AMDTOOLBOX

Official AMD driver. WattMan, Radeon Overlay (Alt+R), FSR toggle, AFMF 2.1, Anti-Lag 2, HYPR-RX, Noise Suppression, Record & Stream. Supports RX 6000–9000.

Complete GPU driver removal — no leftovers in registry, folders or driver store. Run in Safe Mode. Essential before GPU vendor change (AMD↔NVIDIA).

AMD's official cleanup tool. Runs in normal Windows mode — less thorough than DDU, but recommended before Adrenalin downgrade. Removes registry entries and driver remnants.

Automatically detects installed GPU and installs matching Adrenalin driver via AMD Software Installer. No manual GPU model search. Requires .NET Runtime.

Workstation/enterprise driver. ISV-certified for DCC software (Blender, Maya, Autodesk). Also works on consumer RX cards — DDU recommended before switching.

De-facto standard GPU OC tool. Core/Mem offset, VF curve (Ctrl+F), power limit, fan curve, OSD via RTSS. Voltage OC on RDNA 2/3 restricted by AMD SPPT lock — fan & clock offset work fine.

Lightweight WattMan replacement by Igor's LAB / hellm. Direct connection via AMD ADLX library. Core clock, mem clock, voltage, fan curve — runs without Adrenalin software, driver only. Portable.

Completely rewritten in C++/WinRT. 5 sections: OC, Voltage, Fan, Monitoring, Settings. No telemetry, no background process. RDNA4 compatible since April 2025 (ADLX 1.4). ImageSharpening feature included.

Sapphire's own OC tool. Optimized for Nitro+, Pulse, Pure. Core/Mem clock, voltage, fan curves. Best RGB control for Sapphire Nitro+ models. Broad AMD GPU compatibility.

ASUS OC tool for ROG Strix, TUF Gaming, Dual models. OC Scanner (auto-OC), voltage/clock tuning, Aura RGB sync. Fully featured for ASUS cards — for other brands MCT is recommended.

Power user standard for RDNA 1/2. Edits SoftPowerPlayTables: PPT, TDC, EDC, voltage/frequency tables, fan curves. Restricted since Adrenalin 2020 by AMD SPPT lock. RDNA 3/4: PSP lock — no support.

OC/undervolt section integrated in Adrenalin. Power Limit Slider (+PPT), frequency offsets, GPU Voltage Tune (% or mV). Primary undervolting method for RDNA 3/4 — RDNA 4 has especially large UV headroom.

Lightweight tool for AMD GPUs via ADL Overdrive registers (no SPPT-level intervention). Well suited for older RDNA 1/2 cards. Superseded for newer cards by MCT.

Changes GDDR6/HBM memory timings on-the-fly via CLI. MMIO register control. Timing tables for common chips in the GitHub repo. Resets after reboot — autostart script for persistence.

The reference for GPU info. GPU revision, shader units, bus width, exact VRAM chip vendor (important for memory OC!), VBIOS version + backup. Sensor tab for real-time data. Render test for quick check.

Most complete hardware monitor. GPU Hotspot/Junction temp (110°C = normal for AMD!), VRAM temp, PPT/TDC/EDC values, all AMD-specific sensor channels. CSV logging for OC tests. HWiNFO plugin for RTSS OSD.

Standard for in-game OSD. FPS, temps, utilization, frametimes as overlay. With HWiNFO plugin: all AMD sensor data in OSD. Precise frame limiter on board. Stealth mode for anti-cheat.

AMD's built-in in-game overlay. Open: Alt+R. GPU clock, temp, VRAM utilization, FPS, frametimes. Performance Metrics Widget pinnable to desktop. FPS limit settable directly in overlay. No extra download.

Stability tester with automatic error detection. GPU:3D mode detects artifacts and GPU errors live. PSU stress test combines CPU+GPU for maximum system load. Monitoring dashboard on board. Freemium.

Creates comprehensive system report for driver issues/crashes. Collects TDR logs, dump files, system configuration, driver version. Essential when submitting AMD support tickets or Reddit diagnostics.

The benchmark standard. TimeSpy Extreme (4K DX12), FireStrike Ultra (DX11), Port Royal (RT), Speed Way (DX12 RT). Online leaderboards for GPU comparison. Free: TimeSpy + FireStrike Basic via Steam.

Extreme GPU stress test. Drives GPU to maximum temperature — good for cooling check and stability test after OC. 15 minutes is sufficient. Set temp limit in WattMan as safety net. ⚠ No extended operation!

