Home

Upload

After a week of trial, I decided to replace Qwen3.6-35B-A3B with Qwen3.8-Flash-Next to run OpenClaw

18.6K

13

2026-09-07 20:16:22

Installation documentation: https://pan.quark.cn/s/332010914ff9

Qwen3.6-35B-A3B

Qwen3.8-Flash-Next

AMD AI max+ 395

Qwen3.8FlashNext MTP

titan909 Send message

Actual test: A card ROCm7.2 vs ROCm7.1.1 running gpt-oss-120b performance

03:15

Actual test: Impact of AMD GPU ROCm7.2 vs ROCm7.1.1 on ComfyUI's image/video generation performance and stability

04:09

Settings Max+395 to break the 96G VRAM limit and run medical large models with VLLM

07:19

MS-S1MAX Desktop AI Supercomputer Unboxing: Installing Ubuntu, AMD GPU Drivers, Network Card Drivers, and Docker AI Environment

10:36

Actual test: AMD Max+ 395 vs RTX4060Ti AI performance comparison (model training)

07:31

Using AMD Max+395 to do things that 4090 can't do - training large models

06:28

Unboxing, Inspection, and HDD Installation of AI Desktop Supercomputer Mini PC - Beelink GTR9 Pro

04:37

[Debunking Myths] Testing the Impact of Different VRAM Allocation Settings on AI Performance

02:10

Max+395 ROCm7.1.1 compiled VLLM0.12.0 running gpt-oss-120B LLM

07:31

Tested: Max+ 395 ROCm7.1.1 llama.cpp running gpt-oss-120B and Qwen3-Next-80B speed

05:27

Sharing a method for stable A-card operation of ComfyUI Z-Image-Turbo text-to-image generation [Solving A-card ComfyUI abnormal exit issues]

05:20

Max+395 Settings with 120G VRAM Running MiniMax-M2.1 to Generate PC Games

07:31

AI Max+ 395 breaks 96GB VRAM limit to deploy GLM-4.7 for PC game generation

06:44

Qwen3-Next: Simple two-step operation boosts performance by up to 20%

04:51

AI-powered full-process medical report analysis, say goodbye to search engines

08:57

🔥0 Online! Watch how my local AI fully automates phone operations to handle daily tasks

01:36

Tutorial and Demonstration of Running Large Language Model Inference Using NPU

04:36

Testing whether the newly open-sourced Qwen3.6-35B-A3B is as powerful as reported, using OpenClaw

04:22

Solving the issue of Ollama using CPU instead of GPU for computation

03:36

Qwen3.6-27B model + Hermes Agent local deployment, AMD Max+ 395 real-world test: DFlash vs MTP acceleration effects

04:30

Using laptop NPU to run Qwen3.6-35B-A3B for Q&A, coding, and image recognition

06:53

Tested locally deploying Tencent's open-source Hy3 model to develop a 3D FPS shooting game

03:44

Actual test of Max+395 dual-machine running DeepSeek v4 Flash 0731 official version encoding performance and quality

03:32

Tested the runtime performance of Muse-Glimmer-30B combined with the Hermes agent

03:02

Solving the crash issue of DeepSeek-V4-Flash-0731 in cluster deployment, doubling performance

09:21

After a week of trial, I decided to replace Qwen3.6-35B-A3B with Qwen3.8-Flash-Next to run OpenClaw

02:55

AMD cards ROCm running ComfyUI idle CPU usage issue

01:43

【Share】Qwen3.8-Flash-Next's prefill improvement triples and reduces agent execution time by 75%

03:11

Open source general-purpose inference engine installation and deployment tutorial - gufo

03:11

GPU+NPU boosts Qwen3.8-Flash-Next's pp performance to 2100+ tokens/s

03:13

Top