A new graphics card is one of the few PC upgrades that still delivers a visible, immediate payoff, but it is also the upgrade most likely to end with a black screen, a Code 43 error, or a GPU that benchmarks 20% below what the box promised. With RTX 50-series and Radeon RX 9000-series cards now sitting at street prices well above their original MSRPs, getting the install right the first time matters more than ever. This tutorial walks through every stage of installing a GPU in 2026, from checking PSU headroom to confirming the card is running at full PCIe bandwidth, in 13 steps that take about 30 minutes once the parts are in hand.
Whether this is a fresh build or a drop-in replacement for an aging card, the sequence below covers driver cleanup, physical installation, BIOS verification, and post-install benchmarking, plus the pitfalls and troubleshooting fixes that cause most support-forum threads about a graphics card that “won’t post” or “shows up as unknown device.”
Most of the actual hands-on work, removing the old card and seating the new one, takes less than ten minutes. The rest of the 30-minute estimate goes to the parts people skip: downloading drivers before you lose your internet connection, running a proper uninstall utility instead of trusting Windows to clean up after itself, and spending a few minutes in BIOS and a benchmark afterward to confirm the card is actually running the way it should. Those verification steps are what separate a GPU upgrade that “just works” from one that technically boots but quietly runs 15% slower than it should for weeks before anyone notices.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Prerequisites: What You Need Before You Start
Before touching a screwdriver, gather the following. Skipping any one of these is the single biggest reason GPU installs go wrong, and most of it takes longer to explain than it does to actually check.
- A PSU with enough headroom. An RTX 5090 draws up to 575W on its own, so a system running one needs at least an 850W unit with native 12V-2×6 cabling or a certified adapter. Smaller cards like the RTX 5070 or RX 9060 XT are comfortable on a 650W PSU. Don’t just add up the GPU’s rated wattage against the PSU’s label; leave at least 150-200W of margin for the CPU, drives, and fans, plus the transient power spikes modern GPUs pull for fractions of a second under load.
- Case clearance. Measure the card’s length (many RTX 5080 and 5090 cards run 320-358mm) against the distance from the front of your case to the drive cage or front fan shroud. Also check the card’s thickness; a 3.5-slot cooler can physically block an adjacent PCIe slot or an M.2 heatsink depending on your motherboard’s layout.
- A Phillips #2 screwdriver and, ideally, an anti-static wrist strap or at minimum a habit of touching the case chassis before handling components. Static discharge damage to a GPU is rare but not impossible, and it’s a free precaution.
- Display Driver Uninstaller (DDU), latest version, downloaded in advance (not after you’ve already removed your old GPU and lost internet access via onboard graphics). DDU is free and widely used specifically because Windows’ built-in uninstall leaves driver remnants behind that can conflict with a new card, especially when switching vendors.
- GPU-Z, latest version, to verify PCIe link width and clock speeds after install. This is the single most useful free diagnostic tool for confirming a GPU install actually succeeded at full performance, not just that it booted.
- A benchmark tool such as 3DMark (Time Spy or Steel Nomad) or Unigine Superposition to confirm the card performs at its rated tier. Free demo versions of most benchmark suites are enough to get a comparable score.
- Windows 11, version 25H2 or later, or a current Linux kernel (6.12+) if running Linux, since very old OS builds can lack PCIe resizable BAR support that modern GPUs expect, and outdated chipset drivers can cause the same reduced-link-width symptoms as a bad physical seat.
- The correct driver package for your card: NVIDIA’s Game Ready or Studio branch, or AMD’s Adrenalin suite, downloaded directly from the vendor site rather than through Windows Update, which sometimes installs an older or generic driver first and requires a second pass to get the full feature set.
One more thing worth checking before you buy: confirm your motherboard’s primary PCIe x16 slot is wired to the CPU’s full lane count (most ATX boards are) and not downgraded to x8 when an NVMe slot is populated, since that can quietly cap a high-end card’s bandwidth. A pre-install checklist from UnixGaming covers this same lane-sharing issue in more detail if your board’s manual is unclear on which M.2 slots share bandwidth with the primary PCIe slot.
Which GPU Should You Buy in 2026?
