Display Stream Compression

Display Stream Compression (DSC) is a low-latency, visually lossless video compression standard for display links, published by the Video Electronics Standards Association (VESA). VESA and the MIPI Alliance announced the finalized version 1.0 on 22 April 2014, saying it could cut the video data rate by up to 66 percent. DSC compresses the pixel stream between a source, such as a graphics card or a phone's application processor, and a display. It is not a file or streaming codec. It is an optional or required feature of DisplayPort, embedded DisplayPort, MIPI Display Serial Interface (DSI) and HDMI, and VESA describes its typical compression ratio as ranging from 1:1 to about 3:1.

DSC matters for PC-tethered virtual reality because per-eye panel resolutions and refresh rates have grown faster than cable bandwidth. HTC said the HTC Vive Pro 2 (2021) was the first headset to support DSC, and later PC headsets from Bigscreen, Shiftall, Varjo and HTC also use or require it. VESA has also funded perception research at York University on how DSC behaves with stereoscopic images of the kind shown in a head-mounted display.

Reviewed 15 October 2026. Checked every claim against the cited VESA releases, DisplayPort FAQ, 2014 whitepaper, VDC-X calls, NVIDIA blog, HTC/Bigscreen/Shiftall/Varjo support pages, and the York University papers and thesis (full text or abstract and metadata). About review dates.

Origin and design goals

In a 2014 VESA whitepaper, Frederick Walls and Sandy MacInnis of Broadcom wrote that physical-layer link technology had not kept pace with rising display pixel counts. They added that some display driver ICs store whole frames, so higher resolutions also meant larger on-chip frame buffers. Proprietary compression schemes already existed, but the authors described them as often not visually lossless, poorly documented and not interoperable. VESA formed the DSC task group in late 2012, issued a call for proposals in January 2013 and evaluated six proposals.

The whitepaper lists the requirements set for the winning design:

• picture quality good enough that users cannot tell compression is active (visually lossless), at rates as low as 8 bits per pixel;
• a constant compressed data rate;
• independently decodable regions (slices), for products that update only part of the screen;
• support for RGB and YCbCr 4:2:2 or 4:4:4 at 8, 10 or 12 bits per component;
• low-cost real-time hardware, with decoders buildable on older process nodes.

The authors argued that existing standards did not fit. Transform codecs such as MPEG-2 and H.264, and wavelet codecs such as JPEG 2000 and VC-2, need many lines of pixel storage, while JPEG-LS cannot guarantee a constant bit rate and missed the quality target when used in lossy mode.

How it works

DSC is based on delta pulse code modulation (DPCM) with an Indexed Color History (ICH). An encoder needs only a single line of pixel storage plus a rate buffer. Walls and MacInnis describe these main stages:

• Color-space conversion. RGB input is converted to the reversible YCoCg-R color space using only shifts and additions.
• Prediction. Three modes are available: a modified median-adaptive predictor (the predictor used in JPEG-LS, changed so that three consecutive samples can be predicted in parallel); block prediction, which searches previously reconstructed pixels to the left and is optional for decoders because of its hardware cost; and midpoint prediction, which bounds the bits needed for worst-case samples.
• Quantization and reconstruction. A power-of-two quantizer is used, so it can be implemented with a shifter.
• Indexed Color History. Recently used pixel values are stored and can be referenced directly.
• Rate control and flatness detection. The quantization parameter is adjusted from buffer fullness and image activity so the rate buffer neither overflows nor underflows, and transitions from busy to flat areas are signalled to reduce artifacts.
• Entropy coding and substream multiplexing. Residuals are entropy coded in three substreams that are multiplexed without headers, so a decoder can run three entropy decoders in parallel and output three pixels per clock.

VESA rates DSC as a "low" complexity codec. At 8 bits per pixel it gives 3:1 compression for 24-bit color and 3.75:1 for 30-bit color. NVIDIA's implementation also accepts fractional compression ratios between 1.0 and 3.0, and its encoder splits the image into slices of equal width that are compressed separately. According to VESA's DisplayPort FAQ, the decoder adds no more than one raster scan line of latency, which it gives as less than 8 microseconds in a 4K, 60 Hz system.

Versions

Version	Date	Main points
DSC 1.0	Finalization announced 22 April 2014	Adopted into VESA embedded DisplayPort (eDP) 1.4 and MIPI DSI 1.2; offered free to VESA members and for US$350 to non-members.
DSC 1.1	Not stated in the VESA releases cited here	8, 10 and 12 bits per color. In January 2016 VESA named NVIDIA's Tegra X1 and Qualcomm's Snapdragon 820 as mobile processors using DSC 1.1.
DSC 1.2	27 January 2016	Native YCbCr 4:2:0 and 4:2:2 coding without conversion to RGB; 8, 10, 12, 14 and 16 bits per color; backward compatible with DSC 1.1.
DSC 1.2a	Not stated in the VESA pages cited here	Listed in VESA's codec comparison with 8 to 16 bits per color; VESA's DisplayPort FAQ calls it the current DSC version.
DSC 1.2b	Not stated in the VESA pages cited here	Named on VESA's DSC page, which describes it as backward compatible with DSC 1.1, and cited in VESA's 2024 VDC-X call for proposals.

