Stoatworks Labs

Reference

The number on the box is not the bandwidth you get

HDMI 2.1 is 48 Gb/s. DisplayPort 2.1 is 80. Neither figure is video bandwidth — both are raw signalling rates, and the line code takes its cut before a single pixel is carried.

How big a cut varies enormously. TMDS and the older DisplayPort rates use 8b/10b and throw away a full fifth; HDMI's Fixed Rate Link uses 16b/18b and loses 11%; DisplayPort's UHBR rates use 128b/132b and lose about 3%. So a 48 Gb/s HDMI link delivers 42.7 Gb/s of video and an 80 Gb/s DisplayPort link delivers 77.6 — and comparing the headline numbers gets the ratio between them wrong.

Everything on this page is computed from two declared facts per row: the raw link rate with its line code, and each format's standard pixel clock. The fit table is checked at build time against five things the industry already knows — including that HDMI 2.0 carries 4K60 at 8-bit and does not carry it at 10.

The interfaces

InterfaceYearLanesLine codeRawVideo payloadCable
DVI single link19993 TMDS pairs8b/10b4.95 Gb/s4.0 Gb/sDVI-D or DVI-I
DVI dual link19996 TMDS pairs8b/10b9.9 Gb/s7.9 Gb/sDVI-D dual link
HDMI 1.0–1.220023 TMDS pairs8b/10b4.95 Gb/s4.0 Gb/sStandard
HDMI 1.3–1.4b20063 TMDS pairs8b/10b10.2 Gb/s8.2 Gb/sHigh Speed
HDMI 2.0–2.0b20133 TMDS pairs8b/10b18 Gb/s14.4 Gb/sPremium High Speed
HDMI 2.1–2.1b20174 FRL lanes16b/18b48 Gb/s42.7 Gb/sUltra High Speed
HDMI 2.220254 FRL lanes16b/18b96 Gb/s85.3 Gb/sUltra96
DP 1.1 (HBR)20074 × 2.7 Gb/s8b/10b10.8 Gb/s8.6 Gb/sStandard
DP 1.2 (HBR2)20104 × 5.4 Gb/s8b/10b21.6 Gb/s17.3 Gb/sStandard
DP 1.4 (HBR3)20164 × 8.1 Gb/s8b/10b32.4 Gb/s25.9 Gb/sDP8K
DP 2.1 UHBR1020194 × 10 Gb/s128b/132b40 Gb/s38.8 Gb/sDP40
DP 2.1 UHBR13.520194 × 13.5 Gb/s128b/132b54 Gb/s52.4 Gb/sDP54
DP 2.1 UHBR2020194 × 20 Gb/s128b/132b80 Gb/s77.6 Gb/sDP80

The gap between the last two columns is the line code, and it is the single most misleading thing about the headline numbers. Note that HDMI 2.1's 48 Gb/s and DisplayPort UHBR10's 40 Gb/s deliver 42.7 and 38.8 — much closer together than the raw figures suggest.

