- US - English
- China - 简体中文
- India - English
- Japan - 日本語
- Malaysia - English
- Singapore - English
- Taiwan – 繁體中文
Micron technology helps usher in the AI data center of the future
- Near memory
- Main memory
- Expansion memory
- Local SSD data cache
- Networked data lakes
Unlocking the potential of AI data centers
Within every AI server box lives a pyramid or hierarchy of memory and storage to support fast groundbreaking AI. When built with Micron’s leading technology, data center bottlenecks are reduced, sustainability and power efficiencies are increased, and the total cost of ownership is improved.
Interact with the pyramid to explore our AI portfolio
Take a peek inside the box
Frequently asked questions
1 As compared to previous 1α node generation.
2 Based on JEDEC specification.
3 Measured data in pJ/bit compared to commercially available (June 2023) competitive 3DS modules.
4 Empirical Intel Memory Latency Checker (Intel MLC) data comparing 128GB MRDIMM 8800MT/s against 128GB RDIMM 6400MT/s.
5 Empirical Stream Triad data comparing 128GB MRDIMM 8800MT/s against 128GB RDIMM 6400MT/s at 1TB.
6 Empirical OpenFOAM task energy comparing 128GB MRDIMM 8800MT/s against 128GB RDIMM 6400MT/s.
7 Compared to LPDDR5X 8533 Mbps
8 Compared to previous generation
9 MLC bandwidth using 12-channel 4800MT/s RDIMM+ 4x256GB CZ120 vs. RDIMM only.
10 Performance comparisons are based on publicly available data information for performance-focused Gen5 SSDs with 1 DWPD 7.68TB capacity SSDs available at product launch. Sequential and random throughput at QD512. Several public sources note big accelerator memory (BaM) such as: https://www.tomshardware.com/news/nvidia-unveils-big-accelerator-memory-solid-state-storage-for-gpus(opens in a new tab) and GPU-initiated direct storage (GIDS) for graph neural network (GNN) training workloads using NVIDIA H100 GPU as tested in Micron's labs against performance-focused Gen5 SSDs.
11 Checkpoint workload modeled on Llama3 405B parameter LLM. The model represents an 8 GPU server. Checkpoint size is 415GB. Figures represent the time required for one checkpoint, the SSD energy consumed for one checkpoint, and the SSD throughput during checkpoint operations compared to the Solidigm D5-P5336. See Micron 6550 ION SSD AI Tech Brief(opens in a new tab) for details.
12 These comparisons use publicly available competitor information from published sources at the time of the 6550 ION launch, with the 6550 ION using a maximum power of 20W and competitive drives using 25W, resulting in up to 20% less maximum power consumption for the 6550 ION.
13 The Micron 6550 ION offers a capacity of up to 61.44TB. E3.S-capable servers can be configured with up to 20 SSDs in 1U.