Aurora1 is the most advanced and highest performing AV1 encoder in the market as recognized by MSU in their 2019 video encoder study. We designed Aurora1 to cover all major video coding scenarios from VOD, to broadcast quality live streaming, to ultra-low latency interactive real-time communications, and more. Aurora1 supports both 8-bit and 10-bit video, HDR, 8K resolution, and frame-rates up to 120fps.
Aurora1 can be deployed as an SDK with a binary file that takes .yuv or .y4m raw video data and outputs AV1 elementary bitstreams in the format of .webm or .ivf. or maybe integrated with FFmpeg. Aurora1 takes advantage of Visionular Intelligent Optimization technology to achieve the maximum possible visual quality at a predefined bitrate level.
Aurora1 is available in our AuroraCloud service to provide AV1 transcoding as a SaaS solution and can be deployed in the data center with its superior multi-threading performance. Aurora1 is also applicable for desktop PC based encoding applications with only 25% of the CPU needing to utilized. And as a result of Aurora1’s tremendous compute efficiency, it can also be deployed over mobile devices having been specifically optimized for ARM architectures.
Aurora1 presents compelling advantages over AVC, HEVC, and VP9 encoding solutions, as well as its peer AV1 encoders with respect to rate-distortion performance under both subjective and objective quality assessment, encoding speed, and high-efficiency memory consumption.
- 2020 Part 1: FullHD, Objective Evaluation results.
- 2019 Part II: FullHD Content, Subjective Evaluation, as the only AV1 contender in the category.
- 2019 Part III: 4K Content, Objective Evaluation, in the category of “4K Universal 1FPS” under all three objective quality metrics SSIM / PSNR / VMAF, surpassing all other AV1/VP9/265/264 encoders.
Aurora1 provides a set of speed presets for the user scenario of VOD. Compared to open-source AV1 encoders, Aurora1 provides superior encoding efficiency with faster encoding speed and less computational overhead. Aurora1 is able to manifest the standard advantages provided by AV1, to provide superior visual quality at the same bitrate level, while incurring less computational cost.
The following is actual performance data collected from our customers where for 35 1080p video clips that were selected from their video sharing platform, Aurora1 showed a clear performance gain over x265 slower while providing higher compression efficiency and double the speed.
Aurora1 has been optimized to provide live streaming encoding performance while maintaining its coding efficiency advantage over existing coding standards. This demonstrates the real-world operational advantages of AV1 compared to other codec standards such as H.264 (AVC) and H.265 (HEVC).
To show our commitment to pushing the bounds on AV1 live performance we are actively collaborating with the Twitch team to push forward the realization of 1080p60 live streaming using the new AMD Threadripper and Epyc 2 32-core processors.
Aurora1 operates 7X Faster on AMD.
Based on customer testing and internal validation, we are seeing as much as a 700% performance improvement for Aurora1 running on AMD EPYC compared to SVT-AV1 operating on Intel. This performance delta demonstrates that Aurora1 is capable of encoding four (4) live 4Kp60 streams using a single 64 core AMD EPYC Milan processor.
Compared to SVT-AV1, Aurora1 uses 32% fewer CPU cycles and 92% less memory while producing encodes that are the same or higher quality with the same bitrate efficiency. Figure 2 shows the results for Tears of Steel.
Low-Latency Interactive Real-Time Communications
We are rolling out Aurora1 with its fastest presets, including superfast & ultrafast levels, to serve the use cases of low delay interactive real-time communications scenarios. We focused first on regular PCs, making ensure that the CPU usage did not exceed 25-30% of the total CPU for a typical i5 or i7 based machine. This use case is ideal for applications like video conferencing and distance learning, where bandwidth must be reduced as far as possible, while quality is preserved.
Aurora1 is able to encode 720p30 live camera content, and screen content at 1080p30 live, demonstrating its superior performance over existing encoding solutions such as libvpx/VP8/VP9, Openh264, and even x264 and x265.
Aurora1 takes full advantage of the powerful set of Screen Content Coding (SCC) tools available. SCC is essential for slide sharing, virtual whiteboards, and general screen sharing as required by Virtual Desktop Infrastructure (VDI) systems. Superior encoder performance is needed for high visual quality to be realized with low delay, and this is where Aurora1 can really shine.
AV1 Software Decoder Support
The AV1 ecosystem is emerging and with powerful video companies like Netflix already distributing AV1 compatible bitstreams, there is no doubt that in time AV1 will be a significant codec standard.
However, the situation today is that there are no hardware decoders available in consumer devices. Hence, AV1 software decoders will play a significant role both now and over the mid-term to provide playback support in software.
Thanks to the AOMedia funded dav1d project, an open AV1 software decoder initially jointly developed by VideoLan, Mozilla, and Two Orioles, there are feasible user scenarios for AV1 where software decoding solutions can be widely accepted.
Up to today, dav1d can decode 1080p AV1 bitstreams on a single CPU core thread using a typical Snapdragon845 mobile device. Mozilla published their evaluation regarding the most recent dav1d performance data:
- AV1 Multi-Threaded Decoder Comparison 2020-May-19 (libgav1, dav1d)
- AV1 Decoder Comparison 2020-May-10 (final)
We did an extensive performance evaluation over several mobile devices while simulating our customers’ typical usage scenarios, where dav1d proved that AV1 software decoding, even at high resolution, is feasible.
Figure 4. Decoder performance comparing AV1 open-source decoder dav1d against H264 decoder ffmpeg-h264 and HEVC decoder openhevc at encoding bitrates of 2mbps and 1mbps.