Developer Shares Exllamav3 MoE Model Offloading and Optimization on 9070XT
11/17/2026 — 11/17, 20:19·1 sources (all)·1 reports (all)
Key takeaways
- A developer shared a method to run the Qwen 3.6 35B MoE model on an AMD 9070XT GPU using Exllamav3. By increasing the chunk size, they achie…
Background & analysis
On November 17, 2026 at 19:10, a developer posted in the Reddit community LocalLLaMA to share experience running a Qwen 3.6 35B MoE model on an AMD 9070XT 16GB GPU with Exllamav3. In the setup, the experts are offloaded to DDR4 3200Mhz RAM while the model runs at Q4 quantization. The post is presented as a follow-up to the developer's earlier thread on MoE and expert offloading, and it notes that the community is still waiting for Qwen 4.
According to the developer, continued tinkering with the expert-offloading configuration delivered substantial gains over the previous post. Decode speed stayed roughly in line with the earlier attempt, but raising the chunk size to 5120 pushed prefill past 2000 t/s at every context length, with the peak reaching 2606 t/s. The write-up therefore offers a concrete, reproducible tuning configuration for running large models locally. Only this single report is available in the current material, so the event currently rests at the stage where the developer has published the configuration and measured speeds, with no further developments documented yet.
AI-generated from 1 reports · updated 10 hours ago
- Heat index
- 72
- All sources
- 1
- All reports
- 1
- First seen
- 12 hours ago
Reports on this story headlines open the original
A developer on Reddit shared offloading configurations for running the Qwen 3.6 35B MoE model using Exllamav3 on an AMD 9070XT 16GB GPU. By increasing the chunk size to 5120, prefill performance exceeded 2000 tokens per second across all context lengths, peaking at 2606 t/s.
Reddit · LocalLLaMAAI score 54
Other stories people are talking about
- Heat index 1331Nvidia board establishes national security and export control committee25 sources
- Heat index 1027California Sues OpenAI Over Agent Intrusion Into Hugging Face19 sources
- Heat index 1006NewGPT-6.2 Astra and Sol Now Available on ChatGPT, API and Perplexity11 sources
- Heat index 934SurgeJensen Huang says he would return as Nvidia CEO and agrees to testify on Capitol Hill11 sources
- Heat index 852Anthropic guides IPO pricing at top of range or higher, with trading set for November 1911 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

