Recent rumors indicate that both AMD and Intel are developing future CPUs with significantly increased L3 cache, potentially targeting local large language model (LLM) performance rather than solely gaming.
AMD gained a gaming advantage with its Ryzen 5000 series CPUs featuring 3D V-Cache, which provides a substantial L3 cache boost. Intel is reportedly preparing a response with its next-generation Nova Lake designs. A leak last week revealed Intel is working on a CPU with two “Big Last Level Cache” (BLLC) chips, described as Intel’s equivalent of 3D V-Cache.
Shortly after the Intel leak, information surfaced about an AMD dual CCD V-Cache CPU. This AMD processor could potentially launch in 2024, significantly earlier than Intel’s BLLC alternative, which is not expected until 2026 with the Nova Lake CPU line.
Despite the excitement for gaming, AMD previously indicated that incorporating more than one V-Cache chip was neither cost-effective nor significantly beneficial for gaming performance. This is presumably because most games do not utilize more than the eight cores found on a single Ryzen chiplet in their high-end processors.
However, these high-cache CPUs may be primarily designed to accelerate AI performance. According to 1usmus, a notable AMD overclocking tool creator quoted by VideoCardz, “double V-Cache CPUs make sense in the AI age, because they can help accelerate inference in large language models.”
In January 2025, AMD stated that developing such a solution was not economically viable and is a highly specialized niche. Now I can say that it does make sense, because accelerated inference of large language models provides a boost, said 1usmus on August 4, 2025.
Additionally, concerns have been raised that asynchronous cache across cores in Windows could lead to performance degradation over time. A dual V-Cache configuration might allow both AMD and Intel to simplify future designs while simultaneously enhancing AI capabilities.
