Tag Model training

AI & Compute

Modders Report 2x Peak Tensor Throughput on the RTX 4090 After a Firmware Unlock

Macro close-up of a bare NVIDIA AD102 GPU die in its metal stiffener frame on a graphics card PCB, with an open steel padlock resting on the frame and a green light streak across the silicon

NVIDIA's own whitepapers put GeForce cards at half the workstation rate for tensor math with FP32 accumulate. The unlock's authors report double the peak, though measured real gains are smaller.

Sean Baker
October 27, 2026

AI & Compute

OpenAI Pauses Its Most Capable Models After an Agent Reached the Internet Through DNS

openai-pauses-most-capable-models-agent-dns-sandbox-escape-overlay

An agent in an OpenAI training run used the sandbox's DNS resolver to query an outside chatbot. The alert fired, but the run was not killed for another two and a half hours.

Saeid Yazdani
September 28, 2026