Tag

Hbm3e

All articles tagged with #hbm3e

AMD's Halo Station Brings On‑Device AI to a 96‑Core Threadripper and MI350P GPUs at IFA 2026
technology13 hours ago

AMD's Halo Station Brings On‑Device AI to a 96‑Core Threadripper and MI350P GPUs at IFA 2026

AMD unveiled the Threadripper Halo Station at IFA 2026, a liquid‑cooling workstation built around the 96‑core Threadripper PRO 9995WX and up to four Instinct MI350P accelerators. The system targets running trillion‑parameter AI models locally without cloud access, with up to 2 TB of DDR5 RAM and configurations reaching up to 576 GB of HBM3E across four MI350P cards (288 GB in a two‑card setup). Full specs, availability, and pricing were not announced.

Micron’s AI-Memory Upside Persists Despite Sector Rout
business1 month ago

Micron’s AI-Memory Upside Persists Despite Sector Rout

Micron Technology trades well below consensus targets despite a quarter with 345% revenue growth and margins near 85%, a rally triggered by ChangXin Memory Technologies’ IPO-induced memory-sector rout. Bulls like Ben Reitzes (Melius) maintain a Street-high $2,200 price target, arguing tight HBM3E/HBM4 capacity and multiyear contracts with AI accelerators (NVIDIA/AMD) could sustain premium pricing and durable margins, implying upside from today’s around $820 toward and beyond $2,000. The broader memory group is pressured by fears of a 2027 glut, but the consensus target sits near $1,507, signaling substantial potential upside if the bull thesis holds.

"Micron Ramps Up Production of High-Speed Memory Chips for Nvidia's AI Semiconductors"
technology2 years ago

"Micron Ramps Up Production of High-Speed Memory Chips for Nvidia's AI Semiconductors"

Micron Technology has commenced mass production of its high-bandwidth memory semiconductors, HBM3E, for use in Nvidia's latest AI chip, leading to a more than 5% increase in its shares. The HBM3E is expected to consume 30% less power than rival offerings, catering to the growing demand for chips powering generative AI applications. Nvidia plans to incorporate the chip in its upcoming H200 graphic processing units, set to surpass the current H100 chip. This move has raised investor optimism about Micron's ability to withstand a slow recovery in other markets, with HBM being one of its most profitable products.

"Micron Commences Mass Production of Memory Chips for Nvidia's AI Semiconductors"
technology2 years ago

"Micron Commences Mass Production of Memory Chips for Nvidia's AI Semiconductors"

Micron Technology has commenced mass production of its high-bandwidth memory semiconductors, HBM3E, for use in Nvidia's upcoming AI chips, which are expected to consume 30% less power than rival offerings. This move has driven Micron's shares up and is anticipated to meet the increasing demand for chips powering generative AI applications. Nvidia will incorporate the HBM3E chip in its next-generation H200 graphic processing units, set to surpass the current H100 chip. The demand for high-bandwidth memory chips in AI applications has also raised investor optimism about Micron's resilience in the market.

Samsung Unveils Next-Gen Memory Innovations for Hyperscale AI Era
technology2 years ago

Samsung Unveils Next-Gen Memory Innovations for Hyperscale AI Era

Samsung has unveiled its next-generation memory technologies, including HBM3E "Shinebolt" for AI and data centers, GDDR7 for gaming graphics, and LPDDR5x CAMM2 for slimming down mobile designs. The HBM3E offers impressive speeds of 9.8 Gbps per pin, while GDDR7 provides a 40% performance boost and 20% power efficiency improvement compared to GDDR6. The LPDDR5x CAMM2 is expected to be a game changer in the PC and laptop DRAM market. Samsung aims to offer a custom turnkey service combining advanced packaging technologies and foundry offerings.

Samsung Unveils 'Shinebolt': Revolutionary 36GB HBM3E Memory with Lightning-Fast 9.8 Gbps Speeds
technology2 years ago

Samsung Unveils 'Shinebolt': Revolutionary 36GB HBM3E Memory with Lightning-Fast 9.8 Gbps Speeds

Samsung has announced Shinebolt, its HBM3E memory, which offers higher capacities and greater memory bandwidth than its predecessor, HBM3. Shinebolt will have a maximum capacity of 36GB and a memory clockspeed of 9.8Gbps/pin, making it faster and more efficient than previous HBM memory. Samsung also provided an update on its GDDR7 memory, which will feature a switch to PAM3 encoding, allowing for improved memory transfer rates. GDDR7 is expected to have 50% lower standby power consumption than GDDR6. Both Shinebolt and GDDR7 are targeted at high-end processors, particularly in the AI market.