GPU pricing has not been friendly in late 2026. According to an October 2026 GPU price-performance report from GPUPoet, prices rose on 59 of 81 tracked cards that month, and a resale RTX 5090 now runs roughly 2.4 times its original MSRP. A separate ranking from PickGPU lists the RTX 5090 at a composite performance score of 100 out of 100 with 32GB of GDDR7, selling new on the resale market for close to $4,920.
That gap between MSRP and street price changes the calculus for most buyers. The table below lines up official MSRP against the observed street price GPUPoet’s October 2026 tracker reported for each current RTX 50-series tier.
| Card | VRAM | MSRP | Observed Street Price (Oct. 2026) | Typical Board Power |
|---|---|---|---|---|
| RTX 5090 | 32GB GDDR7 | $1,999 | $4,795 | 575W |
| RTX 5080 | 16GB GDDR7 | $999 | $1,233 | 360W |
| RTX 5070 Ti | 16GB GDDR7 | $749 | $895 | 300W |
| RTX 5070 | 12GB GDDR7 | $549 | $557 | 250W |
| RTX 5060 Ti (16GB) | 16GB GDDR7 | $429 | $600 | 180W |
| RTX 5060 | 8GB GDDR7 | $299 | $328 | 145W |
For most people upgrading a gaming PC rather than chasing flagship bragging rights, the RTX 5070 Ti and RTX 5070 are currently trading closest to their original MSRP, which makes them the more rational install targets this cycle. The RTX 5060 Ti 16GB variant is also worth a look if your main bottleneck is VRAM rather than raw shader throughput, since it keeps 16GB while staying well under RTX 5080 pricing. On the AMD side, the Radeon RX 9070 XT and RX 9060 XT occupy similar price-to-performance territory as the RTX 5070 Ti and RTX 5060 Ti respectively, and the install process described in this guide applies to both vendors equally; the physical seating, power connector type, and BIOS checks don’t change based on who made the silicon.
PCIe Generations and Backward Compatibility
One question that comes up constantly when upgrading rather than building fresh: does a PCIe 5.0 card work in an older PCIe 4.0 or PCIe 3.0 motherboard slot? Yes, PCIe is fully backward and forward compatible by design. A PCIe 5.0 GPU will run in a PCIe 4.0 slot at PCIe 4.0 speeds, and in a PCIe 3.0 slot at PCIe 3.0 speeds, both electrically and mechanically. The practical impact depends heavily on the card tier: a flagship card like the RTX 5090 can lose a measurable percentage of performance in certain titles when dropped to PCIe 3.0 x16 bandwidth, while a mid-range card like the RTX 5070 shows a difference closer to the margin of error in most games, since it doesn’t generate enough frame data per second to saturate even an older bus.
This matters for the troubleshooting steps later in this guide. GPU-Z reports two numbers in its Bus Interface field: the card’s maximum supported link (say, PCIe 5.0 x16) and the current negotiated link (say, PCIe 4.0 x16 if your motherboard tops out at Gen 4). Seeing a lower generation than the card supports is completely normal and not a problem. Seeing a lower lane width, x8 or x4 instead of x16, on a slot that should support x16 is the actual warning sign, and that’s the distinction to keep in mind when you check it in Step 12.
Step 1-3: Prep, Safety, and Driver Cleanup
Step 1: Confirm PSU wattage and case clearance
Open your PSU’s spec sticker or manual and check the total continuous wattage and the number of PCIe power connectors it provides. If your new card needs a 12V-2×6 connector (standard on RTX 5090/5080/5070 Ti reference and most partner boards) and your PSU only has legacy 8-pin PCIe connectors, you’ll need the adapter cable that ships in the card’s box, or a native 12V-2×6 cable if your PSU supports ATX 3.1. While you have the side panel off anyway, measure the card against the case’s actual internal clearance rather than trusting the case manufacturer’s “supports up to 360mm” spec sheet number, since front fans, radiator mounts, or drive cages installed in your specific build can eat into that on paper.
Step 2: Download drivers and diagnostic tools in advance
Download the latest GPU driver package, DDU, and GPU-Z to your desktop before you touch any hardware. If the new card is your only display output and something goes wrong mid-install, you don’t want to be stuck without internet access to redownload a driver. This matters more than it sounds: plenty of installs go sideways specifically because someone removed their old card, found the new one wasn’t displaying anything yet, and had no working GPU and no onboard graphics fallback to get back online and grab the files they needed.