The 1.2 update was aimed at televisions and HDR content. VESA said native 4:2:0 coding achieved 2:1 compression of YCbCr 4:2:0 material, against 3:1 when the same content was first converted to RGB. Dale Stolitzka of Samsung Display, chair of the VESA DSC task group, said the update expanded adoption opportunities into automotive infotainment and digital TV.

Adoption in display interfaces

At launch, DSC was written into VESA's eDP 1.4 and MIPI DSI 1.2, interfaces that connect internal panels in devices such as laptops and phones. The 2014 whitepaper said DSC made it possible to carry 8K video at 60 Hz over DisplayPort, 4K at 60 Hz over dual-link MIPI DSI, or WQXGA at 60 Hz over single-link MIPI DSI.

DisplayPort 1.4, published on 1 March 2016, was the first DisplayPort standard to use DSC. VESA said the DSC 1.2 transport enabled up to 3:1 compression, that it had been "deemed, through VESA membership testing, to be visually lossless", and that it allowed modes such as 8K at 60 Hz and 4K at 120 Hz with HDR deep color. Forward Error Correction was added on top of the DSC transport to deal with transmission errors. DisplayPort 2.1b makes DSC mandatory: VESA's FAQ states that DisplayPort 2.1b devices must support at least the HBR2 data rate and DSC, and that DSC is included for optional use in eDP 1.4, HDMI and MIPI DSI 1.2. VESA's codec comparison lists HDMI 2.1, DisplayPort 2.0 and eDP 1.4b as publicly known standards adopting DSC 1.2a.

On the graphics side, NVIDIA wrote in 2020 that its Turing architecture was the first to support DSC.

Use in VR headsets

NVIDIA's 2020 developer article explained the bandwidth problem with a hypothetical future-generation headset: 4000 x 4000 pixels per eye at 90 Hz and 24 bits per pixel would need about 69 Gbit/s, above the 25.9 Gbit/s maximum of DisplayPort 1.4a. With DSC the example stream falls to 23 Gbit/s. NVIDIA said DSC was available to application developers in VRWorks Graphics SDK 3.4, and that the VRWorks DirectMode API used by headset makers is available only to approved headset vendors.

Headset	Reported use of DSC
HTC Vive Pro 2	Announced 11 May 2021. HTC said it was the first headset to support DSC, which let it carry about 719 megapixels per second over DisplayPort 1.2 at full resolution and frame rate. HTC said it worked with NVIDIA and AMD on DSC and that the scheme was backward compatible with DisplayPort 1.2. HTC support says the 4896 x 2448 mode needs a DSC-capable graphics card plus the Link Box (2.0) and headset cable (2.0), and that compression is automatic.
Bigscreen Beyond	Bigscreen cites the bandwidth limits of DisplayPort 1.4, display controllers and MIPI. At 90 Hz the headset uses DSC to compress the signal to 1920 x 1920 per eye and a built-in hardware upscaler scales it to the 2560 x 2560 panels; at 75 Hz the image is shown at 2560 x 2560 without upscaling, which Bigscreen calls the sharpest mode.
Shiftall MeganeX Superlight 8K	Shiftall's FAQ confirms DSC and says its settings vary with resolution and color depth; the headset needs a DisplayPort 1.4 output and supported NVIDIA cards.
Varjo Aero	For laptops, a third-party Mini DisplayPort to DisplayPort adapter must support 8K video and DisplayPort 1.4 with DSC.
Varjo XR-4	Laptop adapters must support 8K video and DisplayPort 1.4 or newer with DSC, connected to a port routed to the dedicated NVIDIA GPU.
HTC Vive Focus Vision	In DisplayPort mode, HTC's error code 331 ("No supported GPU found") tells users to make sure the GPU supports DSC and has at least 6 GB of VRAM.

Bigscreen describes the difference between its 75 Hz and 90 Hz modes as subtle and mainly visible in fine detail such as text. It recommends 75 Hz for movies and reading and 90 Hz for gaming, and calls its through-the-lens comparison photos an approximation of what the eye sees in the headset.

