Hbm4
PulseAugur coverage of Hbm4 — every cluster mentioning Hbm4 across labs, papers, and developer communities, ranked by signal.
- 2026-06-05 research_milestone Nvidia CEO Jensen Huang announced that three memory manufacturers have qualified to supply HBM4, indicating a significant step in the supply chain for advanced AI hardware. source
4 day(s) with sentiment data
-
Samsung, SK Hynix See AI-Driven Memory Demand Surge Amid HBM4 Competition
Samsung and SK Hynix are experiencing increased demand for their memory products, driven by the AI industry's transformation. Competition for HBM4 is intensifying, alongside a growing need for server DDR5 and enterprise…
-
NVIDIA revises Rubin GPU specs, reducing memory and NVLink bandwidth
NVIDIA has revised the specifications for its upcoming Rubin GPU, reducing the HBM4 memory bandwidth and NVLink interconnect bandwidth compared to earlier projections. The Rubin GPU will now offer 19.2 TB/second of memo…
-
Microsoft unveils Maia 300 AI accelerator with 288GB HBM4 memory
Microsoft has unveiled its next-generation AI accelerator, the Maia 300, at OCP 2026. This new chip features eight HBM4 stacks, providing 288GB of memory and 18TB/s of memory bandwidth, along with 3.2TB/s of scale-up bi…
-
Micron pivots to 3D DRAM as Samsung leads HBM revenue
Micron is shifting its strategy away from 2D DRAM manufacturing after two decades, reallocating capital towards 3D DRAM and advanced NAND technologies. This move includes focusing on 1-delta density improvements and exp…
-
AMD eyes deeper Samsung partnership for AI memory supply and chip orders
AMD is seeking to deepen its partnership with Samsung, aiming to secure a more robust supply of High Bandwidth Memory (HBM) for its AI accelerators. Samsung, in turn, is reportedly interested in leveraging this relation…
-
Nvidia adopts custom HBM, freeing up GPU die space for compute
SemiAnalysis reports that Nvidia is adopting custom High Bandwidth Memory (HBM) solutions, significantly reducing the die space required for memory controllers and PHYs. This custom approach, particularly with Samsung's…
-
Samsung custom HBM4 shrinks interface, frees GPU die space · 3 sources tracked
Samsung is developing custom High Bandwidth Memory (HBM) solutions that significantly reduce the physical footprint and improve efficiency compared to standard HBM4. These custom designs replace wide PHY interfaces with…
-
AI boom strains memory chip supply, impacting consumer electronics
The global demand for AI data centers is creating a severe shortage of memory chips, particularly High Bandwidth Memory (HBM). This has led to inventory levels for major manufacturers like Samsung and SK Hynix dropping …
-
Hybrid Bonding for HBM Delayed to Decade's End, Logic Adoption Advances
Hybrid bonding, a copper-to-copper chip joining technique, is in high-volume production for logic chips but its adoption for High Bandwidth Memory (HBM) has been delayed. A recent JEDEC decision allows HBM4 to continue …
-
NVIDIA cuts HBM in Rubin Ultra due to cost surge · 3 sources tracked
NVIDIA has reduced the High Bandwidth Memory (HBM) configuration for its Rubin Ultra system, downgrading from HBM4E 12-Hi (384GB) to HBM4 8-Hi (192GB). This decision stems from recent price hikes in HBM and DRAM, which …
-
Samsung plans integrated compute and memory with zHBM roadmap
Samsung has outlined a three-phase strategy to integrate logic and compute capabilities directly into its High Bandwidth Memory (HBM) architecture. The company's zHBM approach aims to place DRAM directly on top of the p…
-
OpenAI's 'Jalapeño' Accelerator Outperforms Nvidia Blackwell in Efficiency
SemiAnalysis has released details about OpenAI's new AI accelerator, codenamed "Jalapeño." This accelerator is reportedly more power-efficient than NVIDIA's Blackwell architecture, offering superior tokens per megawatt.…
-
d-Matrix unveils 3D DRAM AI accelerator with 100 TB/s bandwidth
d-Matrix has unveiled its Raptor AI accelerator, which features a novel 3D stacking architecture. This design places a TSMC 4nm compute die directly on top of a custom-designed DRAM die, achieving an unprecedented 100 T…
-
OpenAI unveils custom Jalapeño ASIC for inference workloads
OpenAI has developed an in-house inference ASIC named Jalapeño, designed in collaboration with Broadcom. This custom chip aims to provide the optimal compute platform for OpenAI's inference workloads, focusing on perfor…
-
Intel unveils practical AI accelerator Crescent Island for inference
Intel has revealed new details about its Crescent Island AI accelerator, built on the Xe3P architecture, at the Hot Chips symposium. This accelerator is designed for inference tasks, featuring a 350W air-cooled PCIe car…
-
Nvidia Rubin Ultra GPU faces HBM downgrade, capacity slashed from 1TB to 192GB
Nvidia's upcoming Rubin Ultra GPU has reportedly seen a significant downgrade in its High Bandwidth Memory (HBM) capacity. Initial previews suggested 1TB of HBM, but the production version is now expected to ship with 1…
-
AI hardware advances: Cerebras stacks DRAM, HBM faces growing costs
At Hot Chips 2026, leading memory and chip manufacturers discussed the future of AI hardware, focusing on advancements in wafer-scale processing and High Bandwidth Memory (HBM). Cerebras unveiled its roadmap for stacked…
-
Samsung's 2nm chip yields lag, impacting Tesla deal and TSMC competition
Samsung is facing significant challenges with its foundry business, particularly concerning yield rates for its 2nm process, which are reportedly around 55%. Despite securing a substantial $16.5 billion contract with Te…
-
AMD Helios aims to challenge NVIDIA with rack-scale AI system
AMD is reportedly developing a new rack-scale AI system called Helios, designed to compete with NVIDIA's infrastructure offerings. This system is expected to feature 72 MI455X accelerators and a substantial 31TB of HBM4…
-
NVIDIA reportedly testing lower-memory Rubin Ultra GPUs amid HBM shortage
NVIDIA is reportedly exploring configurations for its Rubin Ultra GPU that feature reduced memory capacities, potentially as low as 192 GB. This move comes amid a persistent shortage of high-bandwidth memory (HBM), whic…