What it does

Host to screen, with nothing in between

A host writes data — samples, frames, primitives, overlays — across PCI Express. The renderer turns it into pixels in 2D and drives an HDMI output continuously, at a fixed frame rate, whether or not the host keeps up.

The usual way to do this is a GPU and a windowing system, and the usual result is a latency figure that depends on the operating system's mood. Here the path from a PCI Express write to a pixel leaving the connector is fixed logic, so it is measurable and it does not change under load.

  • Low latency — a bounded, published path from write to pixel, not an average
  • High frame rate — the output timing is generated locally and never waits for the host
  • 2D rendering in fabric — fills, blits, alpha blending, scaling, lines and text
  • No host GPU — and no GPU driver in your software stack
  • Deterministic under load — a busy host costs content, never timing
Host interfacePCI Express, Gen3 or Gen4, x4 or x8
Display outputHDMI, up to 4K at 60 Hz
High-rate mode1080p at up to 240 Hz
LatencyUnder 1 ms from PCI Express write to pixel out
Rendering2D: fill, blit, alpha blend, scale, line, glyph
LayersUp to 8 composited layers with per-layer alpha
Frame buffersDouble or triple buffered, tear-free
Host softwareLinux driver and a C rendering library, source included
PCIe in HDMI out Sub-millisecond No GPU
How it is built

Four stages, all of them ours

Every stage in the path is IP from our own catalogue, which is why the timing budget closes across the whole product rather than block by block.

01

Host link

ALF-CR-PCIE takes the writes at the full link rate, both directions, including at small transfer sizes.

02

Data movement

ALF-CR-DMA moves descriptors and payload into the render buffers without the host polling for space.

03

Render and composite

The 2D engine draws into a layer, composites the layers and hands finished lines to the output stage on the pixel clock.

04

Display output

HDMI timing is generated locally, so the output keeps its frame rate whatever the host is doing upstream.

Latency is measured, not modelled. The figure we publish is from a PCI Express write completing to the corresponding pixel leaving the HDMI connector, captured on hardware. Ask for the measurement setup with the datasheet — it is the only way the number means anything.
Where it fits

What people put on the screen

Instrument displays

Waveforms, spectra and status drawn straight from an acquisition path, updating at a rate an operator can act on.

Operator consoles

Systems where a frame arriving late is a safety question rather than a cosmetic one, and where a GPU driver is not welcome in the stack.

Test and simulation rigs

Closed-loop benches that need a real display fed from generated data with a latency budget the loop can be designed around.

Tell us the resolution and the latency budget.

Frame rate, what has to be drawn, and how much of the path you want to own. That is enough to say which configuration fits and whether it should be the product or the cores.