The post discusses the performance of various AI models on hardware with a 780M GPU and DDR5 RAM running at 5600MT/s. The user mentions that while they can run qwen3.5-35b-a3b, it is very slow in processing tasks despite achieving 16 transactions per second (TPS) at Quality Level 4. They previously used lfm2 24b which was faster but had issues with tool calling and an unusual fixation on quantum computing topics. The post seeks recommendations for alternative models that perform well under these hardware constraints, focusing on better processing speed and functionality without the aforementioned quirks.
- qwen3.5-35b-a3b
- lfm2 24b
- Upgrade GPU to a more powerful model if budget allows, such as NVIDIA RTX 3080 Ti or AMD Radeon RX 6900 XT.
- Ensure DDR5 RAM is running at optimal speed settings by checking BIOS/UEFI configuration.
- Consider using alternative AI models known for better performance under similar hardware constraints.
In homelab environments, this issue impacts systems where computational resources are limited. The specific software versions and configurations affected include GPU drivers (latest stable release recommended) and the RAM's timing settings in BIOS/UEFI.