Cache Wars: Intel eDRAM vs. AMD 3D V-Cache and the Future of Intel Nova-Lake S CPU Performance

Cache Wars: Intel eDRAM vs. AMD 3D V-Cache and the Future of Intel Nova-Lake S CPU Performance

In the relentless pursuit of computing power, CPU manufacturers constantly innovate, pushing the boundaries of raw clock speeds and core counts. However, another critical battleground exists: memory cache. By storing frequently accessed data closer to the processing cores, cache significantly reduces latency and boosts performance. Over the years, Intel and AMD have taken distinct, yet equally innovative, approaches to integrate more memory into their CPUs, most notably Intel's past eDRAM and AMD's current 3D V-Cache technology. As we look towards future architectures like Intel's rumored Nova Lake-S, understanding these historical and present strategies provides vital context for the next generation of CPU performance.


1. Intel's eDRAM Experiment (2013-2015): Powering Integrated Graphics

Intel's foray into deeply integrated memory began with eDRAM (embedded DRAM), primarily seen in select Haswell (2013) and more broadly in Broadwell (2015) processors. At its core, eDRAM was a dedicated L4 cache implemented as a separate die within the same CPU package, communicating via a high-speed interposer.

Why eDRAM? The Integrated Graphics Bottleneck

The primary motivation behind eDRAM was to overcome the severe memory bandwidth limitations faced by integrated graphics processors (iGPUs). Traditional iGPUs had to compete with the CPU for access to system RAM (DDR3 at the time), which offered insufficient bandwidth for demanding graphical workloads like gaming. The 128 MB of eDRAM served as an ultra-fast, dedicated buffer for Intel's Iris Pro Graphics (GT3e/GT4e), providing a much wider and lower-latency data path than system memory.

Performance Impact and Notable CPUs

The eDRAM delivered a dramatic uplift in iGPU performance. Benchmarks showed Iris Pro Graphics with eDRAM often delivering double the frame rates of standard Intel HD Graphics solutions without it. This allowed for playable experiences in many popular older AAA titles (e.g., Grand Theft Auto V, BioShock Infinite, Tomb Raider 2013) at 720p/1080p and high settings in esports titles (e.g., League of Legends, Dota 2, Counter-Strike: Global Offensive), even challenging entry-level discrete GPUs of the era like the NVIDIA GeForce GT 740.

Notable CPUs featuring eDRAM included:

  • Haswell (2013): Select mobile (HQ series) and embedded parts with Iris Pro Graphics 5200.
  • Broadwell (2015): The desktop Core i7-5775C and Core i5-5675C were unique as the only socketed desktop CPUs Intel released with eDRAM (Iris Pro Graphics 6200). Various mobile and server (Xeon E3 v4) parts also utilized it.

The Fade of Mainstream eDRAM

Despite its technical prowess, eDRAM phased out from mainstream desktop CPUs after Broadwell. Reasons included: increased manufacturing cost and complexity (two dies on a single package), additional power consumption and die space, and the market reality that for serious gaming, users would opt for more powerful discrete graphics cards, which offered better performance-per-dollar compared to the eDRAM-equipped integrated solutions.


2. AMD's 3D V-Cache (X3D CPUs): The Modern Cache King for Gaming

Fast forward to the 2020s, and AMD introduced its own revolutionary cache technology: 3D V-Cache, prominently featured in their "X3D" series Ryzen CPUs (e.g., Ryzen 7 5800X3D, Ryzen 7 7800X3D). Unlike eDRAM, AMD's 3D V-Cache involves vertically stacking additional L3 cache directly on top of the existing CPU chiplets using TSVs (Through-Silicon Vias).

How 3D V-Cache Boosts CPU Performance

AMD's approach focuses on the CPU cores themselves, not the integrated graphics (which most high-end gaming rigs use a discrete GPU for anyway). Many games are incredibly sensitive to L3 cache size and latency. Game engines frequently access large working sets of data for game logic, AI, physics, and rendering instructions. If this data can reside in an enormous, low-latency L3 cache instead of having to be fetched from much slower system RAM, the CPU can process instructions significantly faster.

This translates to substantial gains in CPU-bound gaming scenarios, leading to higher average framerates and, crucially, improved 1% low framerates and better frametime consistency, resulting in a smoother overall gaming experience. The X3D CPUs have often proven to be the fastest gaming CPUs in the world due to this technology.

