Cinedeck Hardware Evaluation

Introduction

Broadcast and live productions involve capturing, processing, and distributing video and audio in real time for television, live events, sports, streaming, and social media. To do this, these types of productions require systems that can handle synchronized multi-camera inputs, low-latency processing, continuous recording without interruptions, and speedy distribution of encoded media files.

Recently, we were asked to explore what kind of system would be required to perform live broadcast recording in a mobile production environment. While many of our solutions focus on content creation workflows—particularly in post-production applications and virtual production environments—we also have the ability to test and consult on workflows that aren’t specifically geared toward post-production. Our consultation services through Labs provide customers with the opportunity to explore options for optimizing their workflows and address any additional questions related to system performance and integration.

For this specific customer request, we were tasked with determining the hardware specifications for a mobile broadcast production system capable of recording four simultaneous SDI inputs in UHD (3840×2160) at 59.94 fps in a 10-bit YUV format using Cinedeck. Cinedeck is a multi-channel recording system that supports UHD capture at up to 60 FPS, simultaneously encodes proxy files, and writes directly to local or shared network storage. 

Our evaluation focused on validating modern hardware with Cinedeck to confirm its ability to handle this workflow reliably. Specifically, we needed to verify that the hardware in our test system could encode eight media files simultaneously: four ProRes422HQ master files and four HD proxy files. The proxy files would be distributed over a network to a NAS so that editors and team members could start post-production as soon as the files were uploaded. 

Testing Parameters

A key challenge in broadcast and live productions is ensuring uninterrupted recording across multiple SDI inputs while maintaining sync accuracy, minimizing playback latency, and efficiently encoding media for storage and distribution. System recommendations can vary based on budget, project scope, and the production’s specific needs. Hardware configurations can vary depending on the production environment as each environment presents unique challenges, such as space limitations, power availability, portability, and noise control. Software can also influence system configurations, as different applications leverage hardware in various ways. In other words, there is no one-size-fits-all solution for broadcast and live production environments. Numerous hardware and software combinations can be recommended to meet the technical requirements of each production.

For our testing, we employed an unorthodox method involving two systems to simulate SDI input and output utilizing the Blackmagic Design 8K Pro G2 Capture Cards in each system because we do not currently have access to enough cameras or monitors with built-in SDI output to recreate this customer’s environment. The first system acted as the SDI output source, sending four SDI signals as media sources for recording, while the test system received, synced, displayed, and played back these SDI signals in Cinedeck. While this setup does not fully replicate a standard broadcast production, as different recording software has unique hardware dependencies and processing methods, it allowed us to assess data flow and performance without relying on traditional camera sources.

Hardware Tested

We tested Cinedeck on a platform featuring an AMD Threadripper 7970X 32-core processor (CPU) paired with an ASUS Pro WS TRX50-Sage WIFI motherboard. This setup provides sufficient PCIe bandwidth to support multi-GPU configurations and additional peripherals, such as capture cards for SDI input/output and NICs for data transmission.

For the GPU, we selected the NVIDIA RTX™ 4000 Ada for its 20GB of VRAM and dual 8th-gen NVENC encoders. These encoders offload proxy media encoding from the CPU to the GPU’s dedicated media engines, allowing the CPU to focus more efficiently on processing the ProRes 422 HQ master files. This configuration supports recording and encoding eight simultaneous media files (four UHD masters and four HD proxies). The division of labor between the CPU and GPU ensures optimal resource utilization, enhancing overall performance and system stability, as any disruption in recording, especially long-duration productions, can be detrimental.

For this testing, we encoded four ProRes 422 HQ streams and four HD Proxy streams to a Gen 3 NVMe M.2 Samsung SSD 970 Pro 1TB. While this approach may be impractical in some scenarios, it pushes the disk drive’s limits and helps identify potential bottlenecks related to read/write speeds and disk cache performance. In this case, the customer intended to record HD proxy streams directly to network storage. However, testing with a higher load on the internal NVMe drive allowed us to evaluate a worst-case scenario where the NAS was unavailable, and all files had to be saved locally. It’s worth mentioning that the largest M.2 NVMe drives on the market today offer 8TB of capacity, so longer recording sessions may exceed the available storage on such a drive – requiring additional solutions such as multiple internal drives or external storage like a NAS which this customer planned to use.

