Resolution, Megapixels & Bandwidth

"1-megapixel", "8MP", "4K", "Full HD" — the same camera is often described in several different ways. This page maps resolution to pixel count, pixel count to link bandwidth, and shows which MIG frame grabber and which SerDes link can carry each resolution / format / frame-rate combination.

Use this page when choosing a board (MIG-S2 / S3 / S6) or a GMSL link rate for a given camera module.

1. Basics — pixel count = width × height

pixel count = horizontal pixels × vertical pixels
1 MP (megapixel) = 1,000,000 pixels

Korean, Japanese, and Chinese datasheets often count pixels in units of 10,000 (만). Divide by 100 to get MP: 800만 ÷ 100 = 8 MP.

2. Resolution ↔ pixel count ↔ common name

"4K" means two different things. UHD 3840×2160 (broadcast, automotive) and DCI 4K 4096×2160 (cinema). Unqualified "4K" almost always means UHD.

3. Pitfalls in the numbers

Marketing rounding

• 8 MP is really 8,294,400 (8.29 M), rounded down.
• 5 MP is really 5,038,848 (5.04 M).
• "1 megapixel" (1280×720) is really 921,600 (0.92 M), rounded up.

For bandwidth and buffer calculations always use the actual width × height, never the nominal MP.

Total vs. effective vs. output pixels

A datasheet may say "5.1 MP" while the streamed output is 2592×1944 (5.04 MP).

Pixel count ≠ image quality

• Sensor size (optical format) and pixel pitch determine low-light performance and dynamic range.
• An optical format such as 1/2.44" describes the sensor diagonal and is independent of pixel count.
• Two 8 MP sensors of different size have different pixel pitch, and the smaller one is noisier.

Bayer / color filter

• The "8 MP" of an RGB sensor is the total Bayer pixel count (R, G, G, B together), not 8 M per color channel.
• Automotive variants (RCCB, RGB-IR, …) are counted the same way.

Aspect ratio

The same MP figure can mean different resolutions when the aspect ratio differs.

4. In practice — pixel count is bandwidth

For GMSL / MIPI design what matters is not the MP label but actual pixels × bits per pixel × frame rate.

bandwidth (bps) = width × height × bpp × fps × overhead

These are pure payload figures. Real links add blanking and packet overhead (×1.2–1.3). The effective GMSL2 payload limits used below are 3G = 2.6 Gbps, 6G = 5.2 Gbps, the maximum video payload specified by the SerDes vendor. They already include margin for the sideband channels — see §4.2.

4.1 Frame size and bandwidth by pixel format

Bits per pixel (bpp) by format

Bayer10 : Bayer12 : YUV422 : RGB888  =  10 : 12 : 16 : 24  =  1 : 1.2 : 1.6 : 2.4

Bayer is small not because it is compressed but because each pixel carries one color. Demosaicing happens on the receiving side.

Conversion table (MB = 10⁶ bytes)

bits / frame  = width × height × bpp        bytes / frame = bits ÷ 8
MB/s          = bytes/frame × fps ÷ 1e6      Gbps          = bits/frame × fps ÷ 1e9

The MP classes above are nominal. Real sensors differ (see the table in §2). Many "4 MP" products are actually 2560×1440 (3.69 MP) — about 90 % of the 4 MP figures above.

4.2 Effective interface bandwidth

GMSL2 links carry less video payload than their raw line rate. The vendor-specified maximum video payload is 3 Gbps mode → 2.6 Gbps, 6 Gbps mode → 5.2 Gbps (86.7 %). The difference comes from 9b/10b encoding and guard-band protection, and the stated maximums already leave room for the sideband channels.

GMSL2 uses 9b/10b encoding, an 11 % overhead (8b/10b would be 25 %) — which is why the efficiency is 86.7 % rather than 80 %.