Testing of visually lossless quality

DSC is lossy in the mathematical sense. Its claim is perceptual: VESA's FAQ defines visually lossless as a typical observer in most cases not noticing any difference from the uncompressed image, and says the codec was checked in "clinical testing by VESA in collaboration with member companies" involving nearly 250,000 subjective image comparisons, preferred over objective metrics such as PSNR. The original 2014 evaluation mostly alternated the original and compressed images in place on the same screen, using test patterns, subpixel-rendered text, screen captures, photos and video picked for difficulty. Walls and MacInnis reported that DSC beat five proprietary algorithms and was visually lossless or very nearly so for all tested images at 8 bits per pixel. They also summarized the standard in a 2014 SID paper and described the 1.2 additions for HDR, wide color gamut and 4:2:2/4:2:0 formats in a 2016 journal article.

Later work used the ISO/IEC 29170-2 flicker protocol. In it, a compressed image and its reference are shown side by side, each alternating with the reference at 5 Hz, and the observer picks the one that flickers. Robert Allison, Laurie Wilcox and colleagues presented what they called the first large-scale evaluation of DSC 1.2 under that standard in 2017. A 2018 follow-up led by York University researchers tested HDR content: 176 people were recruited, crops of 20 challenging HDR images were shown on 65-inch 4K televisions, and DSC 1.2a was run at 8 and 10 bits per pixel. The paper notes that the ISO criterion (no observer picks the reference correctly on more than 75 percent of trials) is very sensitive to outliers, so VESA uses a modified criterion based on the mean and standard deviation of responses. Under it, DSC 1.2a was visually lossless for the majority of the images at both rates.

The meaning of "visually lossless" is itself debated. A 2018 paper in the Journal of Electronic Imaging by Allison and co-authors found that no widely accepted definition exists and proposed graded levels of visually lossless quality tied to the test method instead.

Stereoscopic and head-mounted display research

VESA set up a multi-year research contract with York University, led by Allison and Wilcox, to validate and improve DSC and VESA's second codec, VDC-M, for stereoscopic 3D and AR/VR/XR use. In VESA's November 2021 account, phases 1 and 2 (completed November 2019) checked whether artifacts such as flicker seen in 2D also appear in stereoscopic viewing for head-mounted displays. Phase 3 (completed March 2020) tested whether switching off the codecs' color transform helps with images pre-corrected for lens chromatic aberration, since that pre-distortion can break the spatial correlation between red, green and blue pixels. A fourth phase on moving content was under way. VESA said both codecs performed very well and that the work led to refinements; it also released more than 28,000 uncompressed HDR image and video files from the project for free download.

Sanjida Sharmin Mohona's master's thesis at York, supervised by Allison, used the flicker paradigm to test two codecs (listed in its keywords as DSC 1.2 and VDC-M 1.0.7) on a stereoscopic display. For one image set, artifacts were less visible in some stereo images than in 2D; the other set was inconclusive. A 2021 paper by Mohona, Wilcox and Allison built a 3D version of the ISO/IEC 29170-2 flicker test and compared DSC 1.2 artifacts in images shown in the screen plane against images given a disparity offset. Artifacts were often less visible in 3D. The authors concluded that stereoscopic content, including virtual reality content, should tolerate higher compression than equivalent 2D images when the errors are binocularly unmatched or have disparity relative to the screen.

Related VESA codecs

VDC-M (VESA Display Compression-M) is a higher-complexity sibling aimed mainly at mobile panels. It codes 8 x 2 pixel blocks instead of DSC's 3 x 1 pixel groups and adds a 2D transform (DCT) with intra prediction, explicit block prediction and vector entropy coding. Its typical rate for SDR content is 6 bits per pixel, or 4:1 for 24-bit source. VESA lists VDC-M 1.1 as adopted in MIPI DSI-2 1.1 and rates its complexity "medium".

VDC-X is a proposed codec for AR/VR/MR. In an October 2023 call for requirements titled "AR/VR Display Stream Compression", VESA's DSC task group said some recent AR/VR implementations "do not conform well" to several assumptions behind DSC and VDC-M, and that VDC-X would provide bounded variable-rate coding of sparse color component planes. It is not meant to cover every DSC use case, and new designs would be held to the same visually lossless standard. The call for proposals, issued 15 April 2024, lists typical inputs of 4096 x 4096 per eye at up to 180 frames per second for VR/MR and 2048 x 2048 at up to 120 for AR, end-to-end latency targets of 8 ms (VR/MR) and 4 ms (AR), pre-applied lens distortion correction, optional pre-applied chromatic aberration correction, and adjustable sample density for foveation (required for VR/MR, optional for AR). Its tentative schedule ran to adoption in February 2026; the VESA pages cited here do not record whether that adoption took place.

See also

• DisplayPort
• DisplayPort Alt Mode
• HDMI
• VirtualLink
• Foveated rendering
• HTC Vive Wireless Adapter