Real-time 3D benchmark with actual game engine. 8K preset uses up to 8 GB VRAM — ideal for memory OC stability test. 1080p Extreme for CPU-limited systems. Heaven/Valley as free older alternatives.

VR readiness test. Orange Room (DX11 baseline), Cyan Room (DX12), Blue Room (high-end 4K VR equivalent). No VR headset required. Good supplementary benchmark to 3DMark for GPU performance comparison.

Best tool for frametime analysis. ETW-based (no DLL inject). 1%/0.1% Low, percentile plots, sync analysis, OSD, CSV/JSON export. Before/after comparison for OC tests. AV exception recommended for game folder.

Microsoft's ETW-based frame capture tool. Basis for CapFrameX and OCAT. CLI: PresentMon.exe -output_file log.csv -process_name game.exe. GPU busy time, display latency. Open source on GitHub.

AMD's own performance capture tool. Frametimes, GPU utilization tracking, OSD. Based on PresentMon — easier entry than CLI. Output compatible with CapFrameX. Open source on GitHub (GPUOpen-Tools).

Reduces system input latency by syncing CPU frame before GPU rendering. Anti-Lag 2 (API level): requires explicit game support. Anti-Lag 1 (driver level): all games, more overhead. RDNA 4 benefits most.

Frame generation + upscaling for any game via DX hook — no native game support required. LSFG 3.0 with Async FG. ⚠ Increased input latency: not suitable for competitive. Base: ≥60 native FPS recommended.

ML-based upscaling — exclusive to RDNA 4 (RX 9000 / Navi 4x). FSR 4.1 via Adrenalin 26.3.1. Over 60 titles. Adrenalin override for non-native games. Levels: Quality / Balanced / Performance / Ultra Performance.

Spatial + temporal upscaling. Cross-vendor: also runs on NVIDIA and Intel! More frequently integrated in games than FSR 4. FidelityFX SDK on GitHub. Recommended upscaling method for RX 6000/7000.

Driver-level frame generation for all DX11/12 games — no native game support required. AFMF 2.1 improves artifact suppression. ⚠ Do not combine with V-Sync — use RTSS frame limiter instead. Base: ≥60 FPS.

AMD's one-click performance profile. Enables FSR, AFMF, Anti-Lag and Radeon Boost simultaneously. Optimized presets for thousands of games. Overrideable per game. Ideal for maximum FPS without manual tuning.

Community mods replacing DLSS with FSR or upgrading FSR versions in games. Known mods: Cyberpunk 2077 FSR3, Elden Ring, Jedi Survivor and more. ⚠ Always backup original DLLs first!

AMD's official VBIOS flash tool. CLI + GUI. Supports RX 6000–9000. ⚠ Backup first: amdvbflash -s 0 backup.rom. Ring-0 driver (AMDVBFlashDriverInstaller.exe) keep active only temporarily — security risk!

Community-modified AMDVBFlash. Bypasses PSP signature locks on RDNA 2 for power limit increases via VBIOS modding. RDNA 3/4: PSP lock generally prevents modding entirely. Read OC.net thread first!

Largest online archive for GPU VBIOS files. RX 6000–9000 available. ⚠ GPU model AND board partner (e.g. Sapphire Nitro+ vs. XFX Speedster) must match exactly — check SSID in GPU-Z!

Patches AMD drivers: removes DVI/HDMI pixel clock limits (165/330 MHz), BIOS signature verification. For custom resolutions and modded BIOS. ⚠ Vanguard & FACEIT anti-cheat block patched drivers!

Writes MPT modifications directly into GPU BIOS — permanently effective without driver dependency. Workflow: MPT → create SPPT profile → RBE → write into BIOS. Developed by R.B.R.T. (Red BIOS Rebellion Team).

BIOS editor for RX 400/500 series (Polaris). Copies memory straps (timing tables) from faster to slower levels — for Elpida/Micron/Samsung/Hynix chips. Not relevant for RDNA cards.

RDNA 1/2 bug: GPU reset after driver crash fails (black screen, reboot needed). Windows: increase TdrDelay/TdrDdiDelay in registry. Linux: vendor-reset kernel module. RDNA 3/4 generally not affected.

Dynamic FPS limiter — reduces GPU load during calm in-game scenes (menus, static moments). Min/Max range adjustable (e.g. 30–165 FPS). GPU temp drops 10–20°C. Good for quiet operation.

Dynamically lowers render resolution during fast camera movements for more FPS. Resolution drops to 50% — less visible at 1440p+. Combines well with FSR upscaling. Configurable per game.

