While architecture Zen 6 hasn’t even made its market debut yet, AMD would already prepare the sequel with Zen 7. The first information circulating suggests a major evolution, both in terms of performance and internal architecture. Between a significant jump in IPC, a profound overhaul of the cache and configurations that can reach up to 72 cores, the manufacturer seems determined to widen the gap even further.
On paper, Zen 7 could well represent a major turning point for desktop processors, particularly for gamers, where memory management and latency become determining factors.
An expected IPC gain of up to 25% for AMD’s Zen 7 architecture
According to the first indiscretions, Zen 7 would aim for an IPC (instructions per cycle) gain of between 15 and 25%. Concretely, this means that processors would be able to execute more instructions at equal frequency, a key lever for improving overall performance without necessarily increasing consumption.
To achieve this objective, AMD would rely on a new engraving process from TSMC, often referred to as the A14 node. This development would make it possible to increase the density of transistors while optimizing energy efficiency, a balance that has become crucial in modern architectures.
But beyond the finesse of the engraving, a significant part of these gains would come from an in-depth overhaul of the cache hierarchy, an element that is now central to in-game performance.
A cache overhaul at the heart of performance
Zen 7 would introduce a notable evolution of the L2 cache which would increase to 2 MB per core, i.e. double the previous generations. This increase makes it possible to reduce accesses to the main memory and to improve the responsiveness of the processor in loads sensitive to the latency.
But it is especially on the L3 cache side that things become impressive. AMD is reportedly preparing two types of modules for its desktop processors:
The first, unofficially baptized “Silverking†, would remain on a classic configuration with 8 cores and 32 MB of L3 cache. The second, “Silverton†, would mark a clearer break with 16 cores grouped on a single CCD and 64 MB of native L3 cache.
This last point is crucial: by grouping more cores on the same block, AMD reduces inter-CCD communications, which reduces latency and improves performance, particularly in gaming.
Up to 224 MB of L3 cache with 3D V-Cache
As with previous generations, AMD would continue to exploit its 3D V-Cache technology, which consists of stacking cache memory directly on the processor. By combining this approach with the new CCDs, capacities could reach up to 224 MB of L3 cache on some models.
An impressive figure, which far exceeds current standards and which could offer a significant advantage in video games, where cache management plays a determining role in the stability of performance and the reduction of latencies.
Un concept extrême à 72 cœurs en préparation
Among the most notable information is also an internal project called “Steamboat†. This would be based on a 3D stacked architecture, with the cache positioned under the cores in order to maximize the space available for calculation.
A single Steamboat module could integrate up to 36 cores and more than 250 MB of cache. In theory, it would be possible to assemble two of these modules on the same platform to reach a total of 72 cores on the AM5 socket.
In practice, such a processor remains highly unlikely for the general public, due to production costs. This approach would be primarily intended for the server segment, where AMD is already preparing its future EPYC processors, capable of reaching much more extreme configurations.
An architecture designed to widen the gap
With Zen 7, AMD seems to want to go well beyond a simple evolution. Between the massive increase in cache, the densification of cores and the improvement in IPC, all levers are activated to maximize performance, both in gaming and in production.
If this initial information is confirmed, Zen 7 could well redefine the standards of the desktop market. It now remains to be seen how the competition will react to such a rise in power, while the battle around cache and latency is emerging more than ever as the new area of confrontation between AMD and Intel.
