# Making best use of varying GPU generations

**URL:** <https://discuss.vllm.ai/t/making-best-use-of-varying-gpu-generations/396>\
**Category:** NVIDIA GPU Support\
**Created:** [April 7, 2025, 12:46pm UTC](https://discuss.vllm.ai/t/making-best-use-of-varying-gpu-generations/396 "2025-04-07T12:46:51Z")\
**Posts on this page:** 1\
**Showing post:** 4

<div class="post-metadata">

**Author:** ![Phillip-Merritt](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/phillip-merritt/32/1540_2.png) [@Phillip-Merritt](https://discuss.vllm.ai/u/Phillip-Merritt)\
**Post date:** [July 4, 2026, 6:42pm UTC](https://discuss.vllm.ai/t/making-best-use-of-varying-gpu-generations/396/4 "2026-07-04T18:42:59Z")

</div>

Hey Chris,

I’ve been wondering the same thing. I found [this post](https://discuss.vllm.ai/t/run-vllm-on-two-diffrent-gpu/659/2) from May saying that the larger GPUs would be limited to the memory of the smallest GPU.

However, I’m wondering if systems like [DFlash](https://github.com/vllm-project/vllm/pull/44586)could support differing GPU sizes. Such as, putting the larger expert model on your biggest/best GPU and putting the smaller speculative models on your smaller GPU.

---

_[View the full topic](https://discuss.vllm.ai/t/making-best-use-of-varying-gpu-generations/396)._