Comparison of upcoming x86 unified memory systems
AMD's upcoming chips leapfrog Intel with 460-690 GB/s bandwidth
AMD is set to dominate the unified memory market with its upcoming Medusa Halo architecture, slated for summer 2027. This system will feature a 50% performance improvement over the current Strix Halo, achieved by expanding memory channels from 4 to 6 and adopting LPDDR6 memory. The result is a massive bandwidth range of 460-690 GB/s, far outpacing competitors. Medusa Halo will be built on Zen 6 cores and RDNA5 graphics, making it ideal for AI and compute-heavy tasks.
Intel's Nova Lake AX, expected in early 2026, will offer up to 341 GB/s using LPDDR5X or possibly LPDDR6 memory. Meanwhile, AMD's Gorgon Halo refresh, launching this summer, delivers 273 GB/s with LPDDR5X-8533—a 15% boost over Strix Halo's 256 GB/s. These systems highlight a rapid evolution in unified memory, crucial for AI inference and high-bandwidth applications. AMD's aggressive roadmap suggests a clear lead in memory performance by 2027.
- AMD Medusa Halo (summer 2027) offers 460-690 GB/s with LPDDR6 and 6 memory channels, a 50% performance improvement.
- Intel Nova Lake AX (early 2026) maxes at 341 GB/s with LPDDR5X/6.
- AMD Gorgon Halo refresh (summer 2025) delivers 273 GB/s, 15% faster than Strix Halo's 256 GB/s.
Why It Matters
Unified memory bandwidth is critical for AI workloads—AMD's leap could redefine high-performance computing standards.