Two Arc Pro B60s got XeMaya past a one-card coding timeout
Flash-Next loaded on one B60. With two, it took nearly eight minutes longer to start and passed all 24 coding, extraction and tool-decision attempts.
Build logs, practical guides and measured comparisons.
Flash-Next loaded on one B60. With two, it took nearly eight minutes longer to start and passed all 24 coding, extraction and tool-decision attempts.
The model took 13 minutes to load from our hard-drive archive. Once warm, the median passing reply took 2.70 seconds.
On two Arc Pro B60s, eight warm Strata requests recorded 16.0 MB of reads and 0.70 MB of writes on the model SSD. Here’s how to measure your own setup.
Four streams moved the queue faster, with a longer wait for replies to start. Here’s how to test the tradeoff on your own server.
Read the startup token budget, leave room for the answer, and distinguish a model that loads from a workload that runs reliably.
Liquid AI’s small agent-focused model is a timely option to investigate. Task reliability and model-specific backend support still need their own evidence.
Reuse a document or conversation prefix to save prompt processing. Here is how to tell whether that matches your workload—and how to compare it fairly.
Chat, coding agents and batch jobs need different measurements. Here's a checklist for comparing local inference setups.
On our four-task comparison, the median fell from 260 seconds to 162. The faster mode also held roughly 11 GiB of host RAM per card.
Direct GPU transfers and a memory-leak fix turned a slow two-card setup into a working 27B server.