Game Stutters on First Load? Shader Compilation Stutter, PSO Caching and Why Driver Updates Bring It Back

2026-10-22 | GPU Repair | About 11 minutes | HK GPU Fix Editorial

Key Takeaways

Shader compilation stutter is a first-time hitch: a game freezes briefly when it meets a new effect or area, then runs smoothly once that shader is cached. The work runs on the CPU, not the GPU, so a strong graphics card alone does not prevent it. Driver updates can invalidate the cache and bring one rough session back.

Which Problem Is This Article About?

If your game stutters the first time you enter a new area or meet a new effect, and then the same place plays smoothly on a second run, this article is for you. If instead the picture itself looks blurry, smeared, or flickery on thin detail after you turn on DLSS or FSR, that is an upscaling problem, and the settings test is different: toggle the upscaler off and step through its quality modes. The two problems often get confused because both can appear after a change — a driver update in one case, an upscaling setting in the other — but their signatures and their fixes are different. Neither one, on its own, is evidence of a failing graphics card.

What Shader Compilation Stutter Feels Like

The pattern is distinctive. A game reports a good average frame rate, yet play is interrupted by brief freezes or jerky pauses. The pauses tend to happen when you first enter a game, first walk into a new area, or first see a new effect, enemy, weapon, weather state, or lighting state on screen. After that first encounter, the same scene plays smoothly. Restarting the game later is usually smooth too, until something invalidates the cache, as explained below.

The key signature is first time only, then gone. If the same spot hitches every single time you walk through it, in exactly the same way, that points away from shader compilation and towards asset streaming, a CPU limit, or memory pressure. That distinction matters, because the wrong diagnosis leads to the wrong fix: buying a faster graphics card will not cure a problem that lives on the CPU, and clearing a cache will not cure a problem that repeats regardless of cache state.

What a Shader Is, and Why It Must Be Compiled on Your PC

A shader is a small program that runs on the GPU and describes how something is drawn: a surface material, lighting, shadows, transparency, post-processing, particles, ray tracing, and similar work. Games do not ship finished GPU machine code. They ship shaders in an intermediate form, and the graphics driver must translate, or compile, them into machine code for the specific GPU and driver combination in the machine. Every PC is different — different GPU model, different driver version — so this step cannot be fully done once by the developer for all PCs.

In modern graphics APIs, DirectX 12 and Vulkan, compiled shaders are bundled into a Pipeline State Object, or PSO. A PSO is one package that tells the API and the GPU the shaders plus the render state for a draw: blending, depth testing, raster settings, render-target formats, and similar parameters. Creating a PSO is a blocking, multi-millisecond operation. If the game asks for a PSO it has never built before in the middle of gameplay, the frame cannot proceed until compilation finishes. One frame therefore takes far longer than the normal budget — 16.7 milliseconds at 60 frames per second, 8.3 milliseconds at 120 frames per second — and the player perceives a hitch.

One secondary summary of Unreal Engine documentation puts runtime PSO creation at 100 milliseconds or more for some PSOs. Treat that figure as order-of-magnitude only: it varies with shader complexity, CPU speed, and driver, and it should not be quoted as if every PSO takes 100 milliseconds. The important point is the scale of the mismatch. A normal frame budget is measured in single-digit or low double-digit milliseconds, while a live compilation can cost several times that budget, so even one uncached PSO is felt as a freeze.

The compilation work itself runs on the CPU, not the GPU. That is why a faster CPU shortens the spikes but does not remove them, and why a very strong GPU can still stutter badly: the GPU is idle, waiting on the CPU to finish compiling. It is also why the common claim that a stronger GPU fixes stutter is only true when the GPU is actually the limit. For shader stutter, the GPU is not the limit; the missing precompiled pipeline is.

Why DirectX 12 and Vulkan Made This More Visible Than DirectX 11

Under DirectX 11, the driver managed much more of the pipeline state, and shaders were typically compiled up front at load time. The cost was longer loading screens, but gameplay was smoother. DirectX 12 and Vulkan are lower-level APIs: they hand pipeline-state control to the developer, which enables performance features but also makes just-in-time PSO creation the developer's problem. If a developer does not pre-warm every PSO the player might hit, the first encounter compiles live and the player feels it.

