Plex/Jellyfin Hardware: How Many 4K Transcodes Can It Handle?

For 10 simultaneous 4K Plex or Jellyfin transcodes, you need a server with 8 vCPU, 32 GB RAM, at least 500 GB NVMe, and a hardware GPU such as an Intel Arc A380 or NVIDIA RTX A2000; without a GPU, this workload requires roughly 120,000–170,000 PassMark CPU.

Plex and Jellyfin transcoding server requirements in 2026

Transcoding is not the same as streaming video

Plex Jellyfin transcoding server requirements depend on the number of simultaneous transcodes, not the total number of users with access to the library. If a TV, browser, or streaming device supports the source container, video codec, audio, and subtitles, the server delivers the file through Direct Play. In this scenario, even 2 vCPU and 4 GB RAM can serve dozens of connections: the CPU does almost no video encoding, while storage and network capacity handle most of the load.

Transcoding starts when a client does not support HEVC, AV1, TrueHD, DTS, PGS subtitles, or when the original bitrate is too high for the viewer's connection. The server decodes the source video, overlays subtitles if needed, changes the resolution, and encodes the stream again. A single 4K HDR Remux at 40–80 Mbps can be more demanding than ten Direct Play clients.

When sizing a node, use the general approach described in this guide to server requirements for self-hosted applications: separate the service's baseline resource usage from peak consumption during concurrent load.

What counts as a stream

One stream is one active transcoder job, not one user. A user may start a movie in 4K and then switch to another episode, causing Plex or Jellyfin to briefly create an additional process. A production server needs 20–30% headroom across GPU, CPU, I/O, and outbound bandwidth.

• 1080p H.264 → 1080p H.264: a relatively light workload, especially with hardware encoding.
• 4K HEVC → 1080p H.264: requires 4K decoding and downscaling; this is a common source of load.
• 4K HDR → SDR: tone mapping significantly increases GPU and VRAM requirements.
• Burning in PGS/ASS subtitles: can move part of the processing to the CPU even when Quick Sync or NVENC is enabled.
• Audio transcoding: usually uses little CPU, but TrueHD/DTS-HD can sometimes trigger full-file transcoding because of client limitations.

Why Direct Play is cheaper than Plex and Jellyfin transcoding

Direct Play: network matters more than CPU

Direct Play is the best way to support more viewers without buying a powerful graphics card. A 4K HEVC movie with an actual bitrate of 25 Mbps uses about 11 GB of outbound traffic per hour, while a 70 Mbps Remux uses around 31 GB per hour. Five users streaming Remux-quality media can consume 350 Mbps of uplink without creating a single heavy CPU process.

For remote access, limit remote playback quality: 1080p at 8 Mbps provides predictable network use for most devices and reduces emergency transcodes. In Plex, this setting is available in Remote Access and client profiles; in Jellyfin, use Playback and user policy limits.

How to reduce the number of transcodes

Before ordering a GPU, check your library formats and client devices. For broad compatibility, use MP4/MKV with H.264 or HEVC, AAC/AC3 audio, and text-based SRT subtitles. PGS and ASS often require burn-in, especially on Smart TVs. It is also useful to compare platform limitations in Jellyfin vs Plex vs Emby on a VPS: your server choice affects hardware transcoding licensing and library management convenience.

1. Enable Direct Play and Direct Stream on clients.
2. Create Optimized Versions for popular movies: 1080p H.264 8–12 Mbps.
3. Store external SRT subtitles instead of image-based PGS subtitles where appropriate.
4. Set a sensible remote bitrate limit: 8–20 Mbps for 1080p and 20–40 Mbps for 4K.
5. Do not enable tone mapping for every user without testing GPU performance using real HDR files.

Plex transcoding server: CPU requirements vs Quick Sync and NVENC

Software transcoding: PassMark guidelines

If hardware acceleration is unavailable, use CPU Mark rather than vCPU count alone. A practical Plex rule of thumb is that one 1080p H.264 transcode typically requires around 2,000 PassMark. A 4K HEVC to 1080p H.264 transcode needs approximately 12,000–17,000 PassMark per stream; 4K HDR with tone mapping may require even more. These are guidelines rather than guarantees: codec profile, FPS, subtitles, and encoding preset can change the result by several times.