Network performance becomes an additional factor when recording to a NAS or cloud storage. Connection speed, NIC capabilities, and storage read/write speeds must be sufficient to sustain data transfers without interruptions. For users looking to estimate the bandwidth required for network transfers to a NAS or cloud storage solution, start by calculating the total data rate by multiplying the bit rate of a single file by the total number of files being transferred. Then, add 10% overhead to ensure network-related tasks do not throttle the CPU. Finally, convert the total bit rate (keeping in mind the difference between bits and bytes) into the appropriate unit (usually Gb/s) to determine the necessary network speed to transfer the media files without latency.

For example, if a user is transferring eight HD proxy media files with a bit rate of 50 MB/s each, the total bandwidth requirement would be:

• 50 MB/s × 8 files = 400 MB/s
• 400 MB/s × 1.10 (to account for overhead) = 440 MB/s
• 440 MB/s × 8 (to convert from bytes to bits) = 3,520 Mb/s
• 3,520 Mb/s ÷ 1000 (to convert from Mb to Gb) = 3.52 Gb/s

In this case, a 5 Gbps network connection should be sufficient… but if additional network traffic is expected, a 10 Gbps NIC (and accompanying network gear) would be recommended to handle the transfer efficiently.

Cinedeck Hardware Performance

Our goal was to identify a hardware configuration capable of recording four ProRes 422 HQ and four HD Proxy streams without dropped frames. We like to ensure adequate resources are available to handle the additional load on the system, so we also monitored hardware utilization to ensure recording the eight streams did not completely max out the system.

Our initial test confirmed that Cinedeck could record four simultaneous UHD YUV 10-bit ProRes 422 HQ files and four HD proxy files without issues. However, we encountered a warning about an input buffer from the disk cache. To mitigate the risk of bottlenecks when encoding to cache, we recommend utilizing a PCIe Gen 4 NVMe drive with faster read/write speeds, such as the Kingston Fury Renegade.

Out of curiosity, we also tested a dual GPU setup (2x RTX 4000 Ada) to evaluate how Cinedeck handled the additional processing resources. A dual GPU configuration can be beneficial for productions requiring multiple camera inputs or a higher volume of media files. For example, the dual RTX 4000 ADA setup we tested provided a total of 4 NVENC encoders, which could more efficiently handle the encoding of additional proxy files and media streams. This offloaded work from the CPU, allowing it to focus on processing ProRes 422 HQ master files while the GPUs handled the encoding functions. Additionally, if graphics or overlays are involved in the production, dual GPUs can provide more processing power and reduce the risk of bottlenecks.

In the end, for this specific customer’s use case, we found that the second GPU was not necessary and did not provide enough benefit to be worthwhile.

Conclusion

Based on our tests, for this exact workflow of recording four ProRes 422 HQ streams and four HD Proxy streams using Cinedeck, we recommend a platform based on the AMD 32-core Threadripper 7970X CPU and an NVIDIA RTX™ 4000 Ada Generation GPU. While it may exceed the requirements of most common broadcast and live production setups, this configuration mitigates the risks associated with long-duration recording and encoding by preventing CPU throttling from maximum thread utilization. The additional cores provide ample support for processing and network bandwidth demands when distributing media files to external sources.

For those considering hardware options for broadcast production who are uncertain of exactly what their workflow needs, we encourage you to reach out to our consultants. They can assist in selecting the appropriate configuration for Cinedeck and other applications used in broadcast and live production, helping to align system performance with the demands of your workflow, and if needed they can loop in our Labs team for additional insights. After all, our goal is to help you find the right computer to get your work done and not be a hindrance!

---