The new Ampere-based version of the Quadra Video Server includes a 96-core CPU to support de-interlacing, decoding, and video processing for more than 300 live channels. Engineers needing to support interlaced-source content and those wanting to run additional applications on the CPU like Whisper audio transcription, packaging, media file orchestration, or DRM will appreciate the power and affordability.
NETINT is now shipping a new configuration of the Quadra Video Server for applications that require significant CPU and transcoding resources, along with ultra-low cost and power consumption per stream. The new server combines a 96-core Ampere ARM CPU and ten Quadra T1U video processing units (VPUs) in a Supermicro server chassis and costs $19,000. It’s designated the Quadra Video Server – Ampere Edition.
Quadra VPUs offer a broad range of native transcoding-related features, including H.264/HEVC/VP9 decode, H.264/HEVC/AV1 encode, scaling, overlay, and up to 18 TOPS of AI processing. When server deployments leverage these hardware-supported features, CPU utilization for the server is modest. A 32-core or even 8-core CPU can typically suffice.
When an installation requires features not supported in hardware, like deinterlacing or AV1/MPEG-2 decode, the system needs a more powerful CPU to perform these functions. CPU requirements also increase when a customer wants to run additional applications on the server, like Whisper real-time audio transcription, dynamic ABR packaging, DRM, streaming orchestration, or content management.
A system based upon a 96-core Ampere ARM CPU enables the Quadra Video Server to deliver performance unavailable from a similarly priced AMD-powered system. Ampere CPUs are ARM-based CPUs, or Advanced RISC Machines, which are more power efficient and generate less heat than other CPUs. This makes them cheaper to deploy in data centers and similar confined spaces where cooling efficiency is paramount.
The combination of Ampere ARM CPU and Quadra VPUs delivers a powerful platform with customizable features for different critical applications. For example, OTT distributors in some markets need to transcode interlaced source content, and deinterlacing is not available on Quadra.
In servers with low-power CPUs, this deinterlacing load could significantly limit system throughput. As shown in Table 1, with the Ampere 96-core CPU, the Quadra Video Server – Ampere Edition can deinterlace and output 95 1080p30 streams, 195 720p30 streams, and 365 576p30 streams. The table shows HEVC output, but H.264 and AV1 output performance should be very similar.
TABLE 1. The Quadra Video Server – Ampere Edition
significantly boosted deinterlacing performance.
Some video engineers receive source video in MPEG-2 format, which Quadra doesn’t natively decode onboard. Recent tests on the Quadra Video Server – Ampere Edition revealed that decoding a 1080p30 incoming MPEG-2 stream to YUV format took only 30% of the total 9600% available, suggesting that the system should be able to process as much as 320 streams delivered in MPEG-2 format.
TABLE 2. The Quadra Video Server – Ampere Edition delivered
outstanding performance in decoding MPEG-2 1080p video inputs.
AV1 contribution streams comprise a tiny percentage of contribution streams in 2023, but this number should increase over time. AV1 is the codec of choice for multiple large streaming providers, including YouTube, Meta, and Netflix, and AV1 usage will get a boost from Apple’s supporting AV1 decoding on the recent iPhone 15 Pro and iPhone 15 Pro Max. Like MPEG-2, Quadra does not decode AV1 natively.
As shown in Table 3, the Quadra Video Server – Ampere Edition decoded a single AV1 stream using only 100% of the available 9600% CPU cycles, suggesting that the system will be capable of decoding 96 incoming streams at full load.
TABLE 3. The Quadra Video Server – Ampere Edition delivered
solid performance decoding AV1 1080p video inputs.
There are myriad transcoding-related functions, including packaging, DRM, and advertising insertion. Hosting these applications on the transcoding server is cheaper than deploying a separate server, saves rack space, decreases power and storage costs, and can reduce latency.
One application that’s particularly useful is live transcription, which has become enabled via Open AI’s Whisper. Briefly, as detailed here, Whisper is an automatic speech recognition (ASR) system that OpenAI trained on 680,000 hours of multilingual and multitask supervised data with excellent accuracy and performance in multiple languages. OpenAI has open-sourced the models and inference code, and it’s been implemented by multiple vendors.
Ampere and NETINT are working to integrate Whisper transcription with Quadra transcoding on the new server, and early results look very promising. As shown in Figure 1, the Ampere 96-core M96-30 outproduced the NVIDIA GPUs and AMD and Intel CPUs shown, making Ampere the most efficient and cost-effective solution for Whisper audio transcription. Note that in this instance, performance is not linear, and the M96-28 included in the Quadra Video Server – Ampere Edition should perform very similarly to the M96-30 shown in the chart.
FIGURE 1. Whisper transcription performance for the Ampere CPU compared to GPUs and other CPUs.
Like all NETINT transcoders and servers, video engineers can control the new server via FFmpeg, GStreamer, or an SDK, with load management incorporated within the basic software stack. This makes the new server easy to deploy to augment or replace existing transcoding resources or to deploy in greenfield installations.
In terms of the specific configuration, the Quadra Video Server – Ampere Edition includes a 2.8 GHz 96-core Ampere Altra Max M96-28 CPU, housed within a Supermicro MegaDC ARS-110M-NR server platform with ten Quadra T1Us. The system includes 256 GB of DDR4-3200 RDIMM memory, running on Ubuntu 22.04.3 LTS.
The new configuration of the Quadra Video Server is available now. Click here to contact our sales team.