As a result, 5 software 1080p transcodes require roughly 10,000 PassMark, while 5 4K transcodes require 60,000–85,000 PassMark. On a VPS, dedicated rather than oversubscribed vCPU matters: CPU steal can cause video buffering even on a powerful processor.

Plex Quick Sync NVENC streams and real GPU limits

The search query plex quicksync nvenc streams cannot be answered with one universal number. Intel Quick Sync on UHD 630 is generally suitable for planning around 2–4 4K HEVC transcodes or 8–15 1080p streams; Iris Xe handles 4–8 4K or 15–25 1080p streams; Intel Arc A380 supports 8–15 4K or dozens of 1080p streams with suitable codecs. NVIDIA NVENC on professional GPUs and modern RTX cards can also handle dozens of 1080p streams, but session counts depend on the model, driver, codec, and licensing policy.

The question of hardware transcoding streams per gpu should always be tested on your exact Plex or Jellyfin version, driver, and media files. A GPU may decode 10 videos simultaneously but fail under 10 HDR-to-SDR streams with burned-in subtitles. For 4K, plan for 60–75% of verified performance to retain capacity for spikes and background tasks.

In Plex, hardware transcoding requires Plex Pass. Jellyfin does not require a subscription, but it needs a correctly installed driver and container access to the GPU device. This is the core of jellyfin transcoding hardware requirements: a video accelerator without VAAPI, QSV, or NVENC available inside the container will not deliver the expected performance.

Quick pick

Need a dedicated server?

Bare metal with NVMe in 70+ locations — configure and order in minutes.

Browse servers

Plex: how many 4K transcodes can your server handle?

Resource sizing for 2, 5, and 10 simultaneous transcodes

For 10 simultaneous 4K transcodes, 8 vCPU, 32 GB RAM, 500 GB NVMe, a 1 Gbps port, and an Intel Arc A380 or NVIDIA RTX A2000-class GPU are sufficient with hardware encoding.

The figures in this table apply to hardware H.264/HEVC transcoding without constant burn-in of complex subtitles. With software transcoding, a 10-stream 4K configuration becomes economically impractical: it requires a server with dozens of physical cores and a high CPU Mark, whereas a GPU handles the main computational workload.

How to answer the Plex how many 4K transcodes question correctly

For the query plex how many 4k transcodes, the accurate answer is: a specific GPU can handle as many 4K streams as testing with your source files confirms, but for planning, allow for 2–4 streams on older Intel UHD, 4–8 on Iris Xe, and 8–15 on Arc A380 or a comparable NVENC accelerator. Do not add together the advertised decoder and encoder numbers: the weakest stage in the pipeline determines the limit.

Storage, RAM, and outbound bandwidth for a media server

Where to store your media library and transcode cache

The movie library itself can be stored on HDD, NAS, object storage, or an attached volume if sequential reads are stable. However, the Plex/Jellyfin metadata directory, SQLite/PostgreSQL database, previews, logs, and temporary transcode directory are best placed on NVMe. Allocate 20–50 GB for cache on a small installation and at least 100 GB for several simultaneous 4K jobs.

You do not need to keep the library on the system NVMe. For example, an archival video catalog works well with a separated architecture: OS, database, and cache on fast storage; media on high-capacity storage. The same principle applies when choosing capacity for MinIO S3 storage, where metadata performance and data capacity scale independently.

RAM and network: hidden causes of buffering

A minimum of 4 GB RAM is suitable for a test server without heavy transcoding. For 2–5 hardware streams, choose 8–16 GB; for 10 streams and a Docker stack, use 32 GB. Memory does not replace a GPU, but it prevents swap usage during library scans, artwork loading, and operation of several containers.

