What changed: In September 2026 two benchmarks of the same game, Assassin’s Creed Mirage, produced opposite headlines. One recorded the minimum frame rate rising from 51 to 90 FPSframerate. How many images (frames) the game shows per second; higher = smoother motion. 60 fps is a common target. after the protection was fully removed. The other, on a different CPU, found the protected build within a few frames of the unprotected one. Why it matters: The honest answer to “does Denuvo hurt performance” is not yes or no. It is where the checks were placed and which thread they run on — which is why the same technology measures as a 39 FPS swing in one test and as noise in the next. Who should care: Anyone treating the “Denuvo anti-tamper” badge on a Steam page as an automatic reason to wait, and anyone reading a removal-patch benchmark as proof of what the DRM cost.
Denuvo Anti-Tamper is software most players never see running, yet almost everyone has an opinion about its frame cost. Its maker, Irdeto, states plainly that protection “runs outside gameplay-critical paths, such as startup and loading, with no noticeable impact during play.” A decade of community benchmarks says something closer to the opposite. In 2026, tests finally got strong enough to examine the disagreement directly.
The two 2026 Mirage benchmarks disagree, and that is the finding
The strongest test of Denuvo’s cost is a devirtualization: a build with the obfuscation machinery physically stripped out, rather than a crack that only tricks the protection into staying quiet. One of those appeared for Assassin’s Creed Mirage in September, removing both Denuvo and VMProtect.
GameGPU reported the results on an Intel Core i7-13700K, a GeForce RTX 5080, 64 GB of RAM and a Samsung 990 Pro, using the game’s built-in 1080p benchmark on standard settings; PCFormat identifies the tester as the YouTube channel Kevin’s Tech Vids. The gaps were large. Minimum FPS rose 76%, from 51 to 90; the 1% low went from 93 to 128; the 0.1% low from 74 to 105; the average rose 14%, from 162 to 186. The maximum barely moved, 232 to 233. The executable also shrank from 496 MB with Denuvo and VMProtect in place to 102 MB without them.
Days later, DSOGaming ran its own comparison and got almost nothing. On a Ryzen 9 7950X3D, 32 GB of DDR5DDR5. A generation of system RAM (memory); newer and faster than DDR4.-6000 and an RTX 5080, at Ultra High with Adaptive Quality off, the two builds ran together at 4K and 1080p. At 720p — where the site deliberately pushed work onto the CPU, “and if Denuvo and VMProtect were hurting performance, this is where I should have seen it” — the protected Ubisoft Connect build managed a 79 FPS minimum and 212 FPS average against 82 and 218 for the cracked build. DSOGaming found traversal stutters in both, concluded the crack “runs EXACTLY like the official Ubisoft Connect version,” and wrote that it was “skeptical of that one video” among the tests it had seen online.
Two 2026 tests, one game, opposite conclusions. The differences that matter are the CPU (a 13700K against a 7950X3D), the resolution and preset, and the fact that the second test was hunting for the effect in the most CPU-bound configuration it could build. Neither outlet is obviously wrong, which is what makes the pair more useful than either number alone.
What the protection is actually doing inside the binary
Denuvo Anti-Tamper, which Irdeto now markets as Video Game Anti-Piracy, is a licensing and obfuscation layer wrapped around a game’s executable. Cheat Code Central’s explainer, drawing on a reverse-engineering analysis, describes the mechanism: selected functions are pulled out of the protected binary and re-executed inside a custom virtual machine, with pieces of certain instructions physically removed from the compiled code. Those removed pieces — “stolen constants” — are reassembled at runtime from a license file generated from the machine’s hardware identifiers.
The frame-time argument comes from how that layer behaves at runtime. Tech4Gamers reported on an analysis by YouTuber Nathan Baggs, corroborated by a programmer using the handle @valigo, describing the VM’s run-time compilation introducing junk jumps and obfuscation that make the CPU’s instruction pipeline and branch predictor work harder, undoing the game’s optimized loops and memory access. Tech4Gamers’ own framing is worth keeping: the analysis claims the recompiled code “reportedly invalidates CPU cache,” and the findings “have not been officially explained by the creators of Denuvo.”
One thing the technical accounts agree on is that this is not Denuvo Anti-Cheat. That kernel-level product is built for competitive multiplayer detection and runs only during a match; the anti-tamper layer is a separate system for storefronts. Cheat Code Central notes that the kernel-driver backlash aimed at the anti-cheat tool gets applied to the anti-tamper layer constantly — a distinction worth holding if you have read our breakdown of kernel anti-cheat architectures, where the ring-0 story concerns the other product entirely.
