{
 "name": "NerdBench Full",
 "tagline": "Little benchmark application for Apple Silicon and Metal/Vulkan",
 "description": "Little benchmark application for Apple Silicon and Metal/Vulkan.\n\nNerdBench puts your Mac's GPU and CPU through real rendering and compute work – ray tracing, path tracing, rasterization and matrix maths – on Metal, Vulkan and the CPU, so you can compare them side by side and see where the performance comes from.\n\nTHE TESTS\n\nCompute\n• SGEMM – single-precision matrix multiply, in GFLOP/s\n• Matrix Transform 4×4 and Matrix Chain 4×4 – the vertex and hierarchy maths behind every 3D engine\n• Vector Math ALU – normalize, dot, cross, reflect and FMA at full load\n• Memory Bandwidth – streaming reads and writes over a large buffer, in GB/s\n• Parallel Reduction – tree reduction through threadgroup memory\n\nRasterization\n• A textured, lit 3D scene of more than 134,000 triangles, at up to 3840×2160\n• Fillrate / Overdraw – heavy blended overdraw, in GPix/s\n\nRay tracing and path tracing\n• Each in two versions: software BVH traversal that runs on all three backends, and Metal ray tracing with acceleration structures refitted every frame, using the dedicated ray tracing hardware on M3 and later\n• The path tracer computes Monte Carlo global illumination: diffuse bounces, GGX metals, clear and frosted glass with refraction, emissive surfaces and sky light. There is no denoiser – every sample is really traced.\n\nTHE SCENE\n\nA detailed 1990s desktop computer stands in a mirrored room. Three colored lights orbit it while twenty-five glass, plastic, metal and copper objects drift and spin the other way, and the camera circles the room. Turn the animation off and the path tracer refines a still image sample by sample, right in front of you.\n\nTHREE BACKENDS, ONE WORKLOAD\n\n• Metal – the Mac's native graphics API\n• Vulkan – through MoltenVK, to compare a portable API with Metal on the same GPU\n• CPU – multithreaded NEON code with a software rasteriser and the same ray and path tracers, on as many cores as you choose\n\nAll three share the same algorithms, shading and scoring, so a result on one compares directly with the others. Every test runs on every backend, except the two that need Metal ray tracing.\n\nHONEST NUMBERS\n\n• GPU time and end-to-end time for every test, so CPU-side overhead is shown rather than hidden\n• Average, minimum and 1% low frame times, and the number of iterations measured\n• GPU timestamps that cannot be right are discarded, and the run tells you when that happens\n• An overall index: the geometric mean of every test that passed\n\nYOU'RE IN CONTROL\n\n• Pick any set of tests, and set the measure window, warm-up and matrix size\n• Raster resolution up to 3840×2160, ray and path tracing up to 1920×1080\n• 1 to 1024 samples per pixel and up to 32 bounces\n• Ray-traced shadows, animation and live preview, each on or off\n• A live preview of the rendering tests and a frame-time graph\n• Export results as JSON or CSV\n• A command-line mode for scripted runs\n\nPRIVATE BY DESIGN\n\nNerdBench has no account and makes no network connections: no analytics, no ads, no tracking. Results stay in the app until you export them.\n\nFor repeatable results, connect the power adapter, turn off Low Power Mode and quit other apps before a run.\n\nAmiga 4000D object from: https://3dwarehouse.sketchup.com/model/032de160-6e4d-46df-b41c-258f2582fb1e/Commodore-Amiga-4000D\nAuthor of object: Sebastien G.\n\nFull version with memory, AI operation and CPU speed tests.",
 "url": "https://debuts.app/apps/nerdbench-full",
 "category": "Utilities",
 "platforms": [
  "iPhone and iPad"
 ],
 "pricing": "paid",
 "use_cases": [],
 "audience": [],
 "links": {
  "app_store": "https://apps.apple.com/us/app/nerdbench-full/id6815858285?mt=12",
  "google_play": null,
  "website": null
 },
 "icon": "https://is1-ssl.mzstatic.com/image/thumb/Purple211/v4/45/b7/85/45b785e9-aadc-ef7f-dd37-d39317eb5e3d/NerdBench.png/512x512bb.png",
 "screenshots": [
  "https://is1-ssl.mzstatic.com/image/thumb/PurpleSource211/v4/13/92/3d/13923ddb-4361-b26b-0aad-a1390abd8c26/Screenshot_2026-09-28_at_22.43.56.png/800x500bb.jpg",
  "https://is1-ssl.mzstatic.com/image/thumb/PurpleSource221/v4/2c/28/4f/2c284f8c-6e32-b1bf-5635-f7890ddd9400/Screenshot_2026-09-28_at_22.44.03.png/800x500bb.jpg",
  "https://is1-ssl.mzstatic.com/image/thumb/PurpleSource221/v4/9e/f1/12/9ef112f3-95a7-4ba7-391c-fcbbab55149f/Screenshot_2026-09-28_at_22.44.30.png/800x500bb.jpg",
  "https://is1-ssl.mzstatic.com/image/thumb/PurpleSource221/v4/b6/09/9e/b6099ec0-0ce6-b46a-c3ef-ba7e3aa9ecd8/Screenshot_2026-09-28_at_22.46.17.png/800x500bb.jpg"
 ],
 "claimed": false,
 "verified": false,
 "maker": "Dariusz Wisniewski",
 "upvotes": null,
 "followers": null,
 "rating": null,
 "ratings": 0,
 "awards": [],
 "launches": [],
 "ai_ready": 78,
 "badges": {
  "launched": "https://debuts.app/badge/nerdbench-full?type=launched",
  "award": "https://debuts.app/badge/nerdbench-full?type=award",
  "rating": "https://debuts.app/badge/nerdbench-full?type=rating"
 }
}