GMSL3 runs at a fixed 12 Gbps forward rate with a 9.7 Gbps video payload. GMSL3 is less efficient than GMSL2: 80.8 % vs 86.7 %. The link doubles (6 → 12 Gbps) but the payload only grows 1.87× (5.2 → 9.7). Assuming "12G = twice 6G" over-estimates capacity. FPD-Link runs at about 80 % in both generations. FPD-Link III packs 32 bits of video payload into each 40-bit frame — the same 20 % overhead as 8b/10b — so 4.16 × 0.8 = 3.33 Gbps. FPD-Link IV carries a 6.0 Gbps video payload on its 7.55 Gbps link (79.5 %). Efficiency ranking: GMSL2 86.7 % > GMSL3 80.8 % ≈ FPD-Link III 80.0 % ≈ FPD-Link IV 79.5 %. GMSL2 stands out because of its 9b/10b encoding (11 % overhead); the others all carry roughly 8b/10b-class 20 % overhead. The FPD-Link III figure is for the 4.16 Gbps class of deserializer. FPD-Link III parts range from 2.0 to 4.85 Gbps, so check the rate of the specific device. The Thunderbolt 3 figure of 22 Gbps reflects measured PCIe-tunnel throughput (external NVMe at 2.6–2.8 GB/s), not the 40 Gbps link rate.

Applying a conservative 80 % rule instead

If you prefer a conservative 80 % rule (3G = 2.4 / 6G = 4.8 Gbps) over the vendor values, three combinations drop out. All of them are Bayer10 with only 2–4 % headroom.

The vendor values (2.6 / 5.2) already include the sideband margin. Applying a further 80 % on top double-counts that margin. Recommendation: judge against the vendor values (2.6 / 5.2), and treat any combination with less than 5 % headroom as borderline — confirm it by measurement.

4.3 Link capacity matrix

Criteria (all effective payload values): GMSL2 = 3G 2.6 / 6G 5.2 · GMSL3 = 9.7 · FPD-Link III = 3.33 · FPD-Link IV = 6.0 · USB 3.0 (S2) = 3.2 · USB 3.2 (S3) = 8.0 · TB3 = 22

Legend: ✅ fits · 🟡 borderline (under 5 % headroom — verify by measurement) · 🔴 exceeds

30 fps

60 fps

Borderline cases — 8 MP Bayer10 @ 30 fps on GMSL2 3G (2.49), 8 MP Bayer10 @ 60 fps on GMSL2 6G (4.98), and on FPD-Link IV (6.0): 8 MP RGB888 @ 30 fps, 4 MP RGB888 @ 60 fps, 8 MP Bayer12 @ 60 fps (5.88–5.97). With only 2–4 % headroom, blanking and packet overhead can push them over. Measure before committing. GMSL3 (9.7) does not change any verdict, because no combination lands in the 9.7–10.4 range. It does make clear that 8 MP RGB888 @ 60 fps (11.94) is out of reach even for GMSL3. FPD-Link III (3.33) is narrower than its raw rate suggests — 8 MP fits only up to Bayer12 @ 30 fps (2.99); YUV422 and above are blocked. FPD-Link IV (6.0) and GMSL2 6G (5.2) are in the same class. The raw rates (7.55 vs 6.0) look far apart, but the effective payloads differ by only 15 %.

4.4 What to take from the matrix

These verdicts are for pixel data only. The vendor values include sideband margin but blanking and CSI-2 packet overhead are extra, so always measure the 🟡 borderline combinations.

4.5 MIG boards + GMSL capacity matrix

Capture ceiling per board

MIG-S3 has different limits per direction. Capture (RX): GPIF III RX 1,280 MB/s is not limiting — USB is the ceiling (~1 GB/s). Playback (TX): GPIF III TX = SDR 100 MHz × 32 bit = 400 MB/s = 3.2 Gbps — this is the bottleneck. Used as a pattern generator, MIG-S3 is therefore equivalent to MIG-S2 (3.2 Gbps). The tables below are for capture.

Legend: ✅ fits · 🟡 borderline (under 5 % headroom — verify by measurement) · 🔴 exceeds

30 fps

60 fps

Coverage summary

Reading board and link together

5. Quick conversions

① pixels → MP           :  width × height ÷ 1,000,000
② MP → "만 화소" (Korean) :  MP × 100
③ 16:9, height known    :  width ≈ height × 16 / 9   (1080 → 1920,  2160 → 3840)
④ "4K is 4× Full HD"    :  3840×2160 = 1920×1080 × 4   (2× width × 2× height)

Basis of the figures

Open items: measured verification of the borderline cases — 8 MP Bayer10 @ 30 fps on GMSL2 3G, 8 MP Bayer10 @ 60 fps on GMSL2 6G, and the three FPD-Link IV cases at 5.88–5.97 Gbps.

See also: Frame Grabber for board specifications and Deserializer for supported SerDes parts.