Samsung details GDDR7, already in production, with speeds up to 36Gbps, doubling GDDR6's 18Gbps, and plans to release new V-NAND chips with a 230-layer design
Francisco Pires / Tom's Hardware :
Context & Ripple Effects
This announcement caps a fast cadence for Samsung's graphics memory line: just months earlier it had developed a 24Gbps GDDR6 part built on its EUV process, a 30%+ jump over the 18Gbps generation this story measures against. Detailing GDDR7 as already in production at up to 36Gbps moves the roadmap from lab milestone to shipping silicon.
The V-NAND half of the story follows the same playbook on a different axis: a planned 230-layer design sits one rung below the 9th-generation 286-layer chips Samsung later put into mass production, and the eventual first-generation GDDR7 completion landed at 32Gbps rather than the 36Gbps ceiling teased here — both signs of how these roadmaps actually resolved.
First-order effects
- GPU and accelerator designers gain a shipping memory option with double the per-pin bandwidth of the 18Gbps GDDR6 baseline, letting them raise frame buffers' throughput without widening the bus.
- Samsung's own product stack gets refreshed on two fronts at once — graphics DRAM and V-NAND — tightening its position across the memory categories it sells.
Second-order effects
- Rival DRAM vendors are pushed to match GDDR7-class speeds or cede the high-bandwidth graphics slot, repeating the pattern set when Samsung's 24Gbps GDDR6 reset the bar mid-2022.
- NAND competitors face continued pressure on layer counts, since Samsung's stated 230-layer plan signaled that vertical stacking — not planar shrinks — is where density gains come from.
Third-order effects
- If the doubling cadence holds, memory bandwidth becomes the scheduled variable in accelerator design rather than a constraint negotiated around, with each GDDR generation pre-announced far enough ahead for GPU roadmaps to be built against it.
- NAND economics consolidate around whoever scales layers fastest, since the 230-to-286-layer progression shows bit density and I/O speed now arrive together in each new generation.
The trend: Memory vendors are racing bandwidth-per-pin and NAND layer counts on public, pre-announced roadmaps, turning memory specs into the pacing item for next-generation GPU and storage designs.