VPS for video transcoding — from €15.48/mo
Encode video on the card instead of the CPU: 1080p runs 3.7× faster. 5-hour test period before payment
Get a server with NVENC
A VPS for transcoding is a server with a share of an NVIDIA Tesla P4, which carries a separate hardware encoder called NVENC: the card encodes the video and the CPU stays free for everything else. We measured the difference on two identical machines, with and without the card. A 52-second 1080p clip: 23.1 s on the CPU against 6.2 s on the card — 3.7× faster, and in HEVC the gap is 8× (42.8 s against 5.2 s). Time is not even the main point: CPU load falls from 80% to 19%, and with a full GPU pipeline down to 2.4%. One share of the card handles four 1080p streams at once — 15× faster than real time. The machine runs on AMD EPYC 7663 with NVMe, Windows or Linux, access over RDP or SSH. A plan plus GPU bundle starts at €15.48/mo, with a 5-hour test period before payment.
Plan plus GPU bundles
One button puts both the plan and the graphics card in the cart
Start
| 4 vCPU · AMD EPYC 7663 |
| 4 GB RAM |
| 40 GB NVMe |
| GPU NVIDIA 1 GB |
| Windows or Linux · RDP |
| Test period 5 hours |
Pro
| 6 vCPU · AMD EPYC 7663 |
| 8 GB RAM |
| 60 GB NVMe |
| GPU NVIDIA 2 GB |
| Windows or Linux · RDP |
| Test period 5 hours |
Max
| 10 vCPU · AMD EPYC 7663 |
| 16 GB RAM |
| 120 GB NVMe |
| GPU NVIDIA 2 GB |
| Windows or Linux · RDP |
| Test period 5 hours |
NVENC instead of the CPU
Encoding moves to a dedicated chip on the card. Measured: 1080p H.264 in 6.2 s against 23.1 s, HEVC in 5.2 s against 42.8 s.
The CPU stays yours
19% load instead of 80%, and 2.4% with decoding on the card too. The freed cores keep serving your site, database or job queue.
Four streams at once
The encoder chip is shared across the whole card and is not split between profiles: four 1080p clips run 15× faster than real time.
How to start transcoding
Take a bundle with a card
The button puts the plan and the share of the card into the cart together. 1 GB of video memory is enough for encoding: the amount does not affect NVENC speed.
Install the system and ffmpeg
You pick the image yourself, and the NVIDIA driver is already in it. ⚠️ Use ffmpeg 6.1 or 7.0: versions 8 and 9 require a newer driver and fail with an NVENC API error.
Test it on your own file
The test period is 5 hours: run your own clip and compare the time and CPU load before you pay.
Where the card wins and where it does not
It wins wherever speed and throughput matter: bulk re-encoding of an archive, clipping and previews, restreaming, preparing HLS, recording from cameras. The more files there are, the clearer the difference — and the CPU stays free meanwhile.
It does not win in one thing: quality per byte. At the same bitrate software x264 scores SSIM 0.99735 against 0.99634 for NVENC — invisible to the eye on most footage, but for a master copy or an archive the CPU encoder is still better.
For encoding a 1 GB share at €9.99/mo is enough: the encoder chip is shared across the card and does not depend on video memory. More memory is for 3D and emulators, not for NVENC.
Take a bundle from €15.48Frequently asked questions
Didn't find your answer?
Write to us via the contact page.
A 52-second 1080p clip takes 6.2 s against 23.1 s on the CPU — 3.7× faster. In HEVC it is 5.2 s against 42.8 s, so 8×. Measured on two identical machines where only the card differed.
Four 1080p streams at once — 208 seconds of video in 13.9 seconds, which is 15× real time. The encoder chip serves the whole card and is not split between shares; Tesla cards have no session limit.
1 GB is enough for encoding — NVENC speed does not depend on the amount. A bundle with 1 GB starts at €15.48/mo. Take 2 GB only if the same machine also runs 3D or an emulator.
6.1.x or 7.0.x. The driver in the image provides NVENC SDK 12.1, while ffmpeg 8, 9 and nightly builds require 13.1 and fail with 'Driver does not support the required nvenc API version'.
Slightly: SSIM 0.99634 against 0.99735 at the same bitrate. On most footage you cannot see it. If the goal is maximum quality per byte for an archive, keep the CPU encoder.
Both. The Windows image already has the NVIDIA driver installed; for Linux we help install the NVIDIA GRID driver. Access over RDP or SSH, with a KVM console available.
Yes, a full GPU pipeline: CUDA decoding plus NVENC encoding. In our test it took 6.1 s at 2.4% CPU load — the cores are almost untouched.
Yes, 5 hours. The machine is handed over immediately, so you can run your own files and check the real time and load before paying.