SamSung HBM2E Flashbolt - 8GB

Boost the power of AI

Accelerating, expanding, and ensuring supercomputing and AI technology

The advancement of supercomputing and AI-based technologies require the highest caliber of memory that can meet industry needs for bandwidth, capacity, and efficiency. HBM2E Flashbolt can boost the power of AI, handle more Big Data through its expanded capacity, and provide high-bandwidth.

The driving force for high performance computing

Building upon the experience of leading high-end technological advancement across various industries such as AI algorithm, data science, Autonomous driving, 5G and etc., Samsung once again establishes its pioneering role in the memory industry with the first ever HBM2E solution.

High bandwidth for the most advanced tasks

HBM2E Flashbolt features a processing speed of 3.6 Gbps, which is approximately 1.5 times faster than the previous generation HBM.
On top of this rapid speed, HBM2E Flashbolt achieves high bandwidth through TSV technology that helps quickly process large amounts of data, improving AI training efficiency by about 6 times the usual rate*.

*Data based on a comparison of 8 GPUs vs. 72 CPUs at cloud service

Doubled capacity to handle more data

HBM2E Flashbolt offers about twice the capacity of the previous generation HBM solution by stacking eight layers of 10nm-class 16 Gb DRAM dies. A larger capacity allows the development of deeper neural networks that greatly improves the speed of acquiring results for Big Data analytics.

※ Source from Samsung Datasheet (Available upon request)

Higher performance using less power

HBM2E Flashbolt boasts an approximately 18% enhancement in power efficiency and about 1K more
power bumps compared to the previous generation HBM solution. This ensures a more secure provision of power that consumes less energy.

※ Source from Samsung Datasheet (Available upon request)

More reliable with On Die ECC

HBM2E Flashbolt provides high reliability and stability with On Die ECC solution. ODECC enables self-correction of internal errors to enhance recovery of corrupted data and lower the frequency of refreshes. Single bit errors are reduced for overall improvement of data reliability as well.