A server for 24/7 encoding.

At 1080p60 and 8 Mbit/s, a stream that never stops sends about 2.6 TB a month. At 1440p60 it is roughly 5.2 TB, and at 4K60 about 11.3 TB. The arithmetic is simply the bitrate held for every second of the month, and it is the number that decides which server is the right one — more often than the encoder is.

Usually not, and the reason is transfer rather than the card. GPU instances are typically sold with a small included allowance — ours includes 1 TB a month, then charges per gigabyte — while a single continuous stream spends several terabytes. At 4K the bandwidth can cost more than the server. A GPU earns its place when you are encoding many outputs at once or working above 1440p; for one continuous broadcast, a dedicated machine with an unmetered uplink is the cheaper and simpler answer.

Yes, on a machine whose CPU has integrated graphics. It is usually disabled in the stock Linux image rather than in firmware: the i915 module is blacklisted and the kernel boots with nomodeset. With both removed and a reboot, /dev/dri/card0 and /dev/dri/renderD128 appear and FFmpeg can use VAAPI normally. We set this up before handover on any machine bought for encoding, and send you the output of vainfo with the credentials so it is proven before you point a stream at it.

On a dedicated machine with an integrated GPU, yes — that device is the render node the encoder talks to, and it appears once the graphics driver is allowed to load. On a shared VPS it will not, because the host does not pass the integrated GPU through to guests. If your pipeline needs renderD128, you need a whole machine or a GPU instance, not a shared one.

That we do not count the traffic: no monthly allowance, no overage line on the invoice, and no fair-use clause that throttles a sustained outbound stream. It is not a promise of infinite speed — the port speed is the ceiling, and a 1 Gbit/s port carries about 324 TB a month if you saturate it. What it means in practice is that a stream running every hour of the month arrives at the same invoice as one that ran for a week.

For software encoding at 1080p, a large VPS is enough, and an unmetered one keeps the bill predictable. For hardware encoding you need a whole machine: Quick Sync lives on the CPU's integrated graphics, and shared instances do not expose it. That is the real dividing line — not raw speed, but whether the encoder silicon is reachable at all.

Yes. They are ordinary Ubuntu or Debian machines with root, so nginx-rtmp, SRS, MediaMTX, OBS headless or a plain FFmpeg pipeline all run the way they do anywhere else. We do not install or manage the streaming stack for you on a self-managed server; what we make sure of is that the encoder is available and the uplink is not going to meter you.