The measured cases, side by side
Every figure below belongs to the outlet or tester that produced it, and the comparisons are not equivalent — some remove Denuvo, some replace it, and some are before-and-after patch notes.
| Case (tester) | What was compared | Framerate result | Source |
|---|---|---|---|
| Resident Evil 4 Remake | Denuvo against the Enigma DRM that replaced it, 1440p, i5-13400 + RTX 4060 Ti | “a difference of over 20 FPS in some cases” — but against a new DRM, not against no DRM | PC Guide |
| Assassin’s Creed Mirage | Denuvo + VMProtect 1.1.1 against a fully devirtualized build, 1080p, i7-13700K + RTX 5080 | Minimum 51 → 90 FPS; average 162 → 186 FPS | GameGPU / Kevin’s Tech Vids |
| Assassin’s Creed Mirage | Ubisoft Connect against the cracked build, 720p CPU-limited, 7950X3D + RTX 5080 | Minimum 79 vs 82 FPS; average 212 vs 218 FPS — no clear gain | DSOGaming |
| DOOM: The Dark Ages | Denuvo removal in the Revelations patch, 4K path tracing, RTX 5090 | Average unchanged at 30 FPS; minimum fell 27 → 24 FPS | GameGPU |
| Metro Exodus | Denuvo against the unprotected build, DirectX 11 | Average 122.4 → 123.7 FPS, a 1.06% gain | Overlord Gaming |
| Prey | Denuvo against the DRM-free GOG build | Average 148.9 → 180.4 FPS, a 21.16% gain | Overlord Gaming |
| Detroit: Become Human | Denuvo against the post-removal Epic build | Average 81.1 → 81.8 FPS — under one frame | Overlord Gaming |
| Ghostwire Tokyo | Boot time with and without Denuvo | 200 seconds with, 54 seconds without | YetTea, via Cheat Code Central |
| Shadow of the Tomb Raider | Removal patch, low settings | “reportedly added 8 to 17 FPS” | Cheat Code Central |
| Final Fantasy XV | Denuvo on against off | Little meaningful impact; some scenes marginally faster with it | Peter “Durante” Thoman, via Cheat Code Central |
The spread is the answer, not a failure of the table. Overlord Gaming’s own summary of its Metro Exodus numbers was that the two builds “traded blows,” and it found the largest gaps in titles where the protection had been integrated least carefully. TechPowerUp added a data point from the other direction in April 2026: Denuvo’s reputation for performance regressions is real, it wrote, “even though we did not find this to necessarily be true in our testing in 2019.”
Where the evidence is most consistent: the clock, not the frame
Framerate is where the disagreement lives. Loading and memory are where the numbers line up far more often, and they track the mechanism — a startup-time layer that decrypts constants and verifies hardware costs seconds before the first frame, which is exactly the window Irdeto says it targets on purpose.
| Case (tester) | Metric | Result | Source |
|---|---|---|---|
| Rime | First startup / subsequent startups | 50% longer on the first run, 65.8% longer after that | Overlord Gaming |
| Death of the Outsider | Benchmark mission load | 80.88% slower with the protection in place | Overlord Gaming |
| Metro Exodus | Main-menu load on repeat launches | Protected build took roughly 2.9× as long | Overlord Gaming |
| Resident Evil 4 Remake on Steam Deck | System RAM in use | “over 1GB more” after the Enigma switch, per Deck Madness | PC Guide |
The memory row is the one to read carefully. That increase belongs to Enigma, the DRM Capcom moved to when it dropped Denuvo — PC Guide reported a swap, not a removal, and noted Capcom never announced it, with the change visible in SteamDB update logs. PC Guide also notes that its comparison is imperfect, since the original benchmarks it draws on came from a 2024 test.
Denuvo’s own account, and what it conceded
Irdeto’s position is not that the protection is free; it is that the cost is a placement problem. Product Manager Andreas Ullmann’s April 2026 blog sets out the company’s numbers: protection is “typically a very small part of the overall game binary, often around 300–400 functions in a binary that may contain hundreds of thousands,” and “in most games, a large share of protection-related execution is intentionally placed during startup (often the majority, around 75–80%).” The argument is that functions, threads and execution windows decide what a player feels, that placement must be validated in the shipped binary rather than inferred from source, and that integrations need re-profiling after major updates because what was safe in one build may overlap differently in the next. Irdeto says it shares its internal runtime-analysis tool with clients.