Consoles largely avoid this because the hardware is fixed: developers can ship precompiled shaders for one known GPU. PCs cannot, because of the hardware and driver matrix. The obvious question is why games do not simply precompile everything. The answer is combinatorial. A single asset can combine with many variants — dynamic versus static lighting, shadows on or off, ray tracing on or off, anti-aliasing mode, transparency — producing a very large number of possible PSOs. Precompiling all of them would bloat cache size and loading time with variants the player may never encounter. Epic's own documentation, as summarised in secondary sources, concedes that PSO precaching can be burdensome for large projects and can still leave gaps in the cache that lead to hitches. Even a game with a compiling shaders screen can therefore still hitch occasionally on an uncached combination, such as a lighting or visual-effect combination the precache missed.

Precaching quality also varies by title even on the same engine. Two games built on the same engine can behave very differently depending on whether the studio implemented precaching properly. It is therefore not accurate to blame or praise an engine as a whole. Historical examples sometimes cited in sources, such as The Callisto Protocol at launch, Star Wars Jedi: Fallen Order, and Gotham Knights, should be read as historical illustrations of launch-era behaviour, not as claims about how those games behave today after patches.

The Shader Cache: Where Compiled Work Is Kept

After a shader or PSO is compiled once, the result is stored in a shader cache on disk, so the next time the same shader is needed it loads from cache instead of recompiling. That is the whole reason for the first time only pattern. Three layers should not be conflated.

First, the game or engine precompilation screen — the Compiling shaders or Building shaders step at first launch — compiles the game's known PSOs before gameplay. Skipping it does not remove the work; it moves the work into gameplay as hitches. Let it finish completely.

Second, the driver shader cache. The GPU driver caches compiled binaries keyed to the driver version and GPU. NVIDIA Control Panel exposes a Shader Cache Size setting; sources advise leaving it at Driver Default or larger, because a too-small cache forces recompilation of shaders already built. AMD manages its cache size automatically, with no equivalent user limit, according to one source. Cache locations reported by multiple sources, on Windows, include the NVIDIA, AMD, and Intel DXCache folders under the local application data folder, plus the system DirectX cache. No universal cache size in gigabytes should be claimed for any vendor: sources give settings guidance, not a universal size.

Third, platform pre-caching. Steam's Shader Pre-Caching setting, under Settings and Downloads, downloads or prepares shaders ahead of time, but according to one source it applies to Vulkan and OpenGL titles, not DirectX 12 games. It must not be promised as a fix for DirectX 12 stutter.

An emerging development should be described with care, not over-promised. Microsoft's Advanced Shader Delivery, reported in 2025 to 2026 coverage, moves precompilation to Microsoft's cloud: developers capture shaders into a State Object Database via DirectX 12, Windows 11, and the Microsoft Agility SDK, and the store precompiles for many hardware and driver combinations. NVIDIA driver 621.01, reported on 2026-10-21, is reported to enable it on Windows 11, and AMD Adrenalin 26.6.1 and later and a Qualcomm Snapdragon X2 driver have also been named in reports. NVIDIA has been reported as claiming a near-100 percent cache hit rate; treat that as a vendor-reported claim that is uncertain in the field, and coverage as game- and store-dependent. Reported launch supporters via the Xbox PC app include Gears of War: E-Day and Call of Duty: Modern Warfare 4, and one report says Steam purchases can also benefit if the Xbox PC app is installed, which remains uncertain and should be verified before relying on it. This must not be presented as the problem now being solved for all games.

Why a Driver Update or Reinstall Re-Triggers the Stutter

Compiled shader binaries are tied to the driver, and GPU, that produced them. When the driver is updated, reinstalled, or sometimes when the game is patched, old cache entries can be invalidated or discarded and the shaders must be compiled again. Whether a given driver update invalidates the entire cache or only part of it is uncertain and may differ by vendor, driver version, and game, so articles should say can invalidate rather than always wipes everything.

The practical consequence, confirmed by several sources, is to expect one rough session after a driver update. The stutter should then fade across repeated runs of the same areas as the cache rebuilds. Rolling back the driver out of panic is usually the wrong response, because the rollback itself would invalidate the cache again. The same logic applies to clearing the shader cache manually, switching GPU vendor or model, reinstalling Windows, and some game patches that change shaders.

Clearing the cache is not routine maintenance. Sources agree it should only be done when corruption is suspected — stutter that never improves across repeated runs of the same area — because the next session will stutter while the cache rebuilds. Clearing the cache never deletes game settings or save files; it stores compiled shaders only. A game can also have both shader stutter and streaming stutter at once, which is common in open-world games, and some games compile shaders in the background during play, so hitches may be spread over the first hour rather than all at the start. One source describes shader stutter lasting through the first 30 to 60 minutes of a new session in some Unreal Engine 5 titles; present that as that source's observation, not a rule.

