SVT-AV1 Performance on ARM and Apple Silicon

This article explores the compatibility, efficiency, and performance of the libsvtav1 encoder on ARM-based processors, with a specific focus on Apple Silicon (M-series chips). We examine how recent software updates, ARM NEON instructions, and hardware architecture updates have transformed SVT-AV1 into a highly viable encoding option on modern ARM devices.

The Evolution of SVT-AV1 on ARM

SVT-AV1 (Scalable Video Technology for AV1) was originally designed by Intel and the Alliance for Open Media with a primary focus on x86 architectures, utilizing AVX2 and AVX-512 instruction sets for acceleration. In its early stages, running SVT-AV1 on ARM processors resulted in poor performance because the encoder had to fall back on unoptimized C code.

To resolve this, open-source contributors initiated a massive effort to port and optimize SVT-AV1 for ARM64. Through the implementation of ARM NEON (SIMD) assembly code, the encoder can now execute data-parallel operations directly on ARM hardware. This optimization has bridged the performance gap, making libsvtav1 highly efficient on ARM-based servers and consumer devices alike.

Efficiency on Apple Silicon (M1, M2, and M3)

Apple Silicon chips, including the M1, M2, and M3 families, are uniquely suited for running libsvtav1 efficiently. Even though Apple Silicon features dedicated hardware decoders for AV1 (starting with the M3 generation), hardware-based AV1 encoding is still limited or absent on older M-series chips. Therefore, software encoding via libsvtav1 remains highly relevant.

SVT-AV1 runs exceptionally well on Apple Silicon due to several architectural advantages:

• High IPC and Wide Execution Width: Apple’s Firestorm, Avalanche, and Oryon performance cores have incredibly wide execution pipelines. This allows them to process the complex mathematics of AV1 encoding much faster per clock cycle than traditional ARM cores.
• Unified Memory Architecture: SVT-AV1 is a multi-threaded encoder that requires significant memory bandwidth, especially at higher resolutions like 4K. Apple Silicon’s unified memory architecture provides ultra-low latency and high-bandwidth access to RAM, preventing bottlenecks during multi-threaded encoding jobs.
• Thermal and Power Efficiency: Encoding video is a CPU-intensive task that typically drains laptop batteries and generates massive heat on x86 laptops. On Apple Silicon, libsvtav1 leverages the high performance-per-watt ratio of the ARM architecture. Users can encode high-quality AV1 video on a MacBook Air or MacBook Pro silently, with minimal thermal throttling and low power draw.

Real-World Usability and Integration

Today, libsvtav1 is fully integrated into popular media frameworks like FFmpeg and HandBrake, both of which offer native Apple Silicon binaries. When encoding on an M-series Mac, the encoder automatically detects the ARM64 architecture and utilizes the optimized NEON assembly paths.

For most users, encoding at medium-to-fast presets (such as Preset 6 to Preset 8 in SVT-AV1) yields faster-than-realtime encoding speeds for 1080p content, and highly acceptable speeds for 4K content. This makes SVT-AV1 a practical, production-ready choice for ARM-based workflows.