AMD Zen 6 Boosts Chiplet Density to 12 Cores and Enables 144 MB 3D V-Cache Configurations

image 496
🔊 Listen to the summary

AMD’s upcoming Zen 6 architecture will feature up to 12 full-performance cores and 48 MB of shared L3 cache on a single core complex die, laying the groundwork for 24-core desktop Ryzen CPUs and monolithic 12-core designs. The structural upgrade represents a 50 percent increase in both core count and cache capacity per chiplet over Zen 5, which maxes out at eight cores and 32 MB of L3 cache.

The details emerged during an AMD technical presentation focused on its upcoming Epyc Venice server lineup, an enterprise family that shares the same underlying core architecture as future consumer Ryzen parts. AMD also showcased an Epyc Venice-X design that layers 96 MB of stacked 3D V-Cache directly on top of the CCD’s native 48 MB cache. This configuration yields 144 MB of total L3 cache across 12 cores, a substantial increase over the 96 MB found on current generation eight-core X3D chiplets.

For gaming and desktop workloads, a 12-core single-die design addresses latency bottlenecks that affect current dual-chiplet processors. Present 12-core Ryzen models split execution across two separate dies, introducing inter-CCD communication delays and requiring complex operating system scheduling to route games toward cache-heavy dies. Consolidating 12 cores onto one die allows all cores to access a single, unified cache pool.

Beyond capacity upgrades, Zen 6 integrates a revamped front end capable of dispatching up to 18 operations per cycle, backed by an enlarged operation cache housing approximately 6,750 entries. AMD noted that the architecture features upgraded branch prediction and accelerated misprediction recovery. During enterprise benchmarks, a dense Zen 6c variant posted roughly a 27 percent geometric-mean IPC gain in server and artificial intelligence tasks, though clock speeds, power limits, and retail desktop specifications have not yet been disclosed.

Source: Notebookcheck