Skip to content

METASAT’s Hardware Platform Delivering Significant Multicore Performance

The METASAT project has reached more than half its duration, and with it, the first project results are about to be completed. The METASAT hardware platform is almost complete and it contains a multicore processor based on the GPL version of the Frontgrade Gaisler’s NOEL-V
integrated with the SPARROW AI accelerator and the Vortex RISC-V GPU. The prototype platform is implemented on the Xilinx VCU118 FPGA. Recently, in an article we published in the 42nd Digital Avionics Systems Conference (DASC 2023), we compared the performance of
the METASAT platform with several COTS multicore platforms in the market including the GR740 space processor from Frontgrade Gaisler. Even though the METASAT platform is implemented on an FPGA and includes the low performance GPL components of the NOEL-V
(low-area floating point unit and low performance L2 cache), it provides competitive performance compared to the ASIC implementation of the GR740.

In 8-bit AI computations accelerated with METASAT SPARROW accelerator, the FPGA METASAT prototype provides higher performance than the ASIC implementation of the GR740. This is a very promising result considering that a potential ASIC implementation of the METASAT platform using the Frontgrade Gaisler’s high performance commercial component will result in a significant performance boost for next generation space processors.

A QEMU model of the METASAT platform is already available (although incomplete in terms of full SPARROW support which is still on-going), and the attendees of our tutorial held on January 18th 2024 at the HiPEAC conference had the opportunity to experiment with it.

The METASAT hardware platform will be available as open source when the project ends, which will be after December 2024.

Stay tuned in our social media to be the first one to know when this happens!


More Posts

METASAT shines at HIPEAC Conference 2024

The METASAT Consortium proudly took centre stage at the HIPEAC Conference 2024 in Munich, where leaders in High-Performance Embedded Architecture converged to explore the latest