Choose how you want to continue.
Benchmarking Qwen 3.6 35B MoE (3B active) on an RTX 3090
Summary
This blog post provides a detailed benchmarking analysis of the LLM Qwen 3.6 35B MoE model using an RTX 3090 GPU, discussing performance implications of quantization, context lengths, and the efficiency of various configurations. The author shares personal insights and experiences in optimizing the model for both Vulkan and CUDA, highlighting the intricacies of using Mixture of Experts and memory management strategies.
MORE POSTS LIKE THIS
Accelerating LLMs on Debian 13: Setting up Vulkan for llama.cpp
özkan pakdil
·
Mar 22, 2026
llama.cpp vs. vLLM: Choosing the right local LLM inference engine
Red Hat
·
Jun 15, 2026
AUTHOR