Step 3: Uninstall your old graphics drivers with DDU
Boot into Safe Mode (Shift+Restart, then Troubleshoot > Advanced Options > Startup Settings > Safe Mode) and run DDU to fully remove the old driver, including leftover registry entries that a normal uninstall leaves behind. This single step prevents the majority of post-install driver conflicts between old NVIDIA/AMD software and a new card. It matters even when staying with the same brand; going from an RTX 4070 to an RTX 5070 Ti is still technically a generational driver jump, and residual files from the old architecture’s driver branch have been known to cause intermittent crashes that look like hardware problems but clear up completely after a proper DDU pass.
# Check your current GPU and driver version before removing anything (PowerShell)
Get-WmiObject Win32_VideoController | Select-Object Name, DriverVersion, AdapterRAM
# Example output:
# Name DriverVersion AdapterRAM
# ---- ------------- ----------
# NVIDIA GeForce RTX 4070 32.0.15.6636 12884901888Step 4-6: Removing the Old Graphics Card
Step 4: Power down, unplug, and ground yourself
Shut the PC down fully, flip the PSU’s physical power switch off, and unplug the cable from the wall. ASUS’s official support documentation is direct about this step, instructing installers to “ensure that your power supply unit is turned off and the power cord is removed from the power outlet before proceeding.” Touch the case chassis to discharge any static before handling the card.
Step 5: Remove the old card
Disconnect the PCIe power cables from the old card, unscrew the bracket from the case, press the PCIe slot’s retention latch to release it, and pull the card straight out. Don’t rock it side to side against the slot, since that stresses the connector’s pins. If the latch is hard to reach because of a large cooler overhanging it, most motherboards have a small thumb-tab extension on the latch itself specifically for this; avoid prying the card out with a screwdriver, which is a fast way to chip the plastic slot housing.
Step 6: Prepare the new card and check for a support bracket
Large RTX 5080/5090 cards often ship with (or recommend) an anti-sag support bracket, since a 1.5-2kg card hanging off a single PCIe slot over months can bend the slot or the card’s PCB. Install the bracket now if your case has a mounting point for one. While the card is still out of the case, this is also the easiest time to double-check the power connector orientation; most 12V-2×6 connectors on the card only seat one way, and forcing a cable in backward is a far more common cause of bent pins than people expect.
Step 7-9: Installing the New Card
Step 7: Seat the new card in the PCIe x16 slot
Align the card’s connector with the motherboard’s topmost full-length PCIe x16 slot, the one wired directly to the CPU on nearly every ATX board. ASUS’s install guide describes this as inserting the card “vertically downward into the motherboard’s PCIe slot until a click is heard.” That click is the retention latch locking, and the same documentation notes that “the motherboard’s PCIe slot latch will automatically lock… indicating that the graphics card is successfully inserted into the slot.”
Step 8: Secure the bracket and connect power cables
Screw the card’s bracket to the case, per ASUS’s instruction to “secure the graphics card to the chassis with screws.” Then connect the required PCIe power cable(s) directly from the PSU; ASUS’s documentation specifically notes that “if your graphics card features auxiliary power connector(s), connect the PCIe power connectors from the power supply unit (PSU) to the graphics card.” Never daisy-chain two high-draw 8-pin connectors off a single PSU cable rated for one.
Step 9: Close the case and reconnect everything
Double-check every cable is fully seated, close the side panel, and plug your monitor cable directly into the new card’s display output rather than a motherboard video port, which only works with integrated graphics and won’t show POST output from a discrete card. If you’re reusing a DisplayPort or HDMI cable from the old setup, it’s worth confirming it’s rated for the resolution and refresh rate you’re actually running; a cable that worked fine for 1080p60 on the old card can bottleneck a new card capable of 4K at a high refresh rate.
Step 10-13: Drivers, BIOS, and Benchmarking
Step 10: First boot and BIOS/UEFI verification
Power on and enter BIOS/UEFI (typically Del or F2 at boot) before Windows loads. Confirm the GPU is detected in the hardware list and that Resizable BAR (sometimes labeled “Smart Access Memory” on AMD boards) is enabled, since leaving it off can cost several percent of performance on modern cards. This is also a good moment to confirm the boot drive and primary display settings haven’t reverted to defaults after being out of the slot, something that happens more often than it should on boards that reset certain settings after a hardware change is detected.
