Review: NETINT Quadra T1U Video Processing Unit
Throughput Testing: Ladder
For throughput ladder testing, I used the same “Football” video source on a RAM disk with the encoding ladder listed in Figure 9 (below).
Figure 9. Encoding ladder encodes
Figure 10 (below) shows the script used when I performed encodes on the Quadra T1U.
Figure 10. Encoding ladder script for Quadra
Figures 11 and 12 (below) show scripts used for the FFmpeg x264 and x265 encoding ladders.
Figure 11. Encoding ladder script for x264
Figure 12. Encoding ladder script for x265
Figure 13 (below) shows the results of my throughput testing with encoding ladders.
Figure 13. Results for throughput performance testing encoding ladder
I had 12 simultaneous encodes with the Quadra T1U encoding to H.264 and HEVC. Quadra AV1 encoding produced 11 encodes, while FFmpeg x264 produced two, and FFmpeg x265 only one.
Quality Testing: Throughput and Quality Optimized
My last tests focused on quality comparisons with throughput and quality optimized. For the throughput quality testing, I used the configuration from earlier for the throughput testing for single-file encoding and encoding ladders. The Veryfast preset was used for x264 and ultrafast for x265 to achieve reasonable throughput with these software-based transcoders.
For quality-optimized testing, I used the Medium preset for x264/x265 and a high-quality configuration for Quadra. I ran a master script that produced the following files:
- Quadra H.264 (quality and throughput) at 6Mbps
- FFmpeg x264 (quality and throughput) at 6Mbps
- Quadra HEVC (quality and throughput) at 5Mbps
- FFmpeg x265 (quality and throughput) at 5Mbps
For analysis, I downloaded the encoded files and input them in FFmetrics to compute PSNR, SSIM, and VMAF scores.
Figure 14 (below) shows how I analyzed the results in FFmetrics. I added the source up top, then I added the video encoded with Quadra video on top and placed the x264 and x265 videos on the bottom. Next, I ran my reports. Once the metrics were finished, I could see them in FFmetrics.
Figure 14. Results analyzed in FFmetrics
Figure 15 (below) shows my results comparing high throughput for Quadra and x264.
Figure 15. Comparing Quadra H264 with x264 using high-throughput configurations at 6 Mbps
For throughput-optimized quality for H.264, the Quadra had higher quality with PSNR and VMAF (Figure 16, below). It’s important to note that the Quadra video came close to the desired “93” score that video engineers and streaming producers are trying to achieve. FFmpeg did achieve higher quality with SSIM.
Figure 16. Comparing Quadra H264 with x264 using high-quality configurations at 6 Mbps
Using the quality-optimized configurations with H264, Quadra had higher PSNR and VMAF scores, while FFmpeg once again had a higher number for SSIM, as shown in Figure 17 (below).
Figure 17. Comparing Quadra HEVC with x265 using high-throughput configurations at 5 Mbps
When it came to throughput-optimized quality for comparisons of HEVC and x265, Quadra had higher scores with PSNR, SSIM, and VMAF.
Using the quality-optimized configurations for Quadra HEVC and x265, FFmpeg achieved higher PSNR and SSIM scores, while Quadra had a higher VMAF score (Figure 18, below).
Figure 18. Comparing Quadra HEVC with x265 using high-quality configurations at 5 Mbps
NETINT also supplied scripts and results using the Netflix Meridian test file. I verified these results, and with the simpler, more movie-like content, Quadra rated highest in 12 of 12 metric comparisons (four file comparisons each using VMAF, SSIM, and PSNR) and achieved VMAF scores of 95 or higher on all four comparisons (Figure 19, below).
Figure 19. Netflix Meridian test results
Closing Thoughts
In closing, I’ll return to the questions I had for this review. After my testing, I can see how the Quadra T1U could benefit colleges and universities like mine. Universities that encode thousands of weekly videos or live streams could benefit from faster encoding and fewer servers in their data centers.
Could encoding with the Quadra T1U provide a significant reduction in CPU usage for single-file encodes and encoding with encoding ladders?
The Quadra T1U does provide a significant reduction in CPU usage with single-file encodes as well as with encoding ladders. ASIC-based encoding really shines with the Quadra T1U being able to perform 32 simultaneous encodes for single-file encoding, 12 encoding ladder encodes with H.264/HEVC, and 11 encoding ladder encodes with AV1. The Quadra T1U’s competitive advantage increases with newer codecs, which take a longer time to encode.
Could significantly more encodes be performed using the Quadra T1U compared to CPU-based encoding?
As a streaming and encoding techie, I find the number of encodes that are possible with only one Quadra T1U module just amazing. I also like how encodes can be easily run from the terminal using master scripts.
Would the quality of encoding using the Quadra T1U be the same as FFmpeg encodes or better?
The quality of encoding is just as good as encoding using FFmpeg and, in many cases, better. The fact that the Quadra T1U is faster and can do many more encodes makes it that much better.
My last thought is related to green streaming. With so many organizations looking into green and eco-friendly streaming, I can see products like the Quadra T1U leading the way. I didn’t have time in this article to analyze the reduction of CapEx, cost per stream, or the number of servers that a product like the Quadra T1U could produce, but I can see how it could make a significant impact. And if that’s the case, this should be a great product for video engineers who are interested in green and eco-friendly streaming.
Related Articles
Magewell's new Director Mini is a mini marvel. This 5.5" "all-in-one" production tool with a $1,299 street price can mix multiple inputs and do picture-in-picture graphics, audio mixing, recording, streaming, and more, as IEBA's Anthony Burokas explains in this in-depth review.
09 Jan 2024
Whether you use the cloud or a hybrid solution to encode your live or on-demand streaming content, high-efficiency hardware-based encoding has come of age. This article explores the math and science behind these at-scale solutions that are not only good for business, but also good for the environment.
01 Dec 2021