Benchmarks

CreekX performance benchmarks

If you are deciding whether CreekX Media Server fits your server and delay requirements, start with the conditions behind the numbers. These first-party tests show what CreekX 2026.10.13 did in defined setups. They are not customer results or promises for different hardware, networks or workloads.

01technical brief

The 13 October 2026 test setup

CreekX 2026.10.13 was installed using the standard one-line installer. Each test step ran for 60 seconds. CPU figures show CreekX usage after subtracting the test tools’ usage.

  • Cloud server with 4 vCPU, 8 GB memory and a 100 GB NVMe disk
  • Ubuntu 24.04
  • Approximately 890 Mbit/s upload bandwidth
  • A real 1080p, 25 fps H.264 test clip at 5 Mbit/s
02First-party evidence

Recorded incoming streams

This test replayed simultaneous 1080p, 5 Mbit/s inputs and recorded each one without re-encoding. It did not use physical cameras.

10 streams

CreekX used 5% CPU on the four-core server, 0.8 GB of memory and 3.7 MB/s of disk writes.

25 streams

CreekX used 7% CPU, 1.1 GB of memory and 9.3 MB/s of disk writes.

49 streams

All 49 remained live during the test step. CreekX used 13% CPU, 1.5 GB of memory and 30.6 MB/s of disk writes.

  • The test stopped at 49 because of the product’s 50-stream limit, not because the server exhausted its resources.
  • At the tested bitrate, one recorded 1080p input works out to approximately 2.3 GB per hour or 55 GB per day.
  • In this setup, disk throughput and incoming bandwidth were more significant constraints than processor usage.
03First-party evidence

Concurrent viewers

These tests used one live source. The viewer clients ran on the same machine as CreekX, so the results measure local load rather than delivery to viewers across the public internet.

500 WebRTC viewers at 1080p

At 2.5 Mbit/s, all 500 test viewers received the stream. CreekX used 37% CPU.

800 WebRTC viewers at 480p

At 1 Mbit/s, all 800 test viewers received the stream. CreekX used 47% CPU.

400 low-latency HLS viewers at 1080p

The 60-second test recorded no stalls. CreekX used 0.8 processor cores.

04technical brief

Calculated network-port capacity

The viewer tests did not measure delivery over the internet. The following figures are calculations for a 1 Gbit/s server port based on stream bitrate, not observed audience totals.

1080p WebRTC

Approximately 340 viewers per server.

720p WebRTC

Approximately 430 viewers per server.

480p WebRTC

Approximately 850 viewers per server.

05First-party evidence

Local-network delay observations

Delay was measured on 8 October 2026 on a laptop over a local network, with FFmpeg sending the way a tuned OBS does. The supplied test summary does not specify the timing instrument or measurement method, or the number and frequency of observations for each protocol. The median and worst values below record what was observed in that local test. They do not establish typical internet delay or a maximum delay for your deployment.

WebRTC

Median delay was 0.49 seconds. The worst measurement was 0.68 seconds.

Low-latency HLS

Median delay ranged from 1.23 to 1.28 seconds. The worst measurement was 1.40 seconds.

Standard HLS

Median delay ranged from 3.4 to 3.8 seconds. The worst measurement was 4.4 seconds.

06First-party evidence

AV1 data use compared with H.264

This separate test was run on 12 October 2026 using an eight-core machine, FFmpeg 9.0.2 and 720p and 480p test films. Picture quality was compared using VMAF. The supplied test summary does not specify the number or type of films, or the AV1 and H.264 encoder settings. The ranges below are observations from that comparison; these missing details limit how closely another test can reproduce them.

At the same measured picture quality

The AV1 copies used approximately 25-30% less data than H.264.

At a VMAF score one or two points lower

The AV1 copies used approximately 40-65% less data than H.264.

  • These ranges describe this test comparison and do not establish the saving for every video. Detailed test reports and raw data are available on request.
07Scope & limitations

What these tests have not established

The measurements are useful starting points for evaluation, but several capacity questions still require workload-specific testing.

  • Physical-camera compatibility and real-camera capacity were not tested in the 49-stream recording run.
  • The number of 24/7 channels one server can sustain has not been load-tested. Development measurements observed approximately 4-6% of one processor core per channel, but that is not a fixed server limit.
  • Viewer totals across the public internet were not measured. The 1 Gbit/s figures are calculations based on network-port capacity.
  • Results may change with source bitrate, recording retention, output qualities, storage performance, network conditions and other work running on the server.
08Your next step

Test the workload that matters to you

Technical buyers can use the 14-day Media Server trial to test their own sources, viewers and server. No credit card is required, and the initial installation needs internet access. For CreekX Cameras, fully offline sites, more than 50 live streams, several servers or audiences in the thousands, speak with CreekX before testing so the deployment can be sized. Detailed test reports are available through that conversation.

Follow the build