How to Measure It: Frame Time, Not Average FPS

Average frames per second hides this problem. A game can average 120 frames per second and still feel terrible if individual frames spike from around 8 milliseconds to around 50 milliseconds. Teach frame time, measured in milliseconds per frame, and 1 percent and 0.1 percent lows, not just the FPS counter. A constantly noisy frame-time band points to a CPU limit or a sync problem; isolated tall spikes point to compilation, streaming, or interrupts.

Tools named across sources include CapFrameX, built on Intel PresentMon, PresentMon itself, whose GPU Busy metric helps separate GPU limits from CPU limits, MSI Afterburner with RTSS for a frame-time graph overlay, and HWiNFO64. The method is to pick a repeatable 60 to 90 second route, capture several runs, and read the graph rather than just the summary. One source's cold-cache versus warm-cache test on the same route showed nearly identical averages but very different 0.1 percent lows, which illustrates why averages mislead; that test's exact numbers are that source's own bench, not a general fact, and should not be copied as if universal.

Two tests are specific to shader stutter. The cheaper test is to replay the same route twice without changing anything. Shader stutter fades on the second run; streaming, CPU, and VRAM problems repeat. The stronger test, from a PSO-precaching guide, is to clear the driver shader cache and then replay the same route: if hitches come back in the same first-encounter places, compilation is implicated, whereas if they persist unchanged regardless of cache state, look elsewhere, such as streaming, asset loading, or garbage collection. Warn readers that this test deliberately causes a stuttery session, so it should be done knowingly, not as casual maintenance.

Differential Diagnosis: Shader Stutter Versus the Look-Alikes

SuspectSignatureWhat to checkWhat rules it in or out
Shader and PSO compilationHitches on first encounter with an effect or area; fades on repeat runs; returns after a driver update, cache clear, or game patchFrame-time graph with isolated tall spikes early in a session; second run of the same route is cleanRepeats identically every run regardless of cache state points away from this
VRAM exhaustionTexture pop-in, textures going blurry or low-detail, hitches when turning quickly or entering dense areas; VRAM at or near capacityVRAM usage overlay; lower texture quality or resolution and see if it clears; note that VRAM allocated is not the same as VRAM neededImproves when texture settings drop; persists after the cache is warm
CPU bottleneckUnstable FPS with GPU usage well below about 95 percent; dropping resolution or enabling upscaling barely changes FPS; worse in dense areas such as cities, large battles, or multiplayer lobbiesPer-core CPU usage, since one core at 100 percent can bottleneck even if overall CPU looks low; resolution-drop testFPS rises when resolution drops, which points more to a GPU limit
System RAM and page-file pressureHitches under memory pressure; committed memory near total RAM; sustained hard faults per second in Resource MonitorTask Manager, Performance, Memory while reproducing; close background apps as a testNo memory pressure visible points away from this
Traversal and asset streamingHitch recurs every time you cross the same world boundary or spawn the same heavy actor, unlike shader stutter, which is once per sessionSame-route repeat test; whether the game is on a hard drive versus an SSDCache-clear test does not change it
Tearing, V-Sync off and no VRRImage split horizontally, with top and bottom showing different frames; feels responsive but looks broken in places; not a freezeV-Sync and VRR settings; high FPS makes tearing more frequent in appearanceA freeze or hitch is not tearing; tearing does not pause motion
Judder, V-Sync on and no VRR with FPS below refreshUneven cadence, for example 50 fps on a 60 Hz panel means some frames shown once and others twice; motion looks uneven even though captured frame times look fineCompare FPS with refresh rate; enable VRR or cap FPSFrame-time graph alone looks clean, because the problem is display cadence
VRR out of rangeGame runs above the monitor's VRR window, for example uncapped 160 to 190 fps on a 144 Hz panel: V-Sync latency returns, occasional tearing, less consistent feel than a capped rate inside the rangeCap FPS a few fps below refresh; common guidance is at least about 3 fps below; NVIDIA Reflex applies a similar cap automatically in supported games when VRR and V-Sync are onInside-range VRR removes most pacing problems but does not remove large shader spikes
Thermal throttlingClock speed drops and recovers on the graph; performance degrades over a session rather than improvingGPU and CPU temperature and clock overlaysShader stutter improves over a session; throttling worsens

Two misconceptions should be killed explicitly. First, high FPS does not equal smooth: frame pacing matters more than the average. Second, a stronger GPU does not fix stutter on its own: it only helps if the GPU is the limit, and shader compilation is CPU-side work. VRR is also not a cure here. Inside-range VRR removes most pacing problems, but a large shader spike remains visible even with VRR, because no display timing trick can hide a frame that simply is not ready.

