We love the rig porn Offline AI speed tests with $5k, $10k, $15k + hardware as much as anyone.
But what if you haven't got that kind of hardware? Here's a few years old NVIDIA 4060 pushing 130 tokens a second using the smaller EmperoAI 2B version of Qwen 3.8. Straight out of
Less than 10 hrs to go. I will take a 2hr nap.
Day 0 support covered by many. Loading NVFP4 instantly. Personal Eval Suite ready to go. Local AI community/Support growing every week. Sidenote: Qwen 3.8-27b was already moving great on the DGX Spark. Imagine now with the MOE
Qwen 3.8 27B
Prompt: bijanbowen (YouTube Channel: Get this guy to 100k, his videos are the best. Deepseek Harness Time to complete: 55 minutes Average tokens per second: 222 RTX5090
Qwen 3.8 27B
Deepseek Harness Time to complete: ~15 minutes Average tokens per second: 252 RTX5090
You have reached the end of the archive
All of qwen38