Open systems, from the radio to the GPU.
Notes from NextG Systems LLC on building open, programmable, and secure systems, in software-defined wireless and AI infrastructure.
-
A tensor accelerator in PolarFire fabric, 81× over the board's own CPUs
Apache TVM's VTA running in PolarFire SoC fabric: 49 GOPS across ResNet-18's convolution layers at 84 % of the array's theoretical peak, driven from Linux on the RISC-V cores with no kernel module. Plus the one PolarFire-specific RTL requirement that porting VTA depends on.
-
Conformant Is Not Performant: What OpenFlow Can Teach O-RAN
If you were in networking when OpenFlow arrived, O-RAN should feel familiar. Data center networking eventually got openness right with SONiC. Here's what wireless can learn from that, and where the analogy breaks down.
-
Slicing an NVIDIA GPU between tenants who don't trust each other
How we divide one NVIDIA GPU's memory and compute between gVisor sandboxes: quotas in the Sentry, a broker in the open kernel driver, and a credit scheduler that a tenant can't game by forking.
-
Why Now: Open Source, AI, and Open Hardware Meet NextG Wireless
Wireless is about to go through the same transformation data center networking went through a decade ago. We saw it coming, and we're building for it.
-
Stop asking, start predicting: SAGE cuts 5G uplink latency by 2.5×
Our SIGCOMM 2026 paper with EPFL: a real-time AI system at the gNB that predicts when and how much a phone will send, and grants the uplink before it asks.
-
A 5G cell on an M.2 card: bringing up the LiteX M2SDR with OCUDU
What it took to run an OCUDU gNB on an open-source AD9361 SDR with a commercial phone attached: hardware-timed TX, sample-exact RX timestamps, and a TX DMA that finally behaves like a FIFO.