Step 11: Install fresh GPU drivers
Boot into Windows, run the official driver installer you downloaded in Step 2, choose a clean/custom install if the installer offers one, and reboot when prompted. Decline any bundled telemetry or recording-software add-ons during install if you don’t plan to use them; they don’t affect GPU performance either way, but trimming them keeps background processes down on a fresh install.
Step 12: Confirm PCIe link width and clock speed with GPU-Z
Open GPU-Z and check the “Bus Interface” field. It should read something like “PCIe x16 5.0 @ x16 4.0” (the current link, then the card’s max supported link), confirming the card is running at full lane width rather than silently dropping to x8 or x4.
# GPU-Z "Bus Interface" field â what a healthy install looks like
Bus Interface: PCIe x16 5.0 @ x16 4.0
# A card running at a reduced link width looks like this instead,
# usually caused by a loose seat or a bifurcated/shared slot:
Bus Interface: PCIe x8 5.0 @ x16 4.0Step 13: Stress-test and benchmark the new card
Run a 15-30 minute stress test (FurMark or 3DMark’s stress test mode) to confirm stability under sustained load, then run a standard benchmark like 3DMark Time Spy to compare your score against published results for your card’s tier. A score more than 10-15% below the expected range usually points to a thermal, power, or driver issue rather than a defective card. Watch the clock speed and temperature readouts during the stress test too; a card that throttles hard within the first few minutes is telling you something about case airflow or cooler contact, not just about the silicon itself.
# Sample nvidia-smi output confirming the card is recognized and healthy
nvidia-smi
+-----------------------------------------------------------------------------+
| NVIDIA-SMI Driver Version: (latest) CUDA Version: (latest) |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce RTX 5070 Ti Off| 00000000:01:00.0 On | N/A |
| 32% 41C P8 18W / 300W | 612MiB / 16384MiB | 0% Default |
+-------------------------------+----------------------+----------------------+If you plan to undervolt or set a custom power limit afterward (covered in the advanced tips section below), a tool like MSI Afterburner can confirm the card is accepting the new curve correctly before you commit to it for daily use.
# MSI Afterburner / nvidia-smi power limit check (confirm the card accepted a custom limit)
nvidia-smi -q -d POWER
# Example relevant output:
# Power Readings
# Power Limit : 270.00 W
# Default Power Limit : 300.00 W
# Min Power Limit : 180.00 W
# Max Power Limit : 330.00 WGame Ready vs Studio vs Enterprise: Which Driver Branch to Install
NVIDIA and AMD both ship multiple driver branches, and picking the wrong one is a quiet source of stutters and crashes that have nothing to do with the hardware install itself. Game Ready drivers prioritize day-one optimization for new game releases and ship on a tight cadence; Studio drivers trade that cadence for longer validation cycles against creative applications like DaVinci Resolve, Blender, and Adobe’s suite; Enterprise/workstation drivers are validated against CAD and simulation software and update far less often. The practical difference shows up mostly in edge cases: someone running Game Ready drivers on a workstation doing long render jobs might occasionally hit a crash in a specific plugin that Studio’s slower, more conservative validation cycle would have caught first.
It’s worth noting that none of these branches require a different physical install process. The driver branch choice happens entirely in Step 11, after the card is already seated and detected; it has no bearing on how you physically mount the card or which power connector it needs.
| Driver Branch | Update Cadence | Best For | Trade-off |
|---|---|---|---|
| Game Ready (NVIDIA) / Adrenalin (AMD) | Weekly to monthly | Gaming, day-one title support | Occasionally less stable in creative apps right after release |
| Studio (NVIDIA) | Roughly quarterly | Video editing, 3D rendering, content creation | Slower to add support for brand-new games |
| Enterprise / Workstation | Infrequent, long-term support | CAD, simulation, certified ISV workloads | Not optimized for gaming workloads at all |
For a build that’s primarily used for gaming, Game Ready or Adrenalin is the right default. Switch to Studio only if the machine spends more time in a timeline editor than a game launcher.
Common Pitfalls When Installing a GPU
- Skipping the DDU cleanup. Installing a new vendor’s driver (say, switching from AMD to NVIDIA) over leftover files from the old one is the single most common cause of random crashes and black screens after an upgrade. It’s tempting to skip this when staying with the same vendor, but generational driver changes can leave behind stale configuration files that cause the exact same symptoms.
