When deploying live media streaming workflows in the cloud, selecting the right instance type and size is crucial for achieving cost efficiency and reliable performance. We benchmarked three standard encoding ladders — the Apple TN224, Netflix’s one-size-fits-all, and the AWS Elemental example ladder — against three encoding profiles (high quality, balanced, and ultra-low latency) on CPU, NVIDIA, and NETINT Quadra (on Akamai Cloud Computing/Linode).
This document summarizes the results of those tests to help you make informed decisions for your business.
Of course, with Norsk you can run multiple channels on a single server, but the benchmarks below should help you get started on what’s right for you.
Apple TN224 (HLS)
Our first test case is based on the Apple TN224 HLS recommendations (also posted at the end of this article). We benchmarked the nine-rung ladder with resolutions ranging from 234p to 1080p.
| Profile | Hardware | vCPUs | CPU Usage
(x264) |
GPU/VPU Usage | GPU/VPU Type | RAM Usage |
| Ultra-Low Latency (ULL) | Linode Dedicated 32GB | 16 | 60% | – | – | 2.3GB |
| EC2 c6a.4xlarge | 16 | 70% | – | – | 2.2GB | |
| EC2 g4dn | 4 | 40% | 57% | NVIDIA T4 | 1.72GB | |
| Linode RTX 4000, Ada small | 4 | 24% | 22% | NVIDIA RTX 4000 Ada | 1.9GB | |
| Quadra T1U x1 Small | 8 | 34% | 32% | Quadra T1U | 1.4GB | |
| Balanced | Linode Dedicated 32GB | 16 | 80% | – | – | 3.4GB |
| EC2 c6a.4xlarge | 16 | 75% | – | – | 3.17GB | |
| EC2 g4dn | 4 | 40% | 55% | NVIDIA T4 | 1.7GB | |
| Linode RTX 4000, Ada small- | 4 | 24% | 16% | NVIDIA RTX 4000 Ada | 1.85GB | |
| Quadra T1U x1 Small | 8 | 30% | 18% | Quadra T1U | 1.4GB | |
| High Quality | Linode Dedicated 32GB | 16 | 95% | – | – | 3.8GB |
| EC2 c6a.4xlarge | 16 | 95% | – | – | 3.55GB | |
| EC2 g4dn- | 4 | 40% | 54% | NVIDIA T4 | 1.72GB | |
| Linode RTX 4000, Ada small | 4 | 24% | 22% | NVIDIA RTX 4000 Ada | 1.88GB | |
| Quadra T1U x1 Small | 8 | 30% | 19% | Quadra T1U | 1.5GB |
The 32 GB Dedicated CPUs on both EC2 and Linode perform well on the ultra-low latency and balanced profiles but struggle on the high-quality profile, with the bottleneck being CPU. While it’s possible to upgrade to larger instances with more CPUs, switching to a server with GPU acceleration would be preferred to save costs.
Netflix One-Size-Fits-All
Our second ladder is Netflix’s traditional one-size-fits-all (also posted at the end of this article), which has ten rungs with resolutions ranging from 240p to 1080p.
| Profile | Hardware | vCPUs | CPU Usage
(x264) |
GPU/VPU Usage | GPU/VPU Type | RAM Usage |
| Ultra-Low Latency (ULL) | Linode Dedicated 32GB | 16 | 67% | – | – | 2.4GB |
| EC2 c6a.4xlarge | 16 | 82% | – | – | 2.1GB | |
| EC2 g4dn | 4 | 45% | 60% | NVIDIA T4 | 1.75GB | |
| Linode RTX 4000, Ada small | 4 | 25% | 20% | NVIDIA RTX 4000 Ada | 1.9GB | |
| Quadra T1U x1 Small | 8 | 26% | 17% | Quadra T1U | 1.45GB | |
| Balanced | Linode Dedicated 32GB | 16 | 87% | – | – | 3.4GB |
| EC2 c6a.4xlarge | 16 | 88% | _ | – | 3.1GB | |
| EC2 g4dn | 4 | 45% | 57% | NVIDIA T4 | 1.8GB | |
| Linode RTX 4000, Ada small | 4 | 25% | 18% | NVIDIA RTX 4000 Ada | 1.9GB | |
| Quadra T1U x1 Small | 8 | 24% | 10% | Quadra T1U | 1.4GB | |
| High Quality | Linode Dedicated 32GB | 16 | 90% | – | – | 3.7GB |
| EC2 c6a.4xlarge | 16 | 90% | _ | – | 3.45GB | |
| EC2 g4dn | 4 | 45% | 56% | NVIDIA T4 | 1.8GB | |
| Linode RTX 4000, Ada small | 4 | 25% | 15% | NVIDIA RTX 4000 Ada | 1.89GB | |
| Quadra T1U x1 Small | 8 | 24% | 10% | Quadra T1U | 1.4GB |
The x264 instances perform well on the ultra-low latency profile but approach their maximum capacity on the balanced and high-quality profiles. As was the case with the previous ladder, it’s advisable to switch to hardware-accelerated encoding, as they provide better performance and value for money.
