Samsung Begins Commercial Shipments of HBM4 and LPDDR6X Memory
What happened
Samsung and SK Hynix have started commercial shipments of HBM4, with Samsung supplying NVIDIA's latest AI accelerators. Separately, Samsung is also shipping its next-generation LPDDR6X memory to Qualcomm for use in future SoCs. The high demand for advanced memory for AI servers is reportedly creating supply constraints and price turbulence for the broader PC and embedded markets.
Why it matters
- HBM4 doubles the memory interface width to 2,048 bits from HBM3E's 1,024 bits, a fundamental architectural change that enables a significant bandwidth increase to over 1.6 TB/s per stack. - Samsung's HBM4 implementation uses an advanced 1c DRAM process with a 4nm logic base die, achieving processing speeds of 11.7 gigabits-per-second (Gbps), which is 46% faster than the 8 Gbps JEDEC industry standard for HBM4. - NVIDIA's next-generation AI accelerator platform, codenamed "Rubin," is expected to be a primary consumer of HBM4 memory, with future GPUs projected to feature 288GB of HBM4 capacity. - The LPDDR6 standard increases performance by moving to a 24-bit channel width, up from 16 bits in LPDDR5X, and targets data rates as high as 14.4 Gbps, compared to the ~10.7 Gbps peak for L
Key numbers
- Samsung and SK Hynix have started commercial shipments of HBM4, with Samsung supplying NVIDIA's latest AI accelerators.
- Separately, Samsung is also shipping its next-generation LPDDR6X memory to Qualcomm for use in future SoCs.
- - HBM4 doubles the memory interface width to 2,048 bits from HBM3E's 1,024 bits, a fundamental architectural change that enables a significant bandwidth increase to over 1.6 TB/s per stack.
- Samsung's HBM4 implementation uses an advanced 1c DRAM process with a 4nm logic base die, achieving processing speeds of 11.7 gigabits-per-second (Gbps), which is 46% faster than the 8 Gbps JEDEC industry standard for HBM4.
What happens next
- NVIDIA's next-generation AI accelerator platform, codenamed "Rubin," is expected to be a primary consumer of HBM4 memory, with future GPUs projected to feature 288GB of HBM4 capacity.
- Separately, Samsung is also shipping its next-generation LPDDR6X memory to Qualcomm for use in future SoCs.
Quick answers
What happened in Samsung Begins Commercial Shipments of HBM4 and LPDDR6X Memory?
Samsung and SK Hynix have started commercial shipments of HBM4, with Samsung supplying NVIDIA's latest AI accelerators. Separately, Samsung is also shipping its next-generation LPDDR6X memory to Qualcomm for use in future SoCs. The high demand for advanced memory for AI servers is reportedly creating supply constraints and price turbulence for the broader PC and embedded markets.
Why does Samsung Begins Commercial Shipments of HBM4 and LPDDR6X Memory matter?
HBM4 doubles the memory interface width to 2,048 bits from HBM3E's 1,024 bits, a fundamental architectural change that enables a significant bandwidth increase to over 1.6 TB/s per stack. Samsung's HBM4 implementation uses an advanced 1c DRAM process with a 4nm logic base die, achieving processing speeds of 11.7 gigabits-per-second (Gbps), which is 46% faster than the 8 Gbps JEDEC industry standard for HBM4. NVIDIA's next-generation AI accelerator platform, codenamed "Rubin," is expected to be a primary consumer of HBM4 memory, with future GPUs projected to feature 288GB of HBM4 capacity. The LPDDR6 standard increases performance by moving to a 24-bit channel width, up from 16 bits in LPDDR5X, and targets data rates as high as 14.4 Gbps, compared to the ~10.7 Gbps peak for L