- Underestimating PSU headroom. A “750W is enough for anything” assumption breaks down fast with an RTX 5090 pulling 575W on its own, leaving little margin for a modern CPU under full load plus storage and fans. Transient power spikes that last only milliseconds can trip a PSU’s protection circuitry even when the average draw looks fine on paper.
- Forcing the card into the slot at an angle. Large triple-slot cards can look seated when the rear bracket is screwed in but the PCIe connector itself is only partially engaged, which causes the reduced-link-width symptom covered in Step 12. The card should sit level and parallel to the slot before you apply any downward pressure.
- Reusing an old 8-pin-to-12V-2×6 adapter from a different card. Adapters are not universal across vendors and wattage tiers; always use the adapter that shipped with your specific card or a native cable rated for its power draw. Mixing adapters rated for different wattages is one of the few genuinely risky mistakes on this list.
- Plugging the monitor into the motherboard’s video output. This is the most common “my new GPU doesn’t work” support thread, and it isn’t a GPU problem at all, just a cable in the wrong port. It happens most often after a case cable-management pass, when the monitor cable gets re-routed to whichever port is closest.
- Leaving Resizable BAR disabled in BIOS. It’s off by default on many boards after a BIOS update or CMOS reset, quietly costing performance without throwing any error. Since there’s no warning message, this one is easy to miss unless you specifically check for it in Step 10.
- Not removing the anti-static bag residue or plastic film from the card’s heatsink fins before installing, which can trap dust and increase idle temperatures over the card’s first few months. It’s a small thing, but it’s also a thirty-second check that’s easy to forget once you’re focused on the power connectors.
- Assuming a higher-tier card will automatically perform better in every game. If your CPU, RAM, or display’s refresh rate is the real bottleneck, swapping in a much faster GPU can show a disappointingly small gain, which gets misread as an install problem when it’s actually a balance issue elsewhere in the system.
Output Examples: What a Successful Install Looks Like
Three quick checks confirm an install went cleanly. First, Device Manager should show the card under “Display adapters” with no warning icon. Second, GPU-Z should report the correct VRAM, clock speeds, and full PCIe link width as shown in Step 12. Third, a benchmark score should land within the expected range for the card’s tier. Running all three takes about five minutes and catches the overwhelming majority of install problems before they turn into a longer troubleshooting session days later.
# Windows Device Manager status (via PowerShell) for a healthy install
Get-PnpDevice -Class Display | Select-Object FriendlyName, Status
# Healthy output:
# FriendlyName Status
# ------------ ------
# NVIDIA GeForce RTX 5070 Ti OK
# A failed install instead shows:
# FriendlyName Status
# ------------ ------
# NVIDIA GeForce RTX 5070 Ti ErrorIf Status reads “Error” and Device Manager flags the card with “This device cannot start (Code 43),” move straight to the troubleshooting steps below rather than reinstalling the same driver repeatedly, since Code 43 is almost never fixed by a driver reinstall alone. In most real-world cases it traces back to either the power connector not being fully seated or a conflict from a driver uninstall that didn’t fully complete, which is exactly why Steps 3 and 8 matter as much as they do.
Troubleshooting Guide: Fixing GPU Install Problems
Most GPU install problems fall into one of a handful of recurring patterns, and recognizing which one you’re looking at saves a lot of trial and error. The table below maps the symptom you’re seeing to the most likely cause and the fix to try first.
