Marvell Unveils Custom HBM Solution for AI Accelerators, Promising Higher Performance and Efficiency

Marvell Introduces Custom High-Bandwidth Memory Solution

Marvell, a leading technology company, has unveiled its new custom high-bandwidth memory (CHBM) solution at its Analyst Day 2024. This innovative technology, developed in collaboration with major memory manufacturers, is designed to enhance the performance of AI accelerators and optimize cloud infrastructure 1

1

Key Features and Advantages

The CHBM solution offers several significant improvements over standard high-bandwidth memory:

1. Increased Logic Density: Marvell claims its proprietary die-to-die I/O allows for up to 25% more logic in custom XPUs.
2. Enhanced Memory Capacity: The design potentially accommodates up to 33% more CHBM memory packages next to compute chiplets.
3. Power Efficiency: Memory interface power consumption is expected to be reduced by up to 70%.
4. Improved Bandwidth: The new Marvell die-to-die HBM interface boasts a bandwidth of 20 Tbps/mm, a significant increase from the current 5 Tbps/mm in standard HBM 1

   1

   .

Collaboration with Memory Manufacturers

Marvell's CHBM initiative involves partnerships with leading memory manufacturers, including Micron, Samsung, and SK hynix. This collaboration is crucial for ensuring the widespread availability and adoption of custom high-bandwidth memory solutions 2

2

Related Stories

Nvidia unveils NVHBM custom memory promising 30% higher bandwidth and 15% lower power than HBM4e

SK hynix Leads the Charge in Next-Gen AI Memory with World's First 12-Layer HBM4 Samples

SK Hynix Accelerates HBM4 Development to Meet Nvidia's Demand, Unveils 16-Layer HBM3E

Marvell and NVIDIA Collaborate to Enhance AI Infrastructure with Custom Silicon Solutions

Industry Perspectives

Several industry leaders have expressed their support and excitement for Marvell's CHBM solution:

• Will Chu, SVP at Marvell, emphasized the importance of tailoring HBM for specific performance and power requirements in the AI era 2

  2

  .
• Raj Narasimhan from Micron highlighted the potential for increased memory capacity and bandwidth to help cloud operators scale their infrastructure efficiently 2

  2

  .
• Harry Yoon of Samsung Electronics noted that optimizing HBM for specific XPUs and software environments would greatly improve cloud infrastructure performance 2

  2

  .

Future Prospects

Marvell's vision extends beyond the current CHBM solution. The company is already looking ahead to bufferless memory with a bandwidth of 50 Tbps/mm, further pushing the boundaries of memory performance 1

1

As AI continues to drive technological advancements, innovations like Marvell's CHBM are poised to play a crucial role in shaping the future of cloud computing and AI acceleration. The collaboration between Marvell and leading memory manufacturers sets the stage for a new era of customized, high-performance memory solutions tailored to the specific needs of AI workloads.