Frame generation deserves a cross-reference, not confusion. DLSS Frame Generation and FSR Frame Generation raise the displayed FPS counter by inserting generated frames, but they do not speed up the game's simulation. If the base frame rate is unstable or low, roughly under 50 to 60 fps according to one source, generated frames inherit the unevenness and add input latency. Frame generation is not a fix for shader stutter.

What Actually Helps, in Order

  1. Let any first-launch compiling shaders step finish completely; do not skip it. Expect a longer first session.
  2. Play through new areas once and let the cache build over the first sessions, rather than judging a game, or a new driver, on its first run.
  3. Keep the driver shader cache enabled and at Driver Default or larger on NVIDIA; do not shrink it to save disk space unless disk pressure forces it, and expect recompilation afterwards.
  4. Enable Steam Shader Pre-Caching for Vulkan and OpenGL titles, noting that it does not cover DirectX 12.
  5. Cap FPS sensibly with VRR on, to separate pacing problems from compilation spikes. Use an in-game cap where available, which usually adds the least latency, or RTSS, which gives very even pacing at typically about one frame of latency according to one source.
  6. Only clear the shader cache when corruption is suspected, meaning stutter that never improves on repeat runs. After clearing, run a representative scene to rebuild the cache.
  7. Keep the game on an SSD. This helps streaming stutter, not shader stutter, and should not be sold as a shader fix.
  8. A faster CPU or more RAM helps at the margins — shorter spikes, less memory pressure — but is not a cure for a game-level precaching gap. Do not recommend a hardware purchase as the fix for such a gap.
  9. When to suspect hardware or service instead: stutter that never improves on repeat runs, appears in all games including old ones, or comes with crashes, visual corruption, or fan and temperature anomalies. That is no longer a first-run shader story and needs proper diagnosis covering GPU health, thermals, storage, and drivers. This article can point readers to a check-up; it must not promise a diagnosis from symptoms alone.

For inspection and advice, the workshop is at Dali Plaza Room 1209, Shek Mun. Telephone 2186 8010, WhatsApp 6558 6806. Any repair quotation is given as a final quote after inspection. The published three-tier range is HK$400–2,800, with the final quotation subject to inspection.

Frequently Asked Questions

Why does my game stutter after a driver update when it was smooth before?

Because the new driver can invalidate the previous shader cache, so shaders must be compiled again. Expect one rough session while the cache rebuilds, then repeated runs of the same areas should smooth out. Rolling back the driver usually invalidates the cache again, so it is rarely the right first response.

Does a faster graphics card fix shader stutter?

Only if the graphics card is the limit, which for shader stutter it is not. Compilation runs on the CPU, and the GPU waits for it. A faster CPU shortens the spikes, but even that does not remove them if the game has not pre-warmed the needed pipelines. Judge the problem by frame-time spikes and the repeat-run test, not by the GPU model.

Should I clear my shader cache regularly?

No. Clearing the cache is not routine maintenance. It should only be done when corruption is suspected, meaning stutter that never improves across repeated runs of the same area. After clearing, the next session will stutter while the cache rebuilds, so clear it deliberately and then run a representative scene to rebuild it.

Why does the game still hitch sometimes even after the compiling shaders screen?

Because precaching cannot practically cover every possible pipeline combination. A lighting, shadow, ray tracing, or effect combination the precache missed will still compile live on first encounter. That is a gap in that game's precache, not proof that the whole cache is broken, and it should fade on repeat runs.

How do I tell shader stutter from a CPU bottleneck?

Replay the same route twice. Shader stutter fades on the second run; a CPU bottleneck repeats, with GPU usage well below about 95 percent and little FPS change when resolution drops. Also check per-core CPU usage: one core at 100 percent can bottleneck a game even when overall CPU usage looks low.

Does an SSD fix shader stutter?

No. An SSD helps asset streaming stutter, because assets load faster, but it does not fix shader compilation, which is CPU-side pipeline building. Moving a game to an SSD is still sensible for streaming, but it should not be sold as a shader fix.

Is shader stutter a sign my graphics card is failing?

Not on its own. First-time-only hitches that fade on repeat runs and return after a driver update or cache clear fit the shader story. Hardware suspicion needs different signs: stutter that never improves, problems in all games including old ones, crashes, visual corruption, or temperature and fan anomalies.

Related articles

Related services

If you need Chip-level GPU repair, see these service pages:

HK GPU Fix 香港顯示卡專業維修,石門安達利廣場 1209 室,沙田,新界,HK。+852 21868010

← 返回 Blog | HK GPU Fix 首頁