Also, CyberEther is a crucial part of the pipeline enabling zero-copy visualization by rendering directly from GPU memory for real-time spectral data display.

Thanks to Fast Machine Learning Conference for inviting me to present our next-gen GPU-accelerated signal processing pipeline for radio astronomy!

Built with NVIDIA Holoscan + NEXUS orchestration for real-time processing across telescope arrays.

Slides: cdn.luigi.ltd/FastML.pdf

Booting the Raspberry Pi 5 from an NVMe is satisfyingly quick! I enabled PCIe 3.0 x1 mode and it made it a few seconds faster.

New Milestone: Our first validated Coyote server!

Packed into a 4U chassis:

- 192 GB of VRAM – 4x RTX 6000 Ada
- 16 TB NVMe storage
- 800 Gbps of networking

Powering the next-generation signal processing at the Allen Telescope Array. Designed for science, engineered for more!

Playing with CyberEther Python API (called Superluminal) inspired on Matplotlib for real-time visualizations!

Feeding it data straight from a CUDA signal generator built with NVIDIA Warp.

Here’s a similar demo: github.com/luigifcruz/C...

I’m designing a modular baseboard for embedded projects: ESP32 or RP2350, power management, and a CM4-style board-to-board connector.

The idea is to swap in daughterboards for whatever function I need. No custom PCB every time. Here’s the schematic.

Any feedback before layout?

I’m testing a mini version of the server made for our next-generation signal processing pipeline. This handles 400G of data ingest, has 96 GB of VRAM with two RTX 6000 Ada, and up to 64TB NVMe. The full server has twice the capacity on the same 4U chassis. github.com/luigifcruz/s...