x86
x86 is a Complex Instruction Set Computing (CISC) processor architecture originally introduced by Intel in 1978 with the 8086 processor and subsequently extended by multiple vendors including AMD. Core characteristics include a strong emphasis on backward binary compatibility, variable-length instruction encoding, and architectural evolution from 16-bit to 32-bit (IA-32) and then to 64-bit (x86-64). The x86 architecture has long dominated the personal computer and server markets and remains the prevailing instruction set foundation for modern desktop and data center computing.
Core Facts
On x86 architectures, SIMD evolved from SSE (128-bit) to AVX-512 (512-bit), while ARM has NEON and SVE instruction sets
x86 uses a Complex Instruction Set Computing (CISC) design with variable instruction lengths, while ARM uses a Reduced Instruction Set Computing (RISC) design with fixed instruction lengths; the machine code of the two architectures is mutually incompatible.
The current mainstream instruction set architectures include x86/x86-64 (Intel/AMD), ARM, RISC-V, MIPS, and PowerPC.
Timeline (last 90 days)
Windows on Arm has long not been a first-class citizen of CUDA, and the available versions and distribution cadence of the corresponding CTK lag behind x86.
Apple Virtualization.framework can achieve near-native performance when running ARM-architecture Windows or Linux guests, but its support for x86 guests is limited.
When running AI workloads on mobile workstations, common benchmarks include token generation speed (tokens/s), first-token latency, and memory usage under specific models; different architectures (x86 + discrete GPU vs ARM unified memory) emphasize different aspects of energy efficiency and peak performance.
Modern x86 processors provide the SSE and AVX instruction set families, and ARM provides NEON, which in theory can increase a software renderer's pixel processing throughput severalfold.
Jensen Huang pointed out that the design logic of hyperscale CPUs in the x86 era was 'being rented,' where the more cores the better, because the unit of pricing was 'how much money per core.'
On the x86 platform, the 8B model still goes through the generic portable loop in arch-fallback.h, which is the root cause of the sharp performance drop for large models on x86.
The Vera CPU achieves 1.8 times the performance of x86 solutions under specific AI workloads.
Jensen Huang believes that hyperscale CPUs (x86) were originally designed to be rented, whereas AI does not rent cores, AI generates Tokens and sells Tokens.
On the x86 architecture, the CPU and GPU typically share a cache-coherent view, while ARM SoCs handle cache coherency between the CPU and GPU in varying ways.
The author proposed that the next extension direction is to measure the ability of models to optimize code separately for ARM and x86 assembly architectures.
27 more timeline events
All Facts (20)
On x86 architectures, SIMD evolved from SSE (128-bit) to AVX-512 (512-bit), while ARM has NEON and SVE instruction sets
65%Verifiedx86 uses a Complex Instruction Set Computing (CISC) design with variable instruction lengths, while ARM uses a Reduced Instruction Set Computing (RISC) design with fixed instruction lengths; the machine code of the two architectures is mutually incompatible.
65%VerifiedThe current mainstream instruction set architectures include x86/x86-64 (Intel/AMD), ARM, RISC-V, MIPS, and PowerPC.
65%UnverifiedSTOSB指令（操作码0xAA）将AL寄存器的值写入ES:DI指向的内存地址并自动递增DI，仅需1字节编码
95%Unverifiedx86实模式下物理地址计算公式为：段寄存器值 × 16 + 偏移量，最大寻址空间为1MB
95%UnverifiedIntel 80386是第一款32位x86处理器
95%Unverified"Wake up! 16b"是一个仅用16字节x86机器码实现的demoscene作品，由知名sizecoder Hellmood创作
85%UnverifiedWindows on Arm has long not been a first-class citizen of CUDA, and the available versions and distribution cadence of the corresponding CTK lag behind x86.
50%UnverifiedApple Virtualization.framework can achieve near-native performance when running ARM-architecture Windows or Linux guests, but its support for x86 guests is limited.
50%UnverifiedWhen running AI workloads on mobile workstations, common benchmarks include token generation speed (tokens/s), first-token latency, and memory usage under specific models; different architectures (x86 + discrete GPU vs ARM unified memory) emphasize different aspects of energy efficiency and peak performance.
50%UnverifiedModern x86 processors provide the SSE and AVX instruction set families, and ARM provides NEON, which in theory can increase a software renderer's pixel processing throughput severalfold.
50%UnverifiedJensen Huang pointed out that the design logic of hyperscale CPUs in the x86 era was 'being rented,' where the more cores the better, because the unit of pricing was 'how much money per core.'
50%UnverifiedOn the x86 platform, the 8B model still goes through the generic portable loop in arch-fallback.h, which is the root cause of the sharp performance drop for large models on x86.
50%UnverifiedThe Vera CPU achieves 1.8 times the performance of x86 solutions under specific AI workloads.
50%UnverifiedJensen Huang believes that hyperscale CPUs (x86) were originally designed to be rented, whereas AI does not rent cores, AI generates Tokens and sells Tokens.
50%UnverifiedOn the x86 architecture, the CPU and GPU typically share a cache-coherent view, while ARM SoCs handle cache coherency between the CPU and GPU in varying ways.
50%UnverifiedEB 18 is a short jump instruction and 90 is a NOP fill; the combination is equivalent to using an unconditional jump to bypass the problematic probe code.
50%UnverifiedThe native code of Android apps is typically compiled to generate binaries only for ARM, and running them on x86 devices requires relying on dynamic binary translation at the system level (such as Intel's Houdini translation layer).
50%UnverifiedMultithreaded code that happens to work correctly on x86 may expose hidden concurrency bugs after being ported to the Arm platform.
50%UnverifiedThe x86 architecture has a mature ecosystem and strong software compatibility, but has gradually revealed shortcomings in energy efficiency due to the historical baggage accumulated from maintaining backward compatibility.
50%