Nvidia says it can shrink LLM memory 20x without changing model weights

Nvidia's KV Cache Transform Coding (KVTC) compresses LLM key-value cache by 20x without model changes, cutting GPU memory costs and time-to-first-token by up to 8x for multi-turn AI applications.

New KV cache compaction technique cuts LLM memory 50x without accuracy loss

MIT researchers developed Attention Matching, a KV cache compaction technique that compresses LLM memory by 50x in seconds — ...
Nature

Cache Performance and Memory Hierarchy Optimization

The dynamic interplay between processor speed and memory access times has rendered cache performance a critical determinant of computing efficiency. As modern systems increasingly rely on hierarchical ...
TechSpot

AMD 3D V-Cache CPU memory used to create incredibly fast RAM disk

Why it matters: A RAM drive is traditionally conceived as a block of volatile memory "formatted" to be used as a secondary storage disk drive. RAM disks are extremely fast compared to HDDs or even ...
NextBigFuture

Tesla Not Compute Limited for FSD AI Means 100+ Exaflops and 1+ Exabyte of Cache Memory

Tesla indicated in August, 2023 they were activating 10,000 Nvidia H100 cluster and over 200 Petabytes of hot cache (NVMe) storage. This memory is used to train the FSD AI on the massive amount of ...