  • DVI single link — Capped by a 165 MHz pixel clock, which is where the 1920×1200 limit comes from. DVI-I adds analogue pins that carry VGA and nothing digital.
  • DVI dual link — A second set of TMDS pairs in the same shell. The connector looks nearly identical and a single-link cable in a dual-link socket simply loses the top half of the bandwidth.
  • HDMI 1.0–1.2 — Electrically DVI single link with audio and CEC added. This is why passive DVI–HDMI adapters work and passive DisplayPort ones do not.
  • HDMI 1.3–1.4b — 340 MHz pixel clock. 1.4 added the 4K modes at 30 Hz and the Ethernet channel almost nothing ever implemented.
  • HDMI 2.0–2.0b — The last TMDS generation, and sized almost exactly for 2160p60 at 8-bit. One bit of extra colour depth does not fit, which is the origin of a great deal of 4:2:2 subsampling.
  • HDMI 2.1–2.1b — A different signalling scheme entirely — Fixed Rate Link, four lanes instead of three, and a clock recovered from the data rather than sent alongside it.
  • HDMI 2.2 — FRL again, at twice the lane rate, with its own new cable certification. Displays that need it are still rare.
  • DP 1.1 (HBR) — DisplayPort is packetised rather than a raster clocked down a wire, which is what lets one link drive several displays.
  • DP 1.2 (HBR2) — Carries 2160p60 at 8-bit three years before HDMI could, which is why professional video kit went DisplayPort first.
  • DP 1.4 (HBR3) — The last 8b/10b rate, and still the most common on installed graphics cards. Display Stream Compression arrives here and does most of the heavy lifting above 4K.
  • DP 2.1 UHBR10 — The new line code matters more than the rate: 128b/132b wastes 3% where 8b/10b wasted 20%.
  • DP 2.1 UHBR13.5 — The middle tier, and the one most likely to be what a "DisplayPort 2.1" port actually negotiates.
  • DP 2.1 UHBR20 — The headline number. A source, a display and a certified DP80 cable are all required, and a port advertising 2.1 need implement none of it.

What fits, uncompressed

At 10-bit RGB or 4:4:4, which is what you want for anything being graded, keyed or projected. Every cell ispixel clock × 3 × 10 against the interface's payload rate. A cross does not mean no picture — it means no picture without chroma subsampling or Display Stream Compression.

interface ↓   format →1080p604.5 Gb/s1080p1208.9 Gb/s2160p308.9 Gb/s2160p6017.8 Gb/s2160p12035.6 Gb/s4320p6071.3 Gb/s
DVI single link4.0 Gb/s payload-0.5-4.9-4.9-13.9-31.7-67.3
DVI dual link7.9 Gb/s payload+3.5-1.0-1.0-9.9-27.7-63.4
HDMI 1.0–1.24.0 Gb/s payload-0.5-4.9-4.9-13.9-31.7-67.3
HDMI 1.3–1.4b8.2 Gb/s payload+3.7-0.8-0.8-9.7-27.5-63.1
HDMI 2.0–2.0b14.4 Gb/s payload+9.9+5.5+5.5-3.4-21.2-56.9
HDMI 2.1–2.1b42.7 Gb/s payload+38.2+33.8+33.8+24.8+7.0-28.6
HDMI 2.285.3 Gb/s payload+80.9+76.4+76.4+67.5+49.7+14.1
DP 1.1 (HBR)8.6 Gb/s payload+4.2-0.3-0.3-9.2-27.0-62.6
DP 1.2 (HBR2)17.3 Gb/s payload+12.8+8.4+8.4-0.5-18.4-54.0
DP 1.4 (HBR3)25.9 Gb/s payload+21.5+17.0+17.0+8.1-9.7-45.4
DP 2.1 UHBR1038.8 Gb/s payload+34.3+29.9+29.9+21.0+3.1-32.5
DP 2.1 UHBR13.552.4 Gb/s payload+47.9+43.5+43.5+34.5+16.7-18.9
DP 2.1 UHBR2077.6 Gb/s payload+73.1+68.7+68.7+59.8+41.9+6.3

The small figure is headroom in Gb/s — how much payload is left over, or how much is missing. A row that clears a format by a fraction of a gigabit is one colour depth or one refresh step away from not clearing it.

Two cells look wrong and are not. DVI single link and HDMI 1.0–1.2 fail 1080p60 here by half a gigabit, because this table is at 10-bit and both are 8-bit interfaces in practice — at 8-bit they clear it with room to spare, which is why everyone remembers them managing it. DisplayPort HBR2 misses 2160p60 by the same hair for the same reason: it carries 4K60 at 8-bit and not at 10.

