Anyone sitting on a lot of slow system memory and a modest GPU.. try Strata + Qwen3.8 Next.

A user with slow DDR4 memory and a modest 7900XTX GPU found success running Strata + Qwen3.8 Next. They reported speeds of 45-70 tokens/s using IQ3_XXS weights, which are just under 80GB. This performance is comparable to users with 12GB and 16GB cards, while DDR5 owners experience significantly faster results. The user recommends trying this setup, especially if 27B models are not performing well.

This report offers a specific configuration for running large models on modest hardware, unlike general discussions that often assume high-end systems.