In this talk, we report early results in the development of a hardware/software stack enabling direct P2P communication between a Fairwaves XTRX / XYNC SDR and an Nvidia GPU. We are able to saturate a PCIe link (7 Gbps) in loopback mode. IQ samples are streamed from the RF IC at more than 350 Mbps. Implementation of this architecture required the development of a new LiteX-based Artix FPGA gateware, and a new SDR framework written in the Julia programming language. The combination of extremely high throughput, GPU computing power, and efficient Julia code will open the doors to a new generation of phased-array and machine-learning experimentation.