Maxwell Zeff of TechCrunch reports that OpenAI’s o3 performance on benchmarks indicates that test-time compute is becoming a key factor in scaling AI models, prompting new considerations regarding costs and usage. AI founders and investors have identified this as the “second era of scaling laws,” emphasizing the importance of established methods in adapting to this shift. This development raises questions about how to effectively leverage test-time compute for AI model scaling while managing associated costs and usage.
Full Article
PCIe 8.0 Announced With 256 GT/s For AI Workloads
PCI-SIG has announced that PCI Express 8.0 will achieve a raw bit rate of 256.0 GT/s, doubling the capacity of PCIe 7.0, with a target release in 2028. This new specification aims to support high data demands from AI and quantum systems, enabling up to 1 terabyte per second bidirectional throughput while enhancing connector designs, protocol efficiency, and power consumption without sacrificing backward compatibility. Want More Context? 🔎
Read more