Samsung Pools KV Cache Using CXL
lauriewired · x · 2026-07-12
Samsung has published a paper/proposal detailing the offloading of KV-cache to a CXL memory pool.
- The experiment utilized an early CXL 2.0 ecosystem, older switches, and a relatively basic setup.
- By applying simple interleaving of the KV Cache across multiple CXL modules, the GPU supply remained stable, performing almost "like real DRAM".
- While the author notes this isn't SOTA, the implementation is highly reproducible, demonstrating that CXL already offers practical value for scaling inference memory.
More from Infra
- Analyst: Ant has the most fungible AI workload of any model provider today — BenBajarin · 2026-10-26
- Anthropic S-1: training infra edge cut per-task cost 40% and doubled revenue growth — BenBajarin · 2026-10-26
- Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26
- US states halt new data centers as 72% of Americans oppose local builds — YvesMulkers · 2026-10-26
- $5M Seed Round Can't Afford One DeepSeek Training Run: GPU-less AI Startup's Three Paths — FinanceYF5 · 2026-10-26
- HBM demand shifts: low-stack inference plans shelved, Rubin to go 12-high only — BenBajarin · 2026-10-26