AWS Elemental
This is a six-rung ladder from AWS Elemental examples (also posted at the end of this article) with resolutions ranging from 288p to 1080p. Unsurprisingly, it’s the relatively least expensive, given the high number of rungs in the other ladders.
| Profile | Hardware | vCPUs | CPU Usage
(x264) |
GPU/VPU Usage | GPU/VPU Type | RAM Usage(GB) |
| Ultra-Low Latency (ULL) | Linode Dedicated 32GB | 16 | 55% | – | – | 1.8 |
| EC2 c6a.4xlarge | 16 | 45% | – | – | 1.8 | |
| EC2 g4dn | 4 | 30% | 42% | NVIDIA T4 | 1.6 | |
| Linode RTX 4000, Ada small | 4 | 17% | 8% | NVIDIA RTX 4000 Ada | 1.6 | |
| Quadra T1U x1 Small | 8 | 24% | 20% | Quadra T1U | 1.3 | |
| Balanced | Linode Dedicated 32GB, 16 cores | 16 | 60% | – | – | 2.4 |
| EC2 c6a.4xlarge, 16 cores | 16 | 45% | – | – | 2.4 | |
| EC2 g4dn | 4 | 30% | 40% | NVIDIA T4 | 1.6 | |
| Linode RTX 4000, Ada small | 4 | 17% | 10.5% | NVIDIA RTX 4000 Ada | 1.7 | |
| Quadra T1U x1 Small | 8 | 22% | 12% | Quadra T1U | 1.4 | |
| High Quality (HQ) | Linode Dedicated 32GB | 16 | 65% | – | – | 2.6 |
| EC2 c6a.4xlarge | 16 | 60% | – | – | 2.65 | |
| EC2 g4dn | 4 | 30% | 42% | NVIDIA T4 | 1.6 | |
| Linode RTX 4000, Ada small | 4 | 17% | 12% | NVIDIA RTX 4000 Ada | 1.71 | |
| Quadra T1U x1 Small | 8 | 23% | 12% | Quadra T1U | 1.45 |
This ladder performs reasonably well across all the highlighted servers, and, assuming other factors are equal, you should prefer the least costly option if running a similar ladder.
Conclusion
Norsk offers a range of options across software and hardware encoding. We recommend that you assess your requirements and select the most suitable profile and instance type for your workflow. To learn more about how Norsk handles media processing, read “Video and Encoding and Transcoding: Complete Guide with Norsk Solutions,” and for a more in-depth discussion about codecs and codec implementations, read “Cracking the Code(c)” or watch the associated webinar.
Appendix: The Industry Standard Encoding Ladders
Apple’s TN2224 Recommended Encoding Ladder
| Resolution | Bitrate (kbps) | Frame rate |
| 416 x 234 | 145 | 30 fps |
| 640 x 360 | 365 | 30 fps |
| 768 x 432 | 730 | 30 fps |
| 768 x 432 | 1100 | 30 fps |
| 960 x 540 | 2000 | 30 fps |
| 1280 x 720 | 3000 | 30 fps |
| 1280 x 720 | 4500 | 30 fps |
| 1920 x 1080 | 6000 | 30 fps |
| 1920 x 1080 | 7800 | 30 fps |
Netflix’s One-Size-Fits-All Encoding Ladder
| Resolution | Bitrate(kbps) |
| 320x 240 | 235 |
| 384 x 288 | 375 |
| 512 x 384 | 750 |
| 640 x 480 | 1050 |
| 720 x 480 | 2350 |
| 1280 x 720 | 3000 |
| 1920 x 1080 | 4300 |
| 1920 x 1080 | 5800 |
AWS Elemental Recommended Encoding Ladder
| Resolution | Bitrate(kbps) |
| 512 x 288 | 400 |
| 640 x 360 | 800 |
| 768 x 432 | 1200 |
| 960 x 540 | 1800 |
| 1280 x 720 | 2700 |
| 1920 x 1080 | 4100 |
