skinny.

Latest / Misar.Blog Podcast

Llama 4 vs Qwen 3.6 70B on RTX 4090: Speed Benchmarks

Llama 4 Scout vs Qwen 3.6 70B-class on a single RTX 4090: what fits in 24GB, tokens/sec with CPU offload, quant levels, and why Qwen 3.6 32B is often the… From the article "Llama 4 vs Qwen 3.6 70B on RTX 4090: Speed Benchmarks" by Synor, published on Misar.Blog.

In this episode:

0:00 Neither Llama 4 Scout nor Qwen 3.6 72B Fits Cleanly in an RTX 4090’s 24 GB

1:30 Llama 4 Scout Can Run Surprisingly Well on a 4090 if Its Experts

2:00 Qwen 3.6 72B Is Much Harder to Run on a 4090 Because Every One

2:40 For a Single 4090

3:17 Llama 4 Scout Is a Small MoE Model with 109B Total Parameters Across 16…

The skinny

The skinny isn't ready yet — notes appear once the transcript is processed.