SK hynix demonstrates CXL memory systems for AI agents at OCP 2026
SK hynix demonstrates CXL memory systems for AI agents at OCP 2026. At the OCP Global Summit in San Jose, SK hynix demonstrated AI-server systems using CXL memory pooling. In a demonstration with Arcee, Liqid and Marvell, it showed KV-cache transfer and reuse across multiple servers and GPUs; the company reported better system performance than conventional methods relying on network communication.
Event / data period: 2026-10
Published: · Updated:
At the OCP Global Summit in San Jose, SK hynix demonstrated AI-server systems using CXL memory pooling. In a demonstration with Arcee, Liqid and Marvell, it showed KV-cache transfer and reuse across multiple servers and GPUs; the company reported better system performance than conventional methods relying on network communication.
In another demonstration with AMD, Penguin, Xcena and Arcee, the system predicted needed KV caches from conversation and task history and preloaded them into DRAM for faster processing. The demonstration also covered storing the full cache on SSD while retaining fast reads. SK hynix also showed pooled memory for speeding transfers among models in a personal AI-agent service with Pickle, and a CMM-Ax demonstration for long-context LLM inference. These were company demonstrations; the release does not establish broadly deployed systems or provide quantified results for these examples.
Data
SK hynix demonstrates CXL memory systems for AI agents at OCP 2026
At the OCP Global Summit in San Jose, SK hynix demonstrated AI-server systems using CXL memory pooling. In a demonstration with Arcee, Liqid and Marvell, it showed KV-cache transfer and reuse across multiple servers and GPUs; the company reported better system performance than conventional methods relying on network communication.
- · Occurred
Company report on OCP demonstrations
At the OCP Global Summit in San Jose, SK hynix demonstrated AI-server systems using CXL memory pooling. In a demonstration with Arcee, Liqid and Marvell, it showed KV-cache transfer and reuse across multiple servers and GPUs; the company reported better system performance than conventional methods relying on network communication.
Methodology
This event record summarizes the presentation and demonstration reported in the cited primary source. Performance figures reported by the manufacturer are not presented as independent verification or industry-wide results. The event entry is dated to publication of the company report; the source does not identify the precise demonstration day, so the article event period is October 2026.
Updated:
Subjects and places
Subject proposals
- ElectricitySuggested
- Subject suggested from the title or summary: KVnews.skhynix.com ↗
- Artificial intelligenceSuggested
- Subject suggested from the title or summary: AInews.skhynix.com ↗
Geographic references
Only source-provided geographic references are shown. Opening an external map is optional. Optional story maps use OpenStreetMap. Opening a map sends your IP address and this site’s address to the map provider; location data remains readable as text without loading the map.
- United States · Place role: Coverage · Place type: Country or area · Location precision: RegionalEvidence
- Explicit preserved editorial country/region; no city or national centroid inferrednews.skhynix.com ↗