| Symptom | Likely Cause | Fix |
|---|---|---|
| No signal / black screen on boot | Monitor cable plugged into motherboard port instead of GPU | Move the display cable to the GPU’s output |
| Code 43 in Device Manager | Corrupted driver install or PSU power sag at boot | Run DDU in Safe Mode, reinstall driver, confirm PSU has a free, correctly rated connector |
| PC won’t POST with new card installed | Card not fully seated, or incompatible/insufficient PSU connector | Reseat the card firmly until the latch clicks; verify power cable type |
| GPU-Z shows reduced PCIe link width (x8 instead of x16) | Card not fully seated, or slot shares lanes with an occupied M.2 slot | Reseat the card; check motherboard manual for shared-lane M.2 slots |
| Random crashes only under gaming load | Leftover drivers from the previous GPU vendor | Full DDU cleanup in Safe Mode before reinstalling |
| Benchmark score 15%+ below expected | Thermal throttling, power limit, or Resizable BAR disabled | Check temps under load, enable Resizable BAR in BIOS, verify power connector seating |
| Fans spin but no display output at all | Card receiving power but not seated in the data slot correctly | Power down, reseat the card, check for a visible gap at either end of the slot |
| System boots to onboard graphics instead of the new GPU | BIOS still set to prioritize integrated graphics | Enter BIOS and set the primary display adapter to the PCIe slot |
| Coil whine or high-pitched noise under load | Normal electrical behavior on many cards, worse at high/uncapped frame rates | Cap frame rate via V-Sync or an in-game limiter; usually not a defect |
Advanced Tips for a Cleaner GPU Upgrade
Once the card is running and benchmarking correctly, a few optional steps squeeze out extra stability and performance. Undervolting is worth doing on nearly any modern card: dropping voltage slightly while keeping clocks close to stock often cuts power draw and temperatures with little to no performance loss, the same principle covered in our CPU undervolting guide applied to the GPU side instead. On a 300W-class card, a well-tuned undervolt can shave meaningful heat and noise off the cooler without giving up more than a percent or two of raw performance, which matters more in a compact case where airflow is already tight.
It’s also worth revisiting your case fan curve after the swap. A new GPU with a different thermal profile than the old one, even within the same product line, can run noticeably hotter or cooler at idle, and a fan curve tuned for the previous card may now be too aggressive or not aggressive enough.
If you’re running a large card vertically for case aesthetics, make sure the PCIe riser cable you use is rated for the card’s PCIe generation; a Gen 4 card run through a cheap Gen 3-rated riser can silently cap bandwidth the same way a bad seat does. Our vertical GPU mount setup guide covers riser selection in more detail. Builders adding a second GPU or extra NVMe drives off a single PCIe slot should also check whether their board supports PCIe bifurcation before assuming a splitter card will work as expected.
Finally, if you’re installing a high-end card into an older system, double-check your RAM isn’t the real bottleneck; a GPU-bound benchmark can look fine while a CPU- or memory-bound game still underperforms, a gap that’s only gotten more visible as DDR5 pricing has climbed through 2026.
Complete Working Project: A Full RTX 5070 Ti Upgrade Walkthrough
To tie the whole process together, here’s a realistic end-to-end walkthrough of swapping an RTX 4070 for an RTX 5070 Ti in an existing mid-tower build with a 750W PSU. This is a common upgrade path right now, since the RTX 5070 Ti sits close to its MSRP per the pricing table above, and it gives a meaningful VRAM and performance jump over a three-generation-old card without the premium that an RTX 5090 currently commands on the resale market.
- Checked the RTX 5070 Ti’s 300W board power against the 750W PSU’s spec sheet; confirmed two free 8-pin PCIe connectors plus the included 12V-2×6 adapter would cover it with margin for the rest of the system.
- Downloaded the latest NVIDIA Game Ready driver, DDU, and GPU-Z to the desktop before starting.
- Booted into Safe Mode and ran DDU to remove the RTX 4070’s driver completely.
- Shut down, unplugged the PSU, and discharged static by touching the case.
- Disconnected the old card’s power cables, unscrewed its bracket, released the PCIe latch, and removed the RTX 4070.
- Removed the plastic shipping film from the RTX 5070 Ti’s heatsink and seated it in the primary x16 slot until the latch clicked.
- Screwed the bracket to the case and connected the 12V-2×6 adapter, routing the two 8-pin PSU cables cleanly along the case’s cable channel.
- Closed the case, moved the monitor cable to the new card’s DisplayPort output, and powered on.
- Entered BIOS, confirmed the RTX 5070 Ti was detected, and enabled Resizable BAR, which had defaulted to off after a prior BIOS update.
- Booted into Windows, installed the Game Ready driver with a clean install option selected, and rebooted.
- Opened GPU-Z and confirmed “PCIe x16 5.0 @ x16 4.0,” meaning the card was running at full lane width on the system’s Gen 4 slot.
- Ran a 20-minute FurMark stress test with no crashes and peak temperatures in a normal range for an open-air cooler.
- Ran 3DMark Time Spy and landed within the expected score band for an RTX 5070 Ti at stock clocks, confirming the upgrade was fully functional before closing out the project.
