The spec sheet says the machine supports four displays. The spec sheet is describing a desktop — four monitors showing windows, mostly static, occasionally scrolling. It is not describing four surfaces each receiving live video at 60fps for two hours.

Both things can be true, and the difference shows up about ninety minutes in.

Three separate limits, and people conflate them

The display controller limit. How many independent display streams the chip can drive at all. This is a hard property of the specific Mac — on Apple Silicon it varies by chip tier, and the base chips support notably fewer than the Pro and Max. Adding a Thunderbolt dock does not raise it; it adds connectors that compete for the same allowance.

The bandwidth limit. Thunderbolt and DisplayPort links have finite capacity. Two 4K60 displays on one Thunderbolt bus is a lot of data; add a fast SSD to the same bus and something gives. This limit is why a display works alone and drops out when a second is connected to the same port.

The render limit. Whether the GPU can actually composite and present all those pixels every frame. Nothing in the spec sheet speaks to this, because it depends entirely on what you are asking it to draw.

A rig fails at whichever limit it hits first, and they present very differently: the first as a display that will not come up, the second as an intermittent dropout, the third as dropped frames on content while everything looks connected and fine.

Where the render limit actually is

The useful way to think about it is total pixels presented per second.

1 × 1080p60   =  124 million pixels/sec
4 × 1080p60   =  498 million
1 × 2160p60   =  497 million
4 × 2160p60   = 1.99 billion

Four 1080p displays cost the same as one 4K display in raw pixels, which is why the display count is a poor predictor on its own. What matters is the total, and then what you are doing to those pixels: a still image is nearly free, a video layer costs a decode plus a composite, and a stack of blend modes and effects costs the GPU real time on every one of them.

The other thing that costs more than people expect is presenting to separate displays. Each display is its own presentation surface with its own vertical sync. Four displays means four sync deadlines, and they are not aligned. A GPU comfortably rendering 500 million pixels into one canvas can struggle to land the same total across four independently-synced outputs.

The failure that matters: it works, then it does not

The dangerous case is not a rig that fails at load-in. It is one that runs fine for an hour.

Two mechanisms:

Thermal throttling. Sustained GPU load raises temperature, the machine reduces clocks to stay within limits, and headroom that existed at 5pm has gone by 8pm. Fanless Macs are the worst for this, and a Mac mini on a shelf in a rack with no airflow is not far behind. There is more on this in running a show on Apple Silicon.

Accumulated pressure. Memory fragmentation, a slow leak, a growing cache, a log file. A rig with 5% headroom at the start has none after two hours.

Both are only findable one way: run the rig, at full load, for longer than the show, in the room it will be in. A ten-minute test proves nothing about a two-hour show. This is the single most valuable thing you can do with a spare afternoon in a venue.

Stop thinking in displays

The insight that unblocks most of these rigs: a display output is a poor way to deliver video to something that is not a monitor.

An LED processor does not want a display. It wants a specific region of a video frame at a mode it recognises. A second machine does not want a display. It wants a source. A streaming encoder does not want a display either.

Only actual monitors and projectors genuinely need a display output. Everything else is better served by a video transport:

  • SDI, via a DeckLink or UltraStudio. Does not touch the display controller at all. Predictable, uncompressed, and what most hardware destinations natively accept. A four-output SDI card gives you four video destinations that cost you zero displays. What matters when sending SDI out of a Mac covers the details.
  • NDI. Any number of destinations over the network, at the cost of encode on this machine and bandwidth on the switch. Genuinely unlimited in count, definitely not unlimited in capacity.
  • Syphon. If the destination is another application on the same Mac, no output is involved at all.

This reframing is why SpectraMap treats outputs as regions of a canvas that can be sent over any transport rather than as displays. The canvas is the wall; how each piece of it travels is a per-output decision. A rig that needs six destinations on a machine that drives two displays is entirely achievable — it just does not do it with display outputs.

Practical ceilings

Rough guidance, assuming a machine with a fan and adequate memory, and treating these as points where you should test rather than as guarantees:

  • Two 1080p60 destinations with video content — comfortable on almost anything current.
  • Four 1080p60 destinations, mixed stills and video — reasonable on a Pro-tier chip. Test it.
  • Four 1080p60 destinations, all live video with effects — you are near the edge. Prefer SDI over displays and soak test properly.
  • Anything at 4K on more than one destination — spec deliberately, do not assume.
  • Six or more destinations — design for SDI or NDI from the start, or use two machines.

Making the limits visible before show day

  1. Add up the pixels. Total resolution × frame rate across every destination. It is a five-second calculation and it frames everything else.
  2. Count real displays separately from video destinations. Only monitors and projectors need display outputs.
  3. Watch actual frame timing, not smoothness. Your eye will not catch one dropped frame in sixty; instrumentation will.
  4. Soak test at length, at temperature, with sudo powermetrics --samplers smc running so throttling is visible rather than inferred.
  5. Know your degrade path. If the machine cannot hold up, what gets dropped first? Deciding that in advance is much better than deciding it in front of an audience.

Specifying a machine around all of this — memory, thermals, Thunderbolt budget and how many destinations to plan for — is its own exercise.

The honest summary

One modern Mac will carry a serious multi-destination video rig. What it will not do is absorb a plan drawn around display outputs, run at 95% of capacity, and stay there for two hours in a warm room. Build with headroom, deliver to hardware over SDI or NDI rather than as displays, and prove it with a soak test rather than a demo.

Originally published on the SpectraRig blog.