Calculate outbound traffic with this formula: bitrate in Mbps × 0.45 = GB per hour. An 8 Mbps stream uses about 3.6 GB/hour, 20 Mbps uses 9 GB/hour, and 50 Mbps uses 22.5 GB/hour. If 10 users watch video at an average of 12 Mbps for 3 hours every day, the server will deliver approximately 13 TB per month. Check your included traffic allowance and apply quality limits for remote users.

How to set up Jellyfin or Plex with hardware transcoding

Checking Intel Quick Sync and VAAPI

On Linux, check that the Intel GPU device /dev/dri/renderD128 exists, install the driver, and pass the device into the container. For Jellyfin, QSV or VAAPI is usually selected in the Playback dashboard. In a virtual environment, you need GPU passthrough or vGPU rather than a standard virtual adapter.

sudo apt update
sudo apt install -y intel-media-va-driver-non-free vainfo
vainfo --display drm --device /dev/dri/renderD128

docker run --rm \
  --device /dev/dri:/dev/dri \
  jellyfin/jellyfin:latest

The expected vainfo result is a list of supported H.264, HEVC, and, on newer GPUs, AV1 profiles. If the command cannot see the device, the Jellyfin interface may show hardware acceleration even though transcoding is actually running on the CPU.

Test with your media library, not synthetic benchmarks

Create a test folder with at least three file types: 1080p H.264, 4K HEVC SDR, and 4K HDR with PGS subtitles. Start parallel playback sessions, monitor the GPU with intel_gpu_top or nvidia-smi, monitor CPU with htop, and check for buffering on the client. To isolate Plex or Jellyfin alongside monitoring and a reverse proxy, use the approach in this guide to calculating Docker host resources.

# NVIDIA: active processes and encoder utilization
nvidia-smi
nvidia-smi dmon -s pucvmet

# Intel: Video/VideoEnhance engine utilization
sudo intel_gpu_top

Quick pick

Need a dedicated server?

Bare metal with NVMe in 70+ locations — configure and order in minutes.

Browse servers

Frequently Asked Questions

How much CPU do I need for one 4K transcode without a GPU?

For one software transcode from 4K HEVC to 1080p H.264, plan for 12,000–17,000 PassMark CPU and at least 4 dedicated vCPU. HDR-to-SDR, 60 FPS, or burning in PGS subtitles may require more. A typical VPS with 2–4 virtual cores and no GPU will often buffer video under this workload.

Do I need Plex Pass for Quick Sync and NVENC?

Yes, Plex requires an active Plex Pass for hardware transcoding through Intel Quick Sync, NVIDIA NVENC, or AMD VCE/VCN. Jellyfin does not license hardware acceleration separately: if a GPU, driver, and device access are available, you can enable VAAPI, QSV, or NVENC without a subscription. In both cases, at least 8 GB RAM is needed for comfortable operation with multiple streams.

How much storage does the transcode cache need?

For 1–2 users, 20–50 GB of fast NVMe cache is sufficient, but for 5–10 parallel 4K transcodes, allocate 100 GB and configure automatic cleanup. The required size depends on buffer duration and target bitrate: one hour of 1080p video at 8 Mbps can create around 3.6 GB of temporary segments.

Can I run Jellyfin on a VPS without a GPU?

Yes, if most clients use Direct Play or you need no more than 1–2 1080p transcodes. For software transcoding, choose a CPU with around 2,000 PassMark headroom per 1080p stream. A 4K transcode without a GPU uses an order of magnitude more compute resources, so GPU passthrough or a dedicated server with an accelerator is more cost-effective for remote users.

Conclusion

Practical recommendation

For 2–5 transcodes, choose 4–6 vCPU, 8–16 GB RAM, and Intel Quick Sync or NVENC; for 10 4K streams, you need an Arc A380/RTX A2000-class GPU, 8 vCPU, and 32 GB RAM. First maximize Direct Play and limit remote bitrate, then scale the GPU based on tests with real media files.

Bare Metal

Need a dedicated server?

Bare metal in 70+ locations: modern hardware, NVMe, fast provisioning, and card or crypto payments.

Choose a server