Video is the workload that makes bandwidth bills interesting. One viewer watching an hour of 1080p video at 7.8 Mbps pulls about 3.5 GB, and a live event can multiply that by thousands at the same moment. That is why a streaming dedicated server with an unmetered port is one of the few cases where the network, not the processor, is the line item that decides the budget. This guide sizes the port from published bit rates, separates the jobs a streaming server does, and shows where one server stops being enough.
Start from the bit rate, not the viewer count
Everything in streaming capacity follows from one number: the bit rate of the renditions your viewers actually watch. Apple's HLS Authoring Specification publishes one reference set of bit-rate variants, a sensible starting point if you do not have a ladder of your own. The highest variant Apple lists at five common resolutions; the HEVC figures are Apple's SDR column, HDR needs more, and Apple lists 20,000 kbps for 4K HDR:
| Rendition | H.264 average | HEVC SDR average |
|---|---|---|
| 640×360 | 365 kbps | 145 kbps |
| 960×540 | 2,000 kbps | 1,600 kbps |
| 1280×720 | 4,500 kbps | 3,400 kbps |
| 1920×1080 | 7,800 kbps | 5,800 kbps |
| 3840×2160 | — | 16,800 kbps |
Apple's specification names 2,000 kbps as the default variant players start on over Wi-Fi, and 730 kbps over cellular. From there, players move up or down with measured throughput, capped by screen size, so measure your real average from server or CDN logs.
How many viewers a streaming dedicated server can carry
Size for peaks at about 80% of a port's usable payload, which on Ethernet is roughly 94% of line rate after protocol overhead. That leaves headroom for retransmissions and bursts. Each viewer's rate below is Apple's video bit rate plus 128 kbit/s of stereo AAC audio, inside the 32 to 160 kbit/s Apple recommends.
| Per port | 1 Gbps | 10 Gbps | 20 Gbps |
|---|---|---|---|
| Usable at 80% | ~750 Mbps | ~7.5 Gbps | ~15 Gbps |
| Viewers at 2.1 Mbps (540p) | ~355 | ~3,540 | ~7,080 |
| Viewers at 4.6 Mbps (720p) | ~165 | ~1,630 | ~3,260 |
| Viewers at 7.9 Mbps (1080p) | ~95 | ~950 | ~1,900 |
| Viewers at 16.9 Mbps (4K) | ~45 | ~445 | ~890 |
These are concurrent viewers, not monthly ones, and they use average rates. For live events, size on your ladder's peak rate instead, because every viewer pulls the same high-motion segment at the same moment. A channel with 500 people watching at 4.6 Mbps for four hours a day moves about 125 TB a month and needs about 2.3 Gbps at peak. On a metered plan the terabytes are the number on the invoice. On an unmetered 5 Gbps port it is just a busy month.
Three jobs, often on three machines
"Streaming server" covers three different workloads, and they stress different parts of the hardware.
Ingest and transcoding. The server receives the source feed and encodes it into every rung of the ladder. This is CPU-bound, and it scales with the number of channels and renditions, not viewers. Software encoding a full ladder in real time is where core count earns its keep; benchmark your own encoder settings before choosing a build, because presets change the cost by multiples.
Origin. The server stores the segments and playlists and answers requests from players or CDN edges. It needs fast disks or plenty of RAM for the hot segments, and a port sized for whatever is not cached downstream.
Edge delivery. The servers viewers actually connect to. This is bandwidth-bound and is where the port size in the table above applies directly.
A small platform can run all three on one machine. As audiences grow, the edge is the first job to split off, usually onto several servers in different cities or behind a CDN.
Live or on demand
Live concentrates load: everyone watches the same segments at the same moment, so caches work brilliantly but peaks are sharp and unforgiving. Size the port for the biggest event you expect, not the average day.
On demand spreads load across a long tail of content. Peaks are gentler, but the working set is larger, so storage throughput and memory for caching matter more. A library that does not fit in RAM needs disks that can keep up with the port; a 10 Gbps port at full speed needs 1.25 GB/s from storage.
Putting the server near the audience
Streaming is less latency-sensitive than gaming, but distance still affects start-up time and how high a player can climb the ladder over long paths. Put edges near viewers:
- North-west Europe, London and southern Scandinavia: Amsterdam (Netherlands servers).
- Germany and Central Europe: Frankfurt or Vienna (Germany servers).
- Poland, the Baltics, Stockholm and Helsinki: Warsaw.
- France, Iberia and Italy: Paris, Madrid or Milan; Ireland and most of Britain: Dublin.
- South-eastern Europe: Bucharest or Sofia (Romania servers).
- North America as a second region: Ashburn (USA servers).
Choosing a European Server Location works through the trade-offs for every city.
When one server is not enough
Move beyond a single server when any of these is true:
- Peak concurrent viewers times your average rate passes the "Usable at 80%" figure for your port in the table above.
- Viewers are spread across regions far enough apart that one city adds noticeable start-up delay.
- One machine failing would take the whole service off air.
If the port is the only limit, a bigger port on the same server is the cheapest fix: on the Xeon builds, going from 1 to 2 Gbps costs €50 a month more. If distance or redundancy is the problem, add a second server in another city, or put a CDN in front of a single origin. 100Gbps dedicated server: who needs one covers the far end of the scale, where a single origin pushes roughly 9,500 to 35,000 streams, depending on bit rate.
What a streaming dedicated server is not for
NexonHost's Terms of Service prohibit using the service to infringe copyright (section 16.5) and allow suspension or termination for it (sections 10.1 and 17.1). Rebroadcasting channels you do not hold the rights to is the textbook case. Stream content you own or are licensed to carry.
Where to start
A streaming dedicated server from NexonHost starts at €129 a month with an unmetered, guaranteed 1 Gbps port and DDoS mitigation included, in any of twelve cities. On the Xeon builds the €50 gigabit steps up to 2 Gbps for €100, 5 Gbps for €360, 10 Gbps for €500 or 20 Gbps for €1,000 a month, with no per-terabyte charge at any size. Every build and its top port speed is on the unmetered dedicated servers page, the larger ports are compared on the 10Gbps dedicated server page, and port prices are on each build's configure step in the store.
Prices and port options are as of 2 October 2026. Check the store's configure step for current port prices.
Sources
- HLS Authoring Specification for Apple Devices, Apple: item 1.25 bit-rate tables (H.264, HEVC SDR and HDR), items 1.32a and 1.32b (2,000 kbps Wi-Fi and 730 kbps cellular default variants) and the stereo audio table (AAC 32 to 160 kbit/s). Retrieved 2 October 2026.
- Terms of Services, NexonHost, sections 10.1, 16.5 and 17.1 (copyright infringement prohibited; suspension and termination). Retrieved 2 October 2026.
- NexonHost store configurator: port options and prices for the Xeon builds. Retrieved 2 October 2026.
- Computed: usable throughput = line rate × 0.941 (TCP payload efficiency, 1,448 ÷ 1,538 bytes) × 0.8; viewers = usable Mbps ÷ (video bit rate + 0.128 Mbps audio); 500 × 4.628 Mbps = 2.3 Gbps; 500 × 4.628 Mbps × 14,400 s × 30 ÷ 8 ≈ 125 TB; 100 Gbps × 0.941 × 0.8 ÷ 7.928 or 2.128 Mbps ≈ 9,500 to 35,400; 7.8 Mbps × 3,600 s ÷ 8 ≈ 3.5 GB; 10 Gbps ÷ 8 = 1.25 GB/s.
