Chinese contentFractile AI 把 DRAM 直接堆叠在处理器上提升内存带宽
这家叫 Fractile AI 的公司把内存直接焊在芯片上，专治大模型推理时带宽不够的老问题，还拉了头部实验室一起搞。做推理部署或关心算力成本的朋友可以看看这条路线。
#memory bandwidth
6 stories · 1 in 7 days · 1 in 30 days
Stories, products and papers tagged "memory bandwidth", newest first.
Nov 12
Oct 2
Volantis raises $88 million for optical chip memory
This funding enables Volantis to potentially solve critical memory bottlenecks in large AI models, which currently limit performance and increase costs. For developers and companies deploying massive AI models, Volantis's technology could reduce task completion times from hours to minutes, though availability in ASEAN markets remains unclear compared to existing solutions.
Aug 24
Xiaomi Xuanjie O3 LPDDR6 memory chip
Xuanjie O3's LPDDR6 support offers significant memory bandwidth improvements for AI applications in Xiaomi devices across Southeast Asia.
IT之家Source
Jul 24
Industry16:17
Chinese content消息称 SK 海力士在美招募 DRAM - 逻辑 3D 堆叠领域技术人才SK海力士在美招人搞3D堆叠内存，专为设备端AI提升带宽和能效，跟标准LPDDR拉开差距。
IT之家Source
Jul 10
Chinese contentMRDIMM将DDR5服务器内存速率冲向17600 MT/s，带宽逼近DDR6
JEDEC的MRDIMM让DDR5内存速率冲到17600 MT/s，带宽直逼DDR6，老平台就能升级，数据中心省下换插槽的钱。
IT之家Source
Jul 4
Chinese content澜起科技第二子代MRDIMM传输速率达12800MT/s，处于规模试用阶段
澜起科技的第二代MRDIMM内存速度达到12800MT/s，比上一代快45%，是普通RDIMM的两倍，专门解决AI推理时的带宽瓶颈，适合关注服务器硬件的朋友。
IT之家Source