The product page states it bluntly: protection “runs outside gameplay-critical paths, such as startup and loading, with no noticeable impact during play,” with function profiling that “avoids performance-critical paths.” A December 2025 Irdeto blog goes further, arguing that modern security “when implemented correctly, does not meaningfully impact performance,” and citing an unnamed well-known cracker who now describes modern protection as “rarely affects performance” and “barely noticeable.”
When the hypervisor bypass landed in April 2026, Irdeto’s head of communications, Daniel Butschek, told TorrentFreak that strengthened versions were coming and that “for players, performance will not be compromised by these strengthened security measures,” adding that addressing the bypass “will not require Denuvo to move into Ring -1 or deeper kernel level.” Tom’s Hardware framed that against history, noting the performance controversy was “vehemently denied by Denuvo and subsequently mocked online, as cracked versions ran far better.” The promise covers future builds, though, and no benchmark of them exists yet.
Where it shipped, and shipped well
The evidence is not one-sided.
Pearl Abyss added Denuvo to Crimson Desert’s Steam page a week before its March 2026 launch, then argued players had already seen the final cost. “The benchmark videos and performance specs we released were all created with the exact same implementation of denuvo that is in the launch build,” a representative told journalist Paul Tassi, per GamesRadar+. “This includes the performance videos by Digital Foundry.” GamesRadar describes that footage as running at ultra settings in native 4K at “mainly” 60fps on a Radeon RX 7900 XTX.
Hogwarts Legacy has the most detailed third-party analysis of any single title. Maurice Heumann, who spent five months reverse-engineering the protection, instrumented his own hooks to log every Denuvo intervention in execution. His write-up found the interventions were not per-frame: Denuvo interacted “once every few seconds,” sometimes not at all, and the log only accumulated “when major things happen, scene switches, loading screens or similar.” He concluded he was “not able to measure any relevant performance overheads caused by it” and that the protection “seems to be very well integrated into the game.” He flagged his own limits: the measurement is of frequency rather than magnitude, and reflects Hogwarts Legacy only.
Then there is the removal that made things worse. GameGPU benchmarked DOOM: The Dark Ages at 4K with path tracing on an RTX 5090 and found the average frame rate unchanged at exactly 30 FPS after Denuvo was stripped out, with the minimum falling from 27 to 24 FPS in demanding scenes. The site’s reading was that the drop pointed at “optimization issues on the part of the developers,” not the DRM.
Why a removal patch is not a controlled experiment
The commonest mistake is treating a post-removal frame-rate jump as Denuvo’s bill. Cheat Code Central’s caution is the cleanest statement of why: “a frame-rate boost after a Denuvo removal patch doesn’t prove Denuvo caused the original slowdown. Removal patches often ship bundled with unrelated engine and shader fixes.” The reverse case applies too, which is why the Mirage devirtualization matters — it is one of very few tests where the protection code itself was the variable.
It also changes which number to read. Averages hide what players feel; the 1% and 0.1% lows, and frame pacing, are where a CPU-side check shows up as a hitch rather than a lower average. Our frametime metrics glossary covers what those metrics measure, and Mirage is a good illustration: the attention-grabbing version of that story is entirely about minimums and lows while the maximum moved by one frame.
What to take from this
Denuvo is neither free nor fatal, and 2026’s evidence supports both halves of that. The mechanism is real: an obfuscating VM whose runtime checks can undo cache-friendly code, which is why badly placed checks show double-digit minimum-FPS swings and why loading times are the most reproducible cost. The variance is real too, and better attributed to integration quality than to the product’s existence. TechPowerUp notes the aggravating factor publishers keep supplying: studios “have a habit of adding Denuvo to games at the last minute before launch,” so the version benchmarked in previews is not always the version that ships. Crimson Desert’s defence was, in effect, that it had avoided exactly that.
If you are deciding whether to buy at launch, look for a specific kind of evidence rather than a general reputation: a simultaneous comparison, a devirtualized test, a dev statement that the preview build carried the protection, or a 1% low figure from a CPU-bound resolution. Absent one of those, “Denuvo is in this game” tells you less about your frame time than which CPU you own and where the studio put the checks.