Generated by GPT-5-mini| 3DNow! | |
|---|---|
| Name | 3DNow! |
| Designer | AMD |
| Introduced | 1998 |
| Architecture | x86 |
| Extensions | SIMD floating-point |
| Successors | SSE |
3DNow! is a single-instruction, multiple-data (SIMD) instruction set extension introduced by Advanced Micro Devices in 1998 for the x86 architecture, intended to accelerate floating-point vector operations for multimedia, graphics, and signal processing. It was offered as an enhancement to compete with contemporaneous extensions from Intel Corporation and aimed at workloads common to products from NVIDIA, ATI Technologies, Matrox, and workstation vendors like Compaq and Sun Microsystems. The technology was positioned in marketing alongside AMD microarchitectures such as AMD K6, AMD Athlon, and companies like Microsoft and Adobe Systems that produced software benefiting from vector math.
3DNow! extended the x86 instruction set to include packed single-precision floating-point SIMD operations, adding a set of new opcodes to benefit applications in 3D graphics, digital audio, and multimedia playback used by vendors such as RealNetworks, DivX, QuickTime, and Winamp. AMD announced the extension as part of competition with Intel Pentium II, and the initiative involved collaborations with software partners including Id Software and Epic Games to leverage accelerated inner loops in engines like those behind Quake and Unreal Tournament. The extension was included in product roadmaps alongside chipset and platform partners like VIA Technologies, SiS, and ALi Corporation.
3DNow! provided packed floating-point instructions operating on 64-bit MMX registers (aliasing MMX state) to perform operations such as add, subtract, multiply, reciprocal, and reciprocal square root, with instructions inspired by research from institutions like Stanford University and MIT. The design added support for data-parallelism similar to efforts in vector processors from Cray Research and accelerated algorithms commonly used in libraries from Intel Math Kernel Library, OpenGL, and codec implementations by Fraunhofer Society contributors. Implementation details intersected with microarchitectural features present in K7 cores, including pipeline balancing used by Intel Pentium III implementations and memory subsystem considerations found in AMD Hammer family designs.
3DNow! debuted in AMD processors such as the AMD K6-2 and was present in subsequent families including AMD Athlon, AMD Duron, and some Mobile AMD parts, with later support or lack thereof influencing adoption by Intel clones and licensees like Transmeta. Hardware vendors including Asus, Gigabyte Technology, MSI, and OEMs like Dell and Hewlett-Packard shipped motherboards and systems exposing 3DNow! through BIOS and microcode interactions. Competing architectures such as ARM architecture and PowerPC offered different SIMD approaches in chips from Apple and IBM, while embedded platforms from Texas Instruments and Analog Devices pursued distinct DSP solutions.
Benchmarks from independent reviewers such as Tom's Hardware Guide, AnandTech, PC Magazine, PC World, and ZDNet measured gains in specific workloads—3D rendering, texture mapping, and audio decoding—when applications or libraries used 3DNow!-optimized paths from vendors like NVIDIA and ATI Technologies. Games from id Software and Epic Games showed frame-rate improvements in some scenes, while codecs by DivX Networks and Xiph.Org benefitted in decode throughput; however, relative performance varied against Intel Pentium II and Pentium III processors using SSE extensions, with results influenced by compiler-generated code from GNU Project toolchains and commercial compilers from Microsoft Visual C++ and Intel C++ Compiler.
Compiler and toolchain vendors including the Free Software Foundation, Intel Corporation, Microsoft, and Borland added intrinsic support or code generation options to exploit 3DNow! in libraries such as glibc, multimedia frameworks like FFmpeg, and game engines used by Valve Corporation and LucasArts. Audio and video applications from RealNetworks, Apple QuickTime, and Windows Media Player incorporated optimized routines, while middleware projects like SDL and graphics stacks such as Mesa 3D integrated fallback or accelerated paths. Open-source projects hosted by communities around SourceForge and GitHub included assembler-optimized kernels and patches for 3DNow! throughout the early 2000s.
Adoption of newer SIMD extensions like SSE and SSE2 from Intel Corporation and the standardization around 128-bit XMM registers shifted industry momentum away from 3DNow!, and later AMD microarchitectures prioritized compatibility with SSE and subsequent instruction sets used by Intel Core families. Despite decline, 3DNow!'s role influenced cross-vendor discussions at standards forums involving The Khronos Group and techniques used by numerical libraries at Lawrence Livermore National Laboratory and influenced instruction design considerations later seen in AVX and NEON. Retrospectives by journalists at Ars Technica, historians at Computer History Museum, and engineers from AMD and Intel recognize 3DNow! for its early push on SIMD floating-point acceleration and its practical impact on multimedia experiences in the late 1990s and early 2000s.
Category:SIMD instruction sets