Resizable BAR allows CPU access to entire GPU VRAM. Enable in UEFI: Above 4G Decoding + Resizable BAR. Check: GPU-Z → Bus Interface → BAR1 Size = VRAM size. 5–15% performance gain depending on game.

Powerful DirectX hook tool. HDR tonemapping (SDR→HDR), precise frame pacer (especially for AMD CPU+GPU combo), latency minimization, texture mods. Inject mode or Global Service mode. Open source on GitHub.

Creates custom resolutions and CVT timing profiles. Ideal for monitor OC (e.g. 160→170 Hz), 10-bit output tweaks, EDID modifications. restart64.exe for safe EDID restart without bluescreen. By ToastyX.

Contrast-enhancing sharpening algorithm directly in the driver. ~1% GPU load. Recommendation: 80% strength as starting point. Very effective in combination with FSR upscaling. Adjustable in real time via Alt+R overlay.

AI-based microphone noise suppression via GPU inference (~1–2% GPU load). Enable: Adrenalin → Audio. Then select microphone in Windows sound settings as 'AMD Noise Suppression'. Fix for Enable bug in 26.3.1.

Node-based Stable Diffusion frontend. Runs on AMD GPUs via DirectML (Windows) or ROCm (Linux). FLUX, SDXL, SD 1.5 — all models. Best performance on AMD with ROCm build. Most active community of all SD UIs.

Optimized AUTOMATIC1111 fork by lllyasviel. DirectML backend for AMD GPUs on Windows. Faster than A1111 thanks to attention mechanism optimizations. FLUX, SDXL, SD 1.5. ControlNet included.

The original Stable Diffusion WebUI. AMD support via DirectML (--directml flag) or ROCm on Linux. Huge extension ecosystem. For pure AMD Windows users: Forge usually faster. A1111 for extension compatibility.

AMD's 1-click local AI environment — integrated directly in Adrenalin (since 26.1.1). Automatically sets up everything needed for local AI workloads: Python, DirectML, libraries. No manual setup. Windows only.

The two most important AI model sources. Civitai: community checkpoints, LoRAs, embeddings — directly for ComfyUI/Forge/A1111. Hugging Face: FLUX, SDXL, LLMs, all official releases. Both GPU-agnostic.

State-of-the-art open-source AI upscaler. 2×, 4×, 8× scaling, various models. On AMD GPUs via DirectML backend (ncnn variant) or directly in ComfyUI/Forge/A1111 as extras upscaler. Standalone CLI + GUI wrapper available.

Node-based image processing tool focused on upscaling. ESRGAN, SwinIR, RRDB, Real-ESRGAN — all models locally. AMD GPU acceleration via DirectML. Batch processing, flexible pipelines. Ideal for mass upscaling of images.

Professional AI image enhancement suite. Gigapixel AI: up to 6× upscaling, best quality. Photo AI: denoise + sharpness + super resolution combined. Video AI: upscaling, slow-mo, stabilization. AMD GPUs supported via DirectML.

Community database for AI upscaling models (ESRGAN, SwinIR, SPAN, RRDB and more). Filterable by architecture, scale factor and target type (photos, anime, textures, screenshots). Directly usable with chaiNNer and ComfyUI.

AMD's CUDA equivalent for GPU compute. Basis for PyTorch, TensorFlow, Stable Diffusion on AMD under Linux. Supports RDNA 2/3/4. RX 6700+ officially supported — older cards via unofficial patches. Windows: only via WSL2.

AMD's HIP SDK for Windows — brings hipBLAS, hipFFT, MIOpen and other ROCm libraries to Windows. Enables CUDA→HIP ported workloads. Basis for local LLM inference tools with AMD GPU acceleration on Windows.

Simplest LLM inference tool. AMD GPU acceleration via ROCm (Linux) and experimental under Windows. Llama 3, Mistral, Phi-3, Gemma, Command-R and more. 1-line install + model download. REST API for custom apps.

GUI tool for local LLM inference. AMD GPU acceleration via DirectML on Windows (no ROCm required). Supports GGUF models from Hugging Face. Visual chat interface, model management, local API server. Ideal for AMD Windows users without ROCm setup.

LLM inference via llama.cpp with Vulkan backend — works on all AMD GPUs without ROCm. GGUF models. Vulkan backend: universal, slightly less optimized than ROCm. Ideal for RX 5000 and older cards without official ROCm support.

AMD's graph inference engine and deep learning primitive library. MIGraphX: optimizes ONNX models for AMD GPU inference (BERT, ResNet, transformer models). MIOpen: CUDA cuDNN equivalent, basis for PyTorch/TensorFlow on ROCm.