AMD
4 articles in the last 30 days.
-
Advancing chip design with OpenAI models
OpenAI says its models cut standard-cell area about 13% in the Jalapeño vector unit and 10% in the matrix unit, and gained up to 14% on AMD GPU IP through RTL rewrites.
-
Talking Venice with Mahesh Subramony at AMD In-Depth
An AMD server architect explains why Venice moved to 2.5D packaging, which needs about 1.6 TB/s between chiplets and the I/O die. He also covers splitting the I/O die in two and the tradeoffs of 32-core CCDs.
-
Talking Zen 6 with Mike Clark at AMD In-Depth
Zen 6c becomes AMD's dense, top-IPC core with the full 4MB of L3 per core, while Zen 6 is the high-frequency design. Architect Mike Clark explains why the two share optimisations.
-
AMD's Zen 6 Family: Feeding the Beast
Zen 6 looks like an evolution of Zen 5 on the surface, but AMD reworked the integer schedulers and branch handling. The mispredict penalty drops, and one thread can now use both decode clusters.