Qwen Next on 24 + 64 GB VRAM?
一位用户正在咨询如何在配备 24GB 3090 显卡和 64GB CMP 170HX 显卡以及 64GB DDR4 内存的系统上运行 Qwen 3.8 Next。他们正在寻求关于如何配置以利用两块 GPU 并可能将剩余部分卸载到 DRAM 的建议，以期达到可接受的速度。用户指出 CMP 170HX 存在 PCIe 2.0 @ 4x 的瓶颈，但已成功在其上以 BF16 运行 Qwen 3.7 27B 或 Minimax H3，并承认 Qwen Next 的要求更高。
- 发布
- 2026年9月7日 11:10
- 来源类型
- 开发者社区
- 档位
- 社区
- 信源状态
- 正常
时间以 UTC 显示
更多信息
Hey all,
I now have a 3090 24GB + a CMP 170HX unlocked with 64GB. I also have 64 GB DDR4.
Can I run Qwen 3.8 Next in any possible configuration with both GPUs and perhaps offloading remaining stuff to DRAM? Is there a way I can get acceptable speeds?
The only bottleneck with the CMP 170HX is the PCIe speed (2.0 @ 4x), so that needs to be factored in. It’s usually not much of an issue though IME once model is loaded…
I am successfully running either Qwen 3.7 27B or Minimax H3 at BF16 on the CMP 170HX at decent speed, so I know the GPU is good for it. But Qwen Next is another beast.
Ideas?