Marvell Unveils Custom High-Bandwidth Memory Architecture for AI Cloud Acceleration

Power consumption is reduced by 70% and silicon real estate is optimized for performance and lower TCO in the new design.

Author's Avatar
Dec 11, 2024
Summary
  • A proprietary high-bandwidth memory computing architecture from Marvell Technology boosts cloud AI accelerator performance by 25%.
Article's Main Image

Made to speed cloud-based artificial intelligence processors, Marvell Technology, Inc. (MRVL, Financials), a rival of Nvidia (NVDA, Financials), created an HBM computer architecture.

Marvell's stock performance indicates their ongoing improvement. MRVL shares are at $106.68 as of December 11, 2024, 11% less than their 52-week high of $119.88 this month. Rising from its 52-week low of $53.19, the stock gained about 100%.

Created by Micron Technology, Inc. (MU, Financials), Samsung Electronics Co. Ltd. (SSNLF, Financials), and SK hynix Inc. (HXSCF, Financials), the new architecture increases memory capacity and powers innovative XPU designs. According to Marvell, this concept increases computational power by 25% and memory space by 33% while lowering power consumption by 70%.

Marvell creates the HBM memory module that decreases total cost of ownership for cloud operators by integrating compute chips, HBM stacks, and innovative packaging. Senior vice president and general manager of Marvell's Custom, Compute, and Storage Group, Will Chu underlined the requirement of building HBM for speed and power constraints and noted that this is a new approach of constructing AI accelerators.

Field leaders in particular embrace this initiative. To let cloud infrastructure function quicker and use less power, Harry Yoon, Executive Vice President of Samsung Electronics, said that HBM may be altered to work better with specific XPUs and software systems. Working with Marvell to improve HBM technology delighted Vice President of DRAM Technology at SK Hynix America Sunny Kang.

In its fiscal third quarter, Marvell generated $1.52 billion—6.87% more than Wall Street projected ($0.43 per share). The organization is focused on strengthening its data base to manage the rising volume of artificial intelligence projects.

Disclosures

I/we have no positions in any stocks mentioned, and have no plans to buy any new positions in the stocks mentioned within the next 72 hours. Click for the complete disclosure