Chat reacts before the video
WebRTC measured 0.49 seconds median in our tests, and 0.41 seconds with the encoder set for real time. At that delay chat, casters and the venue screen stay on the same moment.
Casters, players and viewers react to the same moment. CreekX delivers WebRTC in under a second in our tests, carries chat, and can send the same match to Twitch and YouTube at the same time.
On a platform stream the audience is 5 to 30 seconds behind the game. That is fine for watching, and it breaks everything interactive: chat celebrates a round the caster has not reached, a prediction closes after the result is known, and a venue screen showing the stream is visibly behind the players sitting under it.
WebRTC measured 0.49 seconds median in our tests, and 0.41 seconds with the encoder set for real time. At that delay chat, casters and the venue screen stay on the same moment.
Restream to several platforms at once while your site plays it live. The server copies the feed to Twitch, YouTube, Kick or any RTMP or SRT destination without re-encoding, so it costs little processor.
Bring remote players or casters in as live guests. A browser room takes up to five guests with their cameras, mixed in your browser and sent live. Game capture itself still comes from the production PC.
One programme feed in, three ways out. The longer walk-through is in the esports tournament streaming guide.
From OBS or a production switcher, over RTMP, or SRT if the venue line is unreliable. Turn B-frames off and set a keyframe every second: with OBS defaults the same server measured 1.70 seconds over WebRTC instead of 0.41. Steps: go live from OBS.
WebRTC to your site, low-latency HLS for large audiences. WebRTC viewers each hold a connection to your server; low-latency HLS is plain HTTPS and can go behind a CDN when the audience outgrows one server’s port.
Platforms at once, VODs afterwards. Turn on recording for the stream and each match lands in the video library when it ends, ready to share or to play as a 24/7 channel between events.
A longer comparison of the two low-delay options: WebRTC and low-latency HLS compared.
| Leg | Protocol | Measured delay | Use it for |
|---|---|---|---|
| Production PC to server | RTMP, or SRT on a weak line | Not measured separately | The programme feed, B-frames off, keyframe every second. |
| Server to your tournament site | WebRTC | 0.41 s median tuned; 1.70 s with OBS defaults | Viewers who chat, predict or sit in the venue. |
| Server to a large online audience | Low-latency HLS | 1.23-1.28 s median, 1.40 s worst | More viewers than one port carries; CDN in front. |
| Server to Twitch, YouTube, Kick | RTMP restream | The platform’s own delay | Reach. The feed is copied, not re-encoded. |
| Venue screens | HLS in the Android kiosk player, or WebRTC in a browser | 3.4-3.8 s / 0.49 s median | Use a browser where the screen must match the stage. |
From our test on a 4 vCPU, 8 GB server: 500 WebRTC viewers of one 1080p stream at 2.5 Mbit/s used 37% of the processor, and the server sustained about 890 Mbit/s of upload.
A sensible split is WebRTC on your own site for the people inside the event, and platforms for the wider audience. Size a final with CreekX.
In a first-party test on 13 October 2026, CreekX 2026.10.13 measured a median delay of 0.49 seconds for WebRTC, with a worst result of 0.68 seconds. Low-latency HLS measured 1.23-1.28 seconds at the median and 1.40 seconds at worst. Standard HLS measured 3.4-3.8 seconds at the median and 4.4 seconds at worst. The test used the standard one-line installer on an Ubuntu 24.04 cloud server with 4 vCPU, 8 GB memory, a 100 GB NVMe disk and about 890 Mbit/s upload bandwidth. The source was a real 1080p, 25 fps H.264 clip at 5 Mbit/s, sent from OBS and viewed on a laptop over the local network. Each step ran for 60 seconds. These are product measurements under the stated conditions, not results from a sports customer deployment or a promise for each viewer.
CreekX has no telemetry and can run with an offline licence and offline installation package supplied through sales. Video and the installed product’s operational account data remain on the customer-controlled server during local operation. Website and sales enquiries are separate. Enabling an off-site backup, external storage, restream destination or another outbound integration creates a corresponding outbound data path. Recorded video and library files are not encrypted by CreekX on the customer’s disks; use server-level disk encryption when encryption at rest is required. Viewer-security features have been checked through CreekX automated tests, but no formal third-party security certification or assessment has been completed.
The 14-day Media Server trial is available without a credit card. It includes the Media Server feature set with trial limits of five 24/7 channels, five restream destinations per stream and 500 library videos. Installation requires an internet connection. Your content is kept when the trial ends. Add a paid licence to continue using licensed features and creating channels after the trial and any applicable licence grace period.
Contact CreekX first if the deployment must be fully offline or air-gapped, will use several servers or more than 50 concurrent live streams, expects thousands of viewers, forms part of a tender, or needs installation, training or Enterprise support. CreekX currently operates one server per installation, so the team should check hardware, network capacity and product fit for a large audience before the trial.
0.49 seconds median over WebRTC in our tests on a local network, 0.41 seconds with B-frames off and a one-second keyframe. The internet adds its own delay, and we have not measured viewers across home networks. The measurements are in where live stream delay comes from.
Yes. Restream the same match to several platforms at once. Platform viewers keep the platform’s own delay; viewers on your site get the WebRTC delay. Steps: YouTube, Facebook and more at once.
No. Add graphics in OBS or your switcher before sending. creekX carries the finished programme.
creekX passes the encoder’s video through without converting it, so the frame rate is whatever OBS sends. Our measurements used 25 and 30 fps sources. We have not measured 1080p60 at game bitrates, so test it on the trial before a final.
The network port decides. A 4 vCPU server sent one 1080p stream at 2.5 Mbit/s to 500 WebRTC viewers at 37% CPU. On 890 Mbit/s of upload that is about 340 viewers at 2.5 Mbit/s, or about 148 at a 6 Mbit/s game bitrate, calculated from the port.
Yes. Leave the public watch link off and send teams a private link that expires, or one-time tokens. Steps: who may watch and stream.
Yes, each as its own stream. Receiving streams is cheap: 49 at once used 13% of a 4 vCPU server. Making several qualities for weak connections costs about 0.6 of a core per match.
Try the trial with one match, then size the final with CreekX.