MCU graphics · side-by-side reference
Three graphics subsystems, one skeleton
Renesas RH850/D1M, NXP i.MX RT1176, and Infineon Traveo T2G all drive HMI displays, and all three share the same five-stage skeleton, assets → render → memory → compose → panel. Read each pipeline down its column; scan a single stage across the row to see what's shared and what isn't. The whole story lives in two rows: how much rendering power the purple stage has, and how much RAM the teal stage costs.
The pipeline, aligned across all three
source
engine
between
display ctrl
output
The two rows that explain everything
Render row, how much power
NXP carries a true vector GPU (GC355) plus a raster blitter, resolution-independent paths, any-angle rotation. Renesas has an OpenVG drawing engine. Infineon has a fixed-function 2.5D blitter with no path geometry at all, so vector shapes must be pre-rasterized into assets. A QML Shape that renders on the RT1176 or D1M may have no hardware path on the T2G.
Memory row, how much RAM
Renesas and Infineon keep this stage tiny, sprites read straight from ROM, or a 128-line OTF ring replaces the framebuffer, to cut BOM cost and stay deterministic. NXP lets it grow into full double-buffered framebuffers in external RAM, buying generality and higher resolution at the price of bandwidth and power.
The compose row is the quiet agreement: every one of the three blends its layers on the fly at scan-out, so no fully composed frame is ever stored, the differences are all upstream, in what feeds that final blend.
Full component comparison
| Component / role | Renesas RH850/D1M | NXP i.MX RT1176 | Infineon Traveo T2G |
|---|---|---|---|
| Positioning | Automotive instrument cluster; ASIL-B safety | General crossover MCU; rich HMI / multimedia | Automotive cluster / body; safety-oriented |
| CPU | RH850 G3-cores, lockstep options | Cortex-M7 ~1 GHz + Cortex-M4 | Dual Cortex-M7 (+ M0+) |
| Vector / geometry GPU | OpenVG 1.1 engine (high-end variants) | GC355 (VeriSilicon Vivante), OpenVG / VGLite | None, no path / vector geometry |
| Blit / raster engine | 2D drawing within the GPU | PXP, blit, scale, fixed rotate, CSC, alpha | Blit engine, fill, blit, blend, affine, decompress |
| Composition / display controller | Sprite engine + video output | LCDIFv2, 8 layers, on-the-fly blend | Layer composition engine + CmdSeq replay |
| Sprite / direct-from-memory | Sprite engine reads images direct from ROM; VRAM wrapper | None dedicated (planes via LCDIFv2) | SpriteLayer (≤8 sub-layers) + OTF replay |
| Framebuffer model | On-the-fly, framebuffer minimized RAM-frugal | Full double-buffered framebuffers framebuffer-based | OTF = none; IBO / LBO-mem when needed RAM-frugal |
| Graphics memory | Internal Video RAM + wrapper; external up to DDR2 | Up to 2 MB SRAM; external SDRAM / HyperRAM / PSRAM | 4 MB line-based internal VRAM; external HyperRAM via SMIF |
| Display out / max | Parallel RGB / TCON to TFT | MIPI-DSI / parallel, up to 1920×1080 @ 30 Hz | RGB / LVDS / MIPI, dual display; OTF layer ≤ 1300×720 |
| Special blocks | JPEG unit, HUD warp (VOWE), stepper driver, sound gen | Camera CSI, extensive crypto | Safety features (JPEG / HUD vary by part) |
| Typical framework | Renesas libs, Qt / others | Qt for MCUs, Embedded Wizard, emWin, LVGL, VGLite | Qt for MCUs (QUL), emWin |
Porting note. The RH850 ↔ T2G move is conceptually easy, both are sprite/OTF, both RAM-frugal, but asset-pipeline-heavy, since you manage compressed images and sprite/line budgets on both. The RT1176 is the opposite: memory and rendering are generous, but you inherit framebuffer management, external-RAM bandwidth planning, and a real GPU driver stack. And "GPU" means three different things across the columns, so vector drawing ports cleanly between NXP and high-end Renesas (both OpenVG) but not to the T2G.