AI accelerators
PulseAugur coverage of AI accelerators — every cluster mentioning AI accelerators across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
Doosan Group invests KRW970B to boost AI accelerator component production
Doosan Group is significantly expanding its copper-clad laminate production capacity in South Korea and China. This investment, valued at approximately KRW 970 billion, is driven by the increasing demand for components …
-
Qualcomm validates stacked AI memory architecture for 2027 accelerators
Qualcomm has successfully validated its High Bandwidth Compute (HBC) memory architecture, a key component for its upcoming Dragonfly AI250 data center accelerators. This stacked memory design places DRAM directly on top…
-
TIRx compiler optimizes ML performance across GPUs and AI accelerators
Researchers have developed TIRx, a new ML compiler designed to optimize performance across diverse hardware like GPUs and AI accelerators. TIRx introduces an explicit execution-scope abstraction and uses named axes for …
-
China's CXMT prioritizes AI memory production despite losses · 1 source tracked
China's Changxin Memory Technologies (CXMT) is prioritizing the production of high-bandwidth memory for AI accelerators, even at a financial loss. This strategic move, supported by significant state subsidies, aims to f…
-
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…
-
Altman-backed Volantis integrates photonics into AI accelerators to bypass memory wall
Volantis, a startup backed by Sam Altman, has unveiled a novel approach to AI acceleration by integrating photonics directly into its hardware. This method aims to overcome the memory wall, a significant bottleneck in c…
-
AI Hardware Supply Chain: Focus on Third Derivative Enablers
SemiAnalysis highlights the critical, often overlooked, "third derivative" companies in the semiconductor supply chain that enable advanced AI hardware. These include material suppliers like Ajinomoto, which dominates t…
-
Samsung unveils zHBM memory stacked directly on AI accelerators
Samsung has unveiled a prototype of its zHBM memory technology, designed to be stacked directly onto AI accelerators. This innovative approach aims to significantly improve data transfer speeds and efficiency for AI wor…
-
New research tackles LLM KV cache optimization for efficient inference · 10 sources tracked
Multiple research papers explore novel techniques for optimizing the Key-Value (KV) cache in large language models (LLMs) to improve inference efficiency and reduce memory overhead. These methods include dynamic cache o…
-
TSMC's AI chip packaging bottleneck creates opening for Intel
TSMC's CoWoS packaging technology, crucial for integrating AI accelerators with High Bandwidth Memory, is facing significant cost and yield challenges due to the immense size of silicon interposers. These limitations ar…
-
AI drives memory chip prices to stratospheric levels, benefiting major manufacturers
Memory chip prices, particularly for high-bandwidth memory (HBM) and NAND flash, are experiencing a significant surge, with Gartner projecting the market to reach $1.6 trillion. This boom is largely driven by the insati…
-
Samsung unveils AI-focused memory tech, impacting consumer prices
Samsung has unveiled new memory and storage technologies, including zHBM, which vertically stacks HBM memory directly above AI accelerators to boost performance up to eight times. The company also introduced V10 BV-NAND…
-
Google eyes 2028 AI accelerator dominance, may tap Intel Foundry
An analyst report from Fubon Research suggests Google plans to produce 12-15 million TPU v9 AI accelerators by 2028, potentially exceeding Nvidia's projected shipments for that year. To meet this ambitious target, Googl…
-
FastTPS method accelerates LLM inference on AI accelerators
A new method called FastTPS has been developed to accelerate the token phase of large language model (LLM) inference on AI accelerators. This method addresses the inherent low parallelism and memory overhead issues, par…
-
Nvidia offers AI accelerators for revenue share
Nvidia is offering a new business model for its AI accelerators, allowing customers to pay with a share of their revenue instead of an upfront purchase. This approach aims to make advanced AI hardware more accessible to…
-
Quantum Neural Networks Compared for Semiconductor Defect Classification
A new research paper explores the application of quantum neural networks (QNNs) for classifying defects in semiconductor wafer maps, a critical step for improving manufacturing yield. The study directly compares continu…
-
Qualcomm integrates AI compute into accelerators, researchers reduce AI math burden
Qualcomm is proposing a new approach to AI infrastructure by integrating compute capabilities directly into their AI accelerators. This design aims to overcome the memory wall by placing computation closer to the data. …
-
Apache TVM launches TIRx compiler for evolving ML kernels and hardware
Apache TVM has launched TIRx, an open-source compiler stack designed for machine learning kernels and evolving hardware. This new system allows for hardware-native DSLs and compilation to GPUs and specialized AI acceler…
-
Frore's LiquidJet Nexus boosts AI accelerator cooling, claims 10% performance gain
Frore Systems has unveiled its LiquidJet Nexus coldplate, designed to cool AI accelerators like Nvidia's Vera Rubin and Grace Blackwell superchips. This new monolithic water block utilizes semiconductor manufacturing te…
-
Astera Labs ships 320-lane PCIe 6.0 switch for 80-accelerator AI clusters
Astera Labs has unveiled its Scorpio X-Series 320 Lane Smart Fabric Switch, a significant advancement in data center interconnectivity. This PCIe 6.0 switch boasts 320 lanes and 20 Tbps of bandwidth, enabling the connec…