Back to articles
DevPickView feed
PC Watchby Ubuntu Japanese Team あわしろいくや
Original published:

Ubuntu Daily, Episode 93: 3x Speed on Arc B580? Surprised by Strata-derived 'XeStrata'

Ubuntu Daily, Episode 93: 3x Speed on Arc B580? Surprised by Strata-derived 'XeStrata'
AI summary
Editorial policy

Amid rising prices for AI-related PC parts, the inference engine 'Strata', which runs large LLMs on 12GB video cards, is gaining attention.

Key points

  • •Prices for PC parts, especially memory and video cards for AI, are rising significantly.
  • •The inference engine 'Strata', which runs large LLMs on 12GB video cards, is attracting attention.
  • •The Strata-derived 'XeStrata' is featured for delivering 3x speed on the Arc B580.
Read the original(opens in a new tab)
#Semiconductors & Infrastructure#LLM / Model Trends#Strata#XeStrata

Related articles

See more
PC Watch

Will DeepSeek-V4.1-Flash-1017 run on Strata? Optimized model coming soon

A 763B-parameter AI model may run on gaming PCs through compression via GSQ-RCO and optimization for local inference using the Strata engine.

PC Watch

Qwen3.8-Flash-Next runs on 32GB! Strata 0.2.0 beta

Software named Strata 0.2.0-beta, which allows the local LLM Qwen3.8-Flash-Next to run on a total of 32GB of memory, was released for free on October 19.

PC Watch

Kazuhisa Nishikawa's Irregular Column: The 9-year-old 'Tesla V100' is hot. Qwen3.8-Flash-Next achieves about 80 tok/s!

The Strata engine allows the Qwen3.8-Flash-Next MoE model to run at approximately 80 tokens per second on a 9-year-old Tesla V100 GPU by utilizing a three-tier memory hierarchy.

PC Watch

Run Qwen3.8-Flash-Next on a Mini PC! Strata Finally Supports Ryzen AI Max+

Local LLM inference engine Strata now supports Ryzen AI Max/Max+ series, enabling Qwen3.8-Flash-Next operation with 128GB of memory.

PC Watch

[Kazuhisa Nishikawa's Irregular Column] I Tried 'Strata,' Which Can Run 'Qwen3.8-Flash-Next' on a Slightly Older Gaming PC!

Report on trial results of the tool "Strata," which runs the 125B parameter "Qwen3.8-Flash-Next" model on a 12GB VRAM video card.