Qwen Next on 24 + 64 GB VRAM?

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?

添加评论
点赞收藏
点踩分享查看原文
评论
?
参与讨论