Current X3D CPUs and Cache Capacities

As of July 2025, the highest total L3 cache available on an AMD Ryzen X3D CPU for consumer desktops is 128 MB. This is found on models such as:

  • AMD Ryzen 9 7950X3D: 128 MB total L3 cache (96 MB on one CCD with 3D V-Cache, 32 MB on the other standard CCD).
  • AMD Ryzen 9 7900X3D: 128 MB total L3 cache (same distribution as 7950X3D).
  • AMD Ryzen 7 7800X3D: 96 MB total L3 cache (all on a single CCD with 3D V-Cache).
  • Recently released Zen 5 based CPUs like the Ryzen 9 9950X3D and 9900X3D also feature 128MB of total L3 cache.

3. The Fundamental Differences: Charting the Approaches

While both eDRAM and 3D V-Cache aimed to enhance CPU performance through integrated memory, their architectural targets, underlying technologies, and implementation strategies were fundamentally different. The following chart highlights these distinctions:

Feature Intel eDRAM (circa 2013-2015) AMD 3D V-Cache (X3D CPUs) Intel Nova Lake-S (Future, ~2026-2027)
(Speculative / Rumored)
Primary Beneficiary Integrated Graphics (iGPU) CPU Cores (especially in games) CPU Cores (especially in games and cache-sensitive applications)
Cache Level L4 Cache Expanded L3 Cache "bLLC" (big Last Line Cache) - likely an expanded L3 cache, potentially integrated into a base tile
Memory Technology DRAM (Dynamic RAM) SRAM (Static RAM) SRAM (expected for competitive cache performance)
Implementation Separate die on the CPU package, connected to iGPU/CPU Vertically stacked directly on top of the CPU's CCD (Core Complex Die) Likely integrated into a "base tile" underneath the core tiles in a chiplet design (similar to how AMD's 9000 series X3D positions its cache)
Capacity (Typical) 128 MB Adds 64 MB or more on top of existing L3 cache (e.g., 96MB, 144MB total L3) Specific total L3 cache size unconfirmed, but expected to be significantly larger than current non-X3D Intel CPUs to compete with AMD's X3D. Some rumors suggest up to 144MB for top SKUs.
Goal Overcome iGPU memory bandwidth bottleneck with system RAM Provide CPU cores with much larger, lower-latency L3 cache Directly challenge AMD's 3D V-Cache for "leadership gaming performance" by providing a large, low-latency cache for CPU cores.
Impact on Gaming Dramatically improved iGPU performance; made integrated gaming viable against entry-level discrete GPUs. Significant uplift in CPU-bound games (higher FPS, better 1% lows) when paired with a discrete GPU. Expected to significantly boost gaming performance, aiming to match or surpass AMD's X3D CPUs in cache-sensitive titles.
Current Status Phased out from mainstream desktop CPUs; a niche historical solution. Actively developed and a key feature of AMD's top gaming CPUs. Rumored for late 2026 / early 2027 launch; details are from leaks and highly speculative.
Disclaimer: Information regarding Intel's Nova Lake-S CPUs, especially concerning specific architectural details like cache implementation and capacity, is based on unconfirmed leaks, rumors, and industry speculation. Official details will only be available closer to the product's launch.

4. Intel's Future Counter: Nova Lake-S and the Evolving Cache Landscape

As AMD continues to leverage its 3D V-Cache technology for gaming dominance, Intel is not standing still. Rumors indicate that Intel's future desktop CPU architecture, codenamed Nova Lake-S (expected around late 2026 to 2027), is being designed as a direct competitor to AMD's X3D line, particularly in gaming.

Speculation suggests that Nova Lake-S will feature a significantly expanded cache solution, potentially referred to as "bLLC" (big Last Line Cache). This is anticipated to be a large, low-latency L3 cache, likely implemented within a chiplet design, possibly on a "base tile" underneath the core-containing tiles. While specific capacities are unconfirmed, the goal is clear: to offer a competitive large cache solution that can rival or exceed the gaming performance benefits seen from AMD's 3D V-Cache, especially in CPU-bound scenarios. This marks a strategic shift for Intel, acknowledging the critical role of large on-die caches for high-end gaming.


The history of CPU development is one of continuous innovation, often driven by intense competition. Intel's eDRAM was an early, albeit ultimately niche, attempt to solve a specific memory bottleneck for integrated graphics. AMD's 3D V-Cache represents a highly successful modern solution for CPU-bound gaming performance, proving the value of massive on-die L3 caches. As Intel gears up for architectures like Nova Lake-S, the "cache wars" are poised to intensify, promising even greater performance for gamers and demanding users in the years to come. These technological battles ultimately benefit consumers, pushing the boundaries of what's possible in personal computing.