A media server is where the creative team's content meets the physical room: a timeline of video, graphics, and live inputs mapped pixel by pixel onto LED walls, projection surfaces, and confidence screens. LML staffs the operators who build that map, program the timeline, and run it live on disguise, Millumin, Hippotizer, Watchout, Pixera, and Picturall. Operating one in a live production is a different discipline from programming one in a studio, and the difference shows up in three questions that come up on every show: what resolution should the content be, what codec should it be delivered in, and what happens when the server fails.

What Resolution Should
the Content Be?

The most common question a media server operator answers before a show, and the most expensive one to answer late. An LED wall is a fixed grid of pixels. Content that does not match that grid gets scaled, and scaled content is soft. Content delivered in an editing codec gets decoded on the CPU, and CPU decode is where multi-layer playback stutters. Describe the wall below and the calculator gives the canvas, the output count, and the data rate the server has to sustain.

The Wall
Distance between LED centers. Smaller pitch means more pixels per foot: 2.6 mm is common for corporate walls, 1.5 to 1.9 for close viewing.
LML Content Spec
What the Server Has to Do
0 × 0
Native
Canvas in pixels
Content at 1.0 pixel per LED pixel
1
4K60 outputs needed
0 MB/s
Per layer, sustained
0 GB
Per 10-minute file
GPU
Decoded on

Data rate per layer at this canvas, by codec

Megabytes per second the storage has to deliver for one full-canvas layer. Reference lines show a single SATA SSD and an NVMe array.

Sustained data rate per full-canvas layer for five codecs at the configured wall
View as table
CodecMB/s per layerLayers from one NVMe array
01 / Resolution

Native or it's soft

A 40-foot wall at 2.6 mm pitch is about 4,690 pixels wide. A 1920-wide master has to be stretched nearly two and a half times to cover it, and every edge in the content blurs across two or three LEDs. Type and logos suffer first. The fix costs nothing at the authoring stage and cannot be bought back later: the content team needs the canvas size before they open their software.

That is why the media server operator issues a content spec in pre-production: canvas dimensions, frame rate, codec, safe areas, and how the canvas splits across outputs. Decks and last-minute videos still arrive at 1080p, and the operator plans a scaled lane for them. Hero content should never be in that lane.

02 / Codec

A delivery codec is not a playback codec

H.264 and ProRes are built for editing and delivery, and both decode on the CPU. One layer plays fine. Three layers of ProRes at a 5K canvas, with a live input and a transition, is where a server starts dropping frames at the worst possible moment. HAP, HAP Q, and NotchLC are built the other way around: large files on disk, decoded on the GPU, with scrubbing and multi-layer playback that stays smooth.

The trade is storage. HAP Q at a big canvas runs hundreds of megabytes a second, which is why media servers run NVMe arrays and why the operator transcodes in prep, not on show day. The calculator above shows the rate for each codec at your wall.

03 / Redundancy

Two servers, one timeline

A single media server driving the main wall is a single point of failure in front of the whole audience. LML operators run a main and an understudy: identical content, identical timeline, both genlocked to the LED processor, with the backup following the main's cues over the network. A failover is a switch on the processor, not a reboot.

The rest is discipline. Outputs matched exactly to the processor's inputs, resolution and refresh confirmed on both. Timecode or a show caller feeding the same cue list the lighting console follows. And a failover actually tested at tech rehearsal, with the client watching, so nobody meets it for the first time during the keynote.

Model notes: canvas pixels are the wall dimension in millimeters divided by pixel pitch. Outputs assume 3840 × 2160 at 60 Hz per output. Data rates use typical bytes per pixel per frame: H.264 0.04, ProRes 422 HQ 0.44, HAP 0.4, HAP Q 0.8, NotchLC 1.2. Reference storage is 500 MB/s for one SATA SSD and 3,000 MB/s for an NVMe array. The sharpness meter is content pixels per LED pixel along the width, capped at 1. Real figures depend on content, bit depth, and the server; LML operators confirm the spec against the actual rig in prep.

Platforms We Operate

The servers LML operators run at a system level, from pixel map to failover.

disguise

The reference platform for large-format corporate and broadcast shows: multi-server understudy setups, timeline and layer-based playback, and native integration with LED processing and tracking systems.

Millumin

Fast, flexible macOS playback for corporate shows, installations, and theatrical work. Strong for cue-based shows where content changes up to the last minute.

Hippotizer

Green Hippo's real-time server line, common on touring and live entertainment shows with heavy effects and lighting-console control over media.

Watchout

Dataton's multi-display timeline system, widely used for blended projection and multi-screen corporate environments.

Pixera

AV Stumpfl's platform for projection mapping and large pixel-space shows, with GPU-based compositing and real-time control.

Picturall

Analog Way's media server family, often paired with Aquilon presentation switching for integrated playback and screen management.

Need a Media Server Operator?

Tell us the platform, the canvas, and the show date, and we will staff the seat and the understudy.

Submit a Booking Inquiry    Back to A/V Services