设计工具
高带宽内存

HBM3E

业界最快的, highest-capacity high-bandwidth memory (HBM) to advance generative AI innovation

暂停

样品现在可用于微米 HBM3E 12高36GB立方体

Today’s generative AI models require an ever-growing amount of data as they scale to deliver better results and address new opportunities. 微米’s 1-beta memory technology leadership and packaging advancements ensure the most efficient data flow in and out of the GPU. 微米’s 8-high and 12-high HBM3E memory further fuel AI innovation at 30% lower power consumption than competition. 

HBM3E 12高36GB立方体

微米 commences volume production of industry-leading HBM3E to accelerate growth of AI

微米's HBM3E helps reduce data center operating costs by consuming about 30% less power than competing offerings. The 8-high 24GB solution will be part of NVIDIA H200 Tensor Core GPUs which will begin shipping in the second calendar quarter of 2024.

美光的HBM3E,现已量产. 推动人工智能革命.

加快人工智能创新速度

生成的人工智能

生成式人工智能为新形式的创造力和表达打开了一个世界, 就像上面的图片, 采用大语言模型(LLM)进行训练和推理. Utilization of compute and memory resources make the difference in time to deploy and response time. 微米 HBM3E provides higher memory capacity that improves performance and reduces CPU offload for faster training and more responsive queries when inferencing LLMs such as ChatGPT.™

由晶体制成的宇航员的3D插图

深度学习

AI unlocks new possibilities for businesses, IT, engineering, science, medicine and more. 随着更大的人工智能模型被部署来加速深度学习, 维护计算和内存效率对于解决性能问题非常重要, 确保所有人受益的成本和权力. 微米 HBM3E improves memory performance while focusing on energy efficiency that increases performance per watt resulting in lower time to train LLMs such as GPT-4 and beyond.

Light blue, pink and yellow data streams combining to form a full variation of datasets on the right

高性能计算

科学家们, 研究人员, 工程师们面临的挑战是发现气候建模的解决方案, 治疗癌症和可再生和可持续能源. 高性能计算 (HPC) propels time to discovery by executing very complex algorithms and advanced simulations that use large datasets. 微米 HBM3E provides higher memory capacity and improves performance by reducing the need to distribute data across multiple nodes, 加快创新步伐.

An artist in a dimly lit room using a stylus to create a colorful abstract design on a tablet

HBM3E专为人工智能和超级计算打造,工艺技术行业领先

微米 extends industry-leading performance across our data center product portfolio with HBM3E. 提供更快的数据速率, 改善热响应, and 50% higher monolithic die density within same package footprint as previous generation.

美光的HBM3 Gen3

HBM3E专为人工智能和超级计算打造,工艺技术行业领先

HBM3E为AI计算核心提供内存带宽

拥有先进的CMOS创新和行业领先的1β工艺技术. 微米 HBM3E提供更高的内存带宽,超过1.2 TB /秒.1

戴着人工智能护目镜的女性模糊图像

HBM3E开启了生成式AI的世界

内存容量增加50%2 per 8-high, 24GB cube, HBM3E enables training at higher precision and accuracy.

霓虹背景上的水花

HBM3E为AI和HPC工作负载提供了更高的每瓦性能

微米 designed an energy-efficient data path that reduces thermal impedance and enables greater than 2.性能/功率比上一代提高5倍.

生成的人工智能 illustration of a modern high technology server room in purple neon colors

HBM3E开创了多模态、数万亿参数人工智能模型的训练

增加内存带宽,提高系统级性能, HBM3E减少了30%以上的训练时间4 and allows >50% more queries per day.5,6

聊天机器人与AI或人工智能技术聊天. Woman using a laptop computer chatting with an intelligent artificial intelligence asks for the answers he wants

微米 HBM3E: The foundation for unlocking unprecedented compute possibilities

微米HBM3E是最快的, 最高容量的高带宽内存,以推进人工智能创新- 8高, 24GB的多维数据集,提供超过1.2tb /s带宽和卓越的电源效率. 美光是您在内存和存储创新方面值得信赖的合作伙伴. 

Cropped shot of African American data engineer holding laptop while working with supercomputer in server room lit by blue light, 本空间

常见问题

微米’s HBM3E 8-high 24GB and HBM3E 12-high 36GB deliver industry-leading performance with bandwidth greater than 1.2tb /s,功耗比市场上任何其他竞争对手低30%. 

微米 HBM3E 8-high 24GB will ship in NVIDIA H200 Tensor Core GPUs starting in the second calendar quarter 2024. 微米 HBM3E 12-high 36GB样品现已上市.

微米’s HBM3E 8-high and 12-high modules deliver an industry-leading pin speed of greater than 9.2Gbps and can support backward-compatible data rates of HBM2 first-generation devices.

微米’s HBM3E 8-high and 12-high solutions deliver an industry-leading bandwidth of more than 1.每个位置2tb /s. HBM3E有1024个IO引脚,引脚速度大于9.2Gbps实现的速率大于1.2 tb /秒.

美光业界领先的hbm3e8 -high每个位置提供24GB的容量. The recently announced 微米 HBM3E 12-high cube will deliver a jaw-dropping 36GB of capacity per placement. 

微米’s HBM3E 8-high and 12-high solutions deliver an industry-leading bandwidth of greater than 1.每个位置2tb /s. HBM3E有1024个IO引脚,引脚速度大于9.实现大于1的速率.2 tb /秒.​

HBM2提供8个独立的通道,运行在3点钟.6Gbps per pin and delivering up to 410GB/s bandwidth and comes in 4GB, 8GB and 16GB capacity. HBM3E提供16个独立通道和32个伪通道. 美光的HBM3E引脚速度大于9.2Gbps,业界领先的带宽超过1gbps.每个位置2tb /s. 微米’s HBM3E offers 24GB capacity using an 8-high stack and 36GB capacity using a 12-high stack. 美光的HBM3E的功耗比竞争对手低30%. 

请参阅我们的 沙巴体育结算平台简短.

特色资源

1.  Data rate testing estimates based on shmoo plot of pin speed performed in manufacturing test environment.
2.  相同堆叠高度的容量增加50%.
3.  Power and performance estimates based on simulation results of workload uses cases.
4.  基于微米内部模型的参考 ACM出版,与当前的运输平台(H100)相比.
5.  基于美光内部模型,参考伯恩斯坦的研究报告, 英伟达(NVIDIA):用自下而上的方法来评估ChatGPT的机会, 2月27日, 2023,与当前的运输平台(H100)相比.
6.  Based on system measurements using commercially available H100 platform and linear extrapolation.

沙巴体育结算平台? Contact our support teams as well as get contact information for our individual locations.

\r\n"}}' id="teaser-1ca09e63de">

客户支持

需要沙巴体育结算平台? Contact our support teams as well as get contact information for our individual locations.

搜索,过滤和下载数据表

Get in-depth information about product features, specifications, functionality, and more.

订购一个微米样品

您的在线资源放置和跟踪订单的美光存储器样品.

下载 & 技术文档

Resources to help you design, test, build and optimize your innovative designs.