That full sequence, from PSU check to final benchmark, is the same 13-step structure covered above, just applied to one concrete card pairing instead of described generically. Total elapsed time, including the benchmark run at the end, came to a little under 40 minutes, with most of that spent waiting on the FurMark stress test and 3DMark run rather than on the physical install itself, which took roughly eight minutes from power-down to case closed.
GPU Installation vs a Full PC Build: Scope Check
It’s worth being clear about what this tutorial does and doesn’t cover. A full PC build involves the motherboard, CPU, RAM, storage, and case assembly in addition to the GPU. If your project is actually a from-scratch build rather than swapping a card into an existing system, the GPU install steps above are still accurate, but you’ll also need to handle CPU socket seating, RAM slot population, and case cable routing before you reach Step 7 here. The install sequence itself, physical seating, power connection, BIOS verification, driver install, and benchmarking, doesn’t change whether the rest of the system is brand new or five years old.
The reverse is also true: someone building a new system from scratch should still run through the driver-cleanup logic in Step 3 if they’re reusing a Windows install from an old drive, since a fresh OS install sidesteps leftover driver conflicts entirely but a migrated drive does not.
Should You Upgrade the PSU Instead of Just the GPU?
If your current PSU is more than four or five years old and you’re jumping up to an RTX 5080 or 5090, it’s worth replacing the PSU alongside the GPU rather than relying on an adapter cable with an aging unit. Capacitor aging reduces a PSU’s real-world headroom below its rated spec over time, and a card pulling 400W+ in transient spikes is a bad place to discover that the hard way. A fresh ATX 3.1-certified PSU with native 12V-2×6 cabling also avoids adapter-related connector issues entirely, which is one less variable to troubleshoot if something goes wrong during first boot.
The cost calculus is straightforward once you factor in the card’s price: spending an extra $100-150 on a quality new PSU to protect a $900+ GPU purchase is a small percentage add-on, and it also buys headroom for a future upgrade down the line without needing to touch the power supply again.
Frequently Asked Questions
Do I need to uninstall my old GPU drivers before installing a new card?
Yes, especially when switching between NVIDIA and AMD. Running DDU in Safe Mode before the physical swap prevents the most common cause of post-install crashes and black screens.
Why does my new GPU show a Code 43 error in Device Manager?
Code 43 typically points to a corrupted driver install or a power delivery issue at boot, not a dead card. Run DDU, reinstall the driver fresh, and confirm the PSU connector is fully seated and correctly rated before assuming the hardware is faulty.
How much PSU wattage do I need for an RTX 5090?
NVIDIA’s reference board power for the RTX 5090 is 575W, so an 850W PSU is a safe baseline once the rest of the system’s power draw is accounted for, with more headroom recommended for overclocked boards.
Why is my GPU running at x8 instead of x16 in GPU-Z?
Usually either the card isn’t fully seated in the slot, or the motherboard shares PCIe lanes between the primary slot and an occupied M.2 storage slot. Reseat the card first, then check your motherboard manual’s lane-sharing diagram.
Should I use Game Ready or Studio drivers?
Game Ready (or AMD Adrenalin) is the right default for gaming-focused systems. Switch to Studio only if the machine is used more for video editing or 3D rendering than gaming.
Do I need a GPU support bracket?
It’s optional but recommended for cards over roughly 1.5kg, which includes most RTX 5080 and 5090 models. A bracket prevents long-term PCB sag that can stress the slot over months of use.
Why does my monitor show no signal after installing a new GPU?
Check that the display cable is plugged into the new GPU’s output port, not the motherboard’s video output, which only functions with integrated graphics.
Is it normal for GPU prices to be above MSRP in late 2026?
Yes. Market trackers including GPUPoet’s October 2026 report show street prices above MSRP on the majority of tracked cards, with the RTX 5090 trading at roughly 2.4 times its original listed price.
Can I install a GPU without removing the old one first?
Only if your motherboard has a second free PCIe x16 slot and you intend to run two cards, which is uncommon for gaming setups today since SLI and CrossFire support has largely disappeared from modern games. For a straight upgrade, remove the old card first.
Do I need to reinstall Windows after a GPU upgrade?
No, a clean driver uninstall with DDU followed by installing the new card’s driver is sufficient for nearly every case. A full Windows reinstall is not required for a GPU swap.
