Next-Gen Console Memory: Why GDDR7 Won Out and How the Unified Pool Works

ByMarcus Vale

Both consoles are going with GDDR7 — and it’s not a close call

The short answer to the GDDR-vs-DDR question: GDDR7, full stop. Leaked specs for both Sony’s next PlayStation and Microsoft’s next Xbox point to GDDR7, which aligns with how console APU design has trended since the PS4 era. DDR memory wins on latency, but a GPU processing textures and framebuffer data needs raw throughput first. GDDR7 delivers around 32 Gbps per pin — roughly 60% more bandwidth than GDDR6. That gap is why DDR doesn’t seriously compete for this role.

What the leaked specs say

On the PlayStation side, the most widely cited leaks describe roughly 30 GB of GDDR7 on a 160-bit bus, putting estimated bandwidth around 640 GB/s — a significant jump from the PS5’s 16 GB GDDR6 at 448 GB/s. Some leaks have floated numbers as high as 40 GB, so nothing is confirmed. Sony’s next console, often referenced under the codename Orion and developed via something called Project Amethyst with AMD, is rumoured to pair that memory with a custom APU on TSMC’s 3nm node: around 52–54 RDNA 5 compute units, roughly 8 Zen 6 CPU cores, targeting somewhere in the 34–40 teraflop range. A 2028–2029 release has been floated due to reported memory supply constraints.

Microsoft’s next console — often referenced as Project Helix, using AMD’s Magnus chiplet design — has appeared in leaks suggesting up to 48 GB of GDDR7. That ceiling is almost certainly for a dev kit or high-end SKU; a realistic consumer figure looks more like 24–32 GB. The leaked silicon points to 68 RDNA 5 compute units and a mix of Zen 6 and Zen 6c CPU cores. Microsoft has reportedly been targeting a 2026–2027 launch window, which would put it ahead of the PS6 if those timelines hold.

The shared pool question — how consoles differ from PCs

On a typical gaming PC, the GPU has its own discrete VRAM and the CPU has system RAM, and moving data between them crosses a PCIe bus with real overhead. Consoles do it differently. Since the PS4 introduced GDDR5 as a unified pool, Sony and Microsoft have used a single shared memory pool where the CPU and GPU both operate from the same physical chips. No separate VRAM. No discrete GPU hanging off a PCIe slot.

The PS5’s 16 GB GDDR6 is a useful reference point. All 16 GB lives in one physical pool, but not all of it runs at the same speed. Sony splits it into a 2 GB region running at full bandwidth and a larger 14 GB region at slightly lower speeds — developers choose where to place assets. The OS takes a slice off the top (roughly 3.5 GB across most firmware versions), and the rest is the working budget games actually get. Developers know exactly what they have. No driver surprises, no contention with a host OS juggling background processes.

Next-gen consoles will almost certainly follow the same pattern. A larger total pool, a portion held back for the OS, and the rest fully exposed to the game. With 30 GB or more to work with, even a generous system reservation leaves developers far more headroom than the roughly 12.5 GB available on PS5.

Why GDDR7 bandwidth matters more in a unified design

On a PC, a GPU on PCIe 5.0 moves data across that bus at around 64 GB/s. In a unified console design, the CPU and GPU address the same memory directly — no bus crossing required. The full pool bandwidth is available to both processors simultaneously. That’s why consoles have historically punched above their spec-sheet weight compared to PC hardware at an equivalent GPU teraflop count.

Micron, one of the primary GDDR7 manufacturers, has published figures showing peak bandwidth above 1.5 TB/s in certain configurations. Whether next-gen consoles approach that ceiling depends on final bus width decisions. But the direction is clear: more bandwidth, same unified model, same OS-shared pool — just with considerably more room to work with.

Sources


Similar Posts