Five checks this model has to pass

  • HDMI 2.0–2.0b carries 2160p60 at 8-bit. HDMI 2.0 carries 4K60 at 8-bit — the format it was sized for.
  • HDMI 2.0–2.0b does not carry 2160p60 at 10-bit. And does not carry it at 10-bit, which is why 4:2:2 and DSC exist.
  • DP 1.2 (HBR2) carries 2160p60 at 8-bit. DisplayPort HBR2 carried 4K60 years before HDMI could.
  • DVI single link does not carry 1080p120 at 8-bit. DVI single link stops at a 165 MHz pixel clock.
  • DP 2.1 UHBR20 carries 2160p120 at 10-bit. UHBR20 carries 4K120 at 10-bit uncompressed.

Work out your own format

The table above is one bit depth and six formats. This runs the same arithmetic on whatever you actually have to get down a cable — including the two levers that usually decide it, bit depth and chroma subsampling.

Try:

This calculator needs JavaScript. Without it, the table above still gives six common formats at 10-bit.

The interfaces and their payload rates come from the same array the table above is built from, so the two cannot disagree. What the calculator adds is a pixel clock for formats nobody standardised: it uses the CTA-861 timing where one exists and CVT reduced blanking otherwise, and tells you which. The difference is real — 1080p60 is 148.5 MHz as a television format and about 138.7 MHz as a computer one, because they blank differently.

None of them specifies a length

This is the same finding as SDI and DMX512, and it is worth stating plainly because the industry is full of confident numbers.HDMI and DisplayPort certify cables to a bandwidth, not to a length. A cable either passes the compliance test at its category or it does not, and how long the manufacturer could make it and still pass is the manufacturer's problem.

The practical consequence is that passive copper gets short as the rate goes up, and the point where it stops working is a property of that specific cable rather than of the standard. Passive HDMI at 48 Gb/s is a two-to-three metre proposition in practice, and at 18 Gb/s perhaps five. Beyond that the market answer is an active optical cable, which is a transmitter, a fibre and a receiver in a connector shell — and which is directional, so it goes in one way round and only one.

DisplayPort is stricter about it than HDMI. VESA's cable certifications are explicitly bandwidth tiers — DP8K, DP40, DP54, DP80 — and a longer cable simply lands in a lower tier. That is a more honest structure than it looks, because it makes the trade visible on the packaging.

The things that reach site

  • Passive DisplayPort-to-HDMI adapters only work from a DP++ source.Dual-Mode DisplayPort lets a port transmit TMDS instead of DisplayPort packets when it detects a passive adapter — so the adapter is just a level shifter and a pin remap. A source without DP++ needs an active adapter that genuinely converts, and the two are visually identical. This is the single most common "the adapter is faulty" call that is not a faulty adapter.
  • DVI-I carries analogue on pins that DVI-D does not have. A DVI-I male will not enter a DVI-D female, which is a mechanical mercy. The analogue pins are VGA, so a passive DVI-I to VGA adapter works and a passive DVI-D to VGA adapter cannot exist, whatever is being sold.
  • Display Stream Compression is visually lossless, not lossless.DSC is how 8K and high-refresh 4K fit down links that cannot carry them uncompressed, and VESA's claim is that it is perceptually transparent. For broadcast, grading or anything being measured, that is a claim to test rather than assume — and DSC either engages or it does not, with nothing on screen to say which.
  • HDCP failures look like cable failures. A black screen, an intermittent drop on source switch, or a display that works on one input and not another is as likely to be a key exchange as a connector. HDCP 2.2 is the gate for most 4K content and a single non-compliant device in the chain — a splitter, an extender, an older matrix — takes the whole path down.
  • EDID is negotiated once, at hotplug. A display that came up before the source, or a matrix that presents its own EDID rather than the display's, is why a format that should fit does not appear in the source's list. The bandwidth table above says what is possible; EDID decides what is offered.

Sources

Assembled 7 September 2026 with AI assistance. The fit table iscomputed from declared link rates, line-coding efficiencies and pixel clocks, and validated against five known-good cases at build time — but it models bandwidth only. It does not know about DSC, chroma subsampling, EDID negotiation, HDCP, or what your specific source and display have chosen to implement, any of which can stop a link that this table says fits comfortably.

Companion pages: SDI over coax and Ethernet cabling.