How to achieve hyper-velocity with ffmpeg video transcoding

• Share on Facebook
• Share on LinkedIn
• Share on X

Overview

FFmpeg is the silent engine behind a vast portion of the digital world’s video content. As an industry-standard, open-source multimedia framework, it powers everything from video streaming services and social media platforms to online video editors and cloud storage backends.

However, the very tasks that make FFmpeg so essential; transcoding, compression, and analysis; are incredibly time-consuming, creating significant bottlenecks for companies that process video at scale.

This case study demonstrates how Incredibuild successfully accelerated FFmpeg’s video processing tasks by over 87%, transforming a time-intensive process into a high-speed, parallel operation.

The Challenge: The video processing burden

Hundreds of companies, including video streaming giants, social media platforms, and online learning providers, rely on FFmpeg for server-side video processing. The core challenge is efficiency. A single task, such as converting a large video file for storage or streaming, can take a significant amount of time, delaying content availability and tying up valuable computing resources.

For companies processing hundreds or thousands of videos daily, these delays accumulate, leading to:

• Bottlenecks cousig long execution times for media processing pipelines.
• Inefficient resource utilization as powerful machines sit idle, while compute intensive tasks are processed on less powerful hardware in adjacent machines.

The PoC aimed to prove that Incredibuild could take these, single-machine workloads and distribute them across a network of machines, providing a new level of acceleration for the multimedia industry.

The Solution: workload distribution

The solution was to break a single, large video processing task into multiple chunks and distribute these chunks to a network of “helper” machines. This allowed the transcoding work to be done in parallel, not only on the thread level – but also leveraging the combined power of multiple machine’s  CPUs and GPUs

For this PoC, two different transcoding scenarios were tested across both Linux and Windows environments:

1. CPU-Based Transcoding: A common video compression task (H264 to H265 HEVC transcoding) was performed on large video files.
2. GPU-Based Transcoding: A second test leveraged the power of Incredibuild to distribute NVENVC GPU-based transcoding tasks, proving that the technology could accelerate even the most advanced, hardware-specific workloads.

The Results: drastic time save

The results of the PoC demonstrated a massive reduction in processing time, with consistent, dramatic improvements across both operating systems and processing methods.

1. CPU-Based Transcoding (H264 to H265 HEVC)

Using a single large 1GB video file, the test compared the time it took to complete the transcoding task on a single machine versus using a grid of 10 helpers.

These results show that a task that previously took nearly an hour could be completed in just over seven minutes, exhibiting a high progression coefficient of performance scaling with added compute nodes.

2. GPU-Based Transcoding (H265/hevc_nvenc)

Even with the phenomenal speed of GPU-based transcoding, a single machine still introduces a bottleneck. The PoC proved that we were able to achieve a near-linear scale improvement using a network of 3 machines.

In this test, we’ve demonstrated our ability to accelerate not just CPU-bound tasks, but also to make an already fast GPU-based process even faster, proving its value in a hybrid computing environment.

Conclusion: A game-changer for video processing

The FFmpeg PoC clearly and powerfully demonstrates Incredibuild’s ability to solve a major industry challenge. For any company that processes video at scale. From social media giants to small online video editors, that aim to maximize their rendering farms utilization, be these on-premises deployments or cloud setups.