Streaming & media
Ingest RTMP or SRT, transcode with FFmpeg on CPU cores or NVENC, and serve HLS to viewers from a dedicated server for streaming with an unmetered port and NVMe.
Traffic runs at full port speed with no gigabyte meter; a 1 Gbps port used flat out moves about 324 TB in a 30-day month.
Ryzen cores handle x264 and x265 ladders, while the RTX 4000 SFF Ada adds NVENC for H.264, HEVC and AV1 and spares the CPU.
HLS and DASH write a small file every few seconds per rendition; NVMe absorbs that churn while recordings land at full speed.
Upgrade any 1 Gbps plan to a 10 Gbps port, or pick an EPYC 7443P or 9354P, which come with 10 Gbps as standard.
Recommended servers
From a first deployment to a busy production setup. Live stock and prices; in-stock picks are online in minutes.
Ingest and restream
8C / 16T · 3.8–5.5 GHz
Eight Zen 5 cores take RTMP or SRT from your encoders, push copies to several platforms and keep room for a light transcode. All 8 locations.
Our pickCPU transcoding ladder
16C / 32T · 4.3–5.7 GHz
16 cores up to 5.7 GHz for full x264 ladders on live inputs or slow x265 presets for VOD, with 2 × 2 TB NVMe for segments and recordings.
GPU transcoding (NVENC)
20 GB VRAM · EPYC 4344P · 8C / 16T · 3.8–5.3 GHz
Ada-generation NVENC encodes H.264, HEVC and AV1 in hardware, so many renditions run side by side while the 8-core EPYC handles packaging.
Sizing
Viewer counts are plain arithmetic for serving straight from the server; a CDN in front changes the math, and transcoding load depends heavily on presets and resolution.
| Workload | Typical need | Suggested plan |
|---|---|---|
| Restream one channel to 3–5 platforms | No transcode; outbound = bitrate × destinations | AMD Ryzen 5 7600 |
| Live channel to ~150 viewers at 5 Mbps | About 750 Mbps out: fits 1 Gbps with headroom | AMD Ryzen 7 9700X |
| 1080p ABR ladder, 3–4 renditions per input | Several cores per input with x264 | AMD Ryzen 9 9950X |
| Many live channels in H.264, HEVC or AV1 | NVENC for encoding, CPU for packaging | NVIDIA RTX 4000 SFF Ada |
| Origin for 1,000+ concurrent 1080p viewers | 5 Gbps+ at 5 Mbps each: a 10 Gbps port | AMD EPYC 7443P |
| VOD archive of tens of terabytes | Parity RAID on 4 × 16 TB, NVMe for hot titles | 64 TB Storage Server |
Expert notes
Short, practical advice from our engineers for this workload.
Port speed caps how many people you can serve at once, not how much you send per month. A 1 Gbps port moves 125 MB/s, about 324 TB over a 30-day month at full rate, and an unmetered port lets you use all of it.
Software encoders get the most quality out of each bit, at a high CPU cost: one 1080p x264 rendition at a sensible preset can keep several cores busy, and x265 asks for more. That suits VOD, where encoding time is cheap, and a handful of live channels on a 16-core Ryzen.
For many live renditions, move encoding to the GPU. The RTX 4000 SFF Ada carries Ada-generation NVENC with H.264, HEVC and AV1 support, and as a professional card it is not bound by the session cap NVIDIA sets on GeForce cards. In FFmpeg, keep frames on the GPU from decode to encode:
-hwaccel cuda -hwaccel_output_format cuda to decodescale_cuda=1280:720 to resizeh264_nvenc, hevc_nvenc or av1_nvenc to encodeEven if everything starts on one box, think in four roles, so each can move to its own server later:
Root access means you install exactly what you need. These are proven choices for this workload.
FAQ
Divide the port speed by the average bitrate and keep about 20% headroom. At 5 Mbps per 1080p viewer, 1 Gbps serves roughly 160 concurrent viewers; at 2.5 Mbps for 720p, about 320. For larger audiences, add the 10 Gbps upgrade ($129/mo), choose an EPYC 7443P or 9354P with 10 Gbps included, or put a CDN in front of your origin.
Yes. Traffic is not counted or billed by the gigabyte, so you can run the port at full speed around the clock within the acceptable use policy. Two limits are worth knowing: bandwidth cannot be resold as IP transit, and you need the rights to the content you distribute. Live events, VOD libraries and 24/7 channels you own are exactly what the port is for.
Use the CPU when quality per bit matters most and channel counts are small, typically VOD encodes with x264, x265 or SVT-AV1 at slower presets. Use NVENC on the RTX 4000 SFF Ada when many live renditions run at once or you want real-time AV1 without dedicating a large CPU to it. Plenty of pipelines mix both: GPU for live, CPU for the archive.
Yes. A restream server takes one ingest and pushes a copy to each destination without re-encoding, so the CPU stays mostly idle and outbound traffic equals your bitrate times the number of platforms. nginx-rtmp-module and SRS handle this natively, and FFmpeg’s tee muxer does the same from the command line. Keep stream keys in config files, not in shell history.
Multiply bitrate by duration: one hour at 5 Mbps is about 2.25 GB per rendition. A ladder of 5, 3 and 1.5 Mbps needs about 4.3 GB per hour, so 30 TB holds roughly 7,000 hours. The 64 TB Storage Server gives about 32 TB usable on its 4 × 16 TB drives in RAID 6 or RAIDZ2, with NVMe for the most-watched titles.
Pick one of the recommended in-stock servers and it can be online in minutes. Need something bigger or different? Our engineers quote custom builds within one business day.