fails
Two ways, and they solve different problems.
Fast path: Device Manager > Display adapters > your GPU > Driver tab > Roll Back Driver. This works when Windows still has the previous version cached, which it will right after an update. If the button is greyed out, the cached package is gone.
Slow path: Device Manager > Update Driver > Browse my computer for drivers > Let me pick from a list of available drivers on my computer. If an older version is listed, you can install it directly. If not, download it from the manufacturer’s site. Run the old installer over the new one and reboot.
After the rollback, verify: Device Manager should show the version number you wrote down at the start. Run your baseline benchmark again and confirm you’re back to where you were. Then check Event Viewer one more time to make sure the errors stopped.
One preventive habit that pays off
The single best long-term fix is boring: don’t update a working driver unless you have a reason. A new feature, a game that specifically needs it, a bug that’s biting you. NinjaOne’s best-practices list says the same thing, and after a few years of watching release-note-driven updates go wrong, it’s hard to argue. Also keep a folder of known-good driver versions for your card. When a bad update hits, you’re ten minutes from fixed instead of an evening.
Windows Update also has a habit of quietly reverting GPU drivers it decides are problematic, and of serving versions several releases behind the manufacturer’s. If you see your carefully installed driver replaced after a reboot, that’s why. It’s not a bug in your GPU. It’s Windows deciding for you.
Testing a driver is mostly discipline. Note the old version, benchmark once, play for an hour, read the logs. That’s it. Do it every time and a bad driver costs you twenty minutes instead of an afternoon.