TECHNICAL
demanding computational tasks while reducing the energy needed to complete them. Within its 512-bit vectors, applications can pack up to eight double- precision (64-bit) floats (FP64), or 16 single-precision (32-bit) floats (FP32), or 32 half-precision floats (16-bit, FP16 or BF16), doubling the number of elements processed per vector compared to FP32. AVX-512’s wide-vector register width and mixed-precision capability make it ideal for workloads in AI/machine learning, video/ media processing, signal processing, and other data-parallel domains, maximising compute and memory efficiency per clock cycle. The Intel Xeon 6 SoC supports up to 72 Performance cores built on Intel 3 process technology, up to eight channels of DDR5 memory operating at up to 6400MT/s, and integrated Intel® Ethernet supporting up to 200Gbps through up to eight ports for low-latency network connectivity. An optimised software ecosystem and component integration lower server costs, with dramatic performance and energy efficiency improvements compared to predecessor processors. Additional information about the built-in Intel Media Transcode Accelerator can be found in the solution brief, “Revolutionising Live Streaming: Built-in Media Transcode Accelerator Delivers 14x Performance per Watt Gains and Ultra Low-Latency.” 11 Benchmarks and testing Extensive testing by VisualOn shows how Optimizer performs compared to other CAE solutions and how it enhances video encoding on Intel CPUs and the built-in Intel Media Transcode Accelerator. All the
benchmarking below was conducted by VisualOn.
Better visual quality
The Optimizer drastically improves visual quality without increasing video bitrate, as illustrated in Figure 3 based on VisualOn’s testing. 10 Higher quality enhances user experience and improves key performance indicators (KPIs).
Comparative results: Other CAE solutions
Compared to results from other CAE per- title encoding solutions from the Netflix Technology Blog, 5 VisualOn Optimizer delivers much lower average bitrate (Figure 4) and higher average VMAF (Figure 5). 12
Seamless integration
Other CAE solutions are typically tightly bound to a particular encoder. The Optimizer is encoder-agnostic, seamlessly integrating into existing streaming workflows without disrupting operations. Since VisualOn Optimizer runs on CPUs and GPUs, often no additional hardware is required. This makes it suitable for wide deployment across multiple use cases. Optimizer on Intel Xeon 6 processors The Intel Xeon 6 SoC is based on a power-efficient microarchitecture that delivers performance and flexibility for video processing workloads. It incorporates Intel Media Transcode Accelerator, a built-in media accelerator based on a dedicated ASIC module and Intel® AI Accelerator engines, including Intel® Advanced Vector Extensions 512 (Intel® AVX-512). The Intel Media Transcode Accelerator delivers hardware accelerated video encoders that operate alongside the CPU cores, increasing throughput dramatically while consuming less power. Intel AVX-512 extends vector processing capabilities with wider vectors, extensible syntax, and rich functionality. With ultra-wide 512-bit vector operations, Intel AVX-512 can handle extremely
Comparative results: Intel Xeon 6 SoC
The following benchmarks show comparative results of CPU-based encoding running without and with VisualOn Optimizer averaged across different content categories over all ABR ladder rungs, and for the top rung of an ABR ladder. The benchmark was completed on a single-socket server based on the Intel Xeon 6 SoC. The input test suite is the same as used in the Netflix Technology Blog, 5 using an average of results across all content types for the top-rung results. Data was collected for the software (CPU) and hardware accelerated video acceleration API (VAAPI) versions of the AVC, HEVC, and AV1 encoders. The results (Table 1) show the bitrate and VMAF score for encoding using FFmpeg only and FFmpeg with Optimizer. The results reveal that bitrate is reduced by at least 12 percent and up to 44 percent, while VMAF scores generally improve. 13 Table 2 shows the results for the number of maximum parallel encoding sessions that can run in real-time on the “Tears of Steel” 1080p video clip on the Intel- based server without and with VisualOn
Figure 4: VisualOn Optimizer delivers lower average bitrate compared to other per-title CAE solutions. 12
Volume 48 No.1 MARCH 2026
93
Made with FlippingBook - Online magazine maker