Testing Qwen 3.6 35B MoE with 3B Active Users on RTX 3090
benchmarks gpu llama mistral qwen
| Source: HN | Original article
Qwen 3.6 model benchmarked on RTX 3090. It has 35B parameters and 3B active.
Benchmarking has been conducted on Qwen 3.6 35B MoE, a Mixture of Experts model, using an RTX 3090. This model boasts 35B total parameters, with 3B active, and utilizes the BF16 tensor type, requiring approximately 70 GiB of space for the weights alone.
This development matters as it signifies a notable step in optimizing AI models for local operation on consumer-grade hardware. The Qwen 3.6-35B-A3B model is particularly noteworthy for its ability to deliver high-quality performance with reduced active parameters, making it a viable option for coding and speed-sensitive tasks on devices like the RTX 3090.
As we look to the future, it will be interesting to observe how this model performs in real-world applications and whether its efficiency can be further improved. With its potential to provide powerful local AI assistance without incurring cloud expenses or requiring code to be sent to external servers, the Qwen 3.6 35B MoE is certainly a model to watch.
Sources
Back to AIPULSEN