In this AI era, I think there’s a real opportunity for Framework to offer more powerful GPU options in future generations.
For any future dGPU, I’d really love to see at least 16GB of VRAM. An RTX 5090- or RTX Pro–class option would be amazing, although I realize that would depend heavily on NVIDIA and the possibilities of your partnership with AMD. Even a high-end AMD GPU with significantly more VRAM would already be a huge step in the right direction.
I’m really excited about what the next generation of Framework hardware could look like, especially from an AI and GPU perspective.
I also feel that the current AI 300 offering is somewhat limited without a more powerful Strix Point/Strix Halo–class option. With Framework now offering a new high-end board for the 12", I’m wondering whether something similar could eventually come to the 16".
I’d be very interested to understand how much flexibility there is in the AMD partnership and whether higher-end GPU options are something Framework is exploring.
I really hope you’re working on this, Framework. There are definitely customers like me ready to upgrade as soon as you offer something more powerful.
already doing that but not sufficient
I need moaaaarrrrr
I’d like to be able to run, let’s say, Qwen3.8 27B or the next to be released Qwen3.8 next model 100% on the laptop at Q8 and decent speed without touching the eGPU… That would be fantastic !
I agree that better GPUs with more VRAM is needed to make the FW16 a good AI system. I’d love an upgrade that had LPCAMM2! I also wouldn’t mind USB4v2 or Thunderbolt 5 (I’m not sure if the AI 300 series is using USB4v2 or not). I know my 7840 is still using USB4v1.
If I recall correctly, framework said when this was asked before that their GPU module solution for the 16 is physically / space limited to 4 dram dies, which means 12GB is the most you can expect until >3GB GDDRx dies hit the market.
We can probably expect that approximately in 20never unless the tech giants decide they’ve finally built enough and stop siphoning all dram capacity into HBM.
I thought it limited it to 8GB and they got it to the 12GB by upgrading it to 3GB dies. I remember it being mentioned in a video somewhere. (Edit: I read it wrong and we’re saying the same thing.) Either way I think it’s only an issue if they persist on using the same GPU shell (which they’ve updated anyway). I don’t see why they couldn’t make the shell slightly thicker for a new GPU and allowing for more VRAM. I feel like most people who want a better GPU are fine with the tradeoffs of a slightly thicker expansion bay.
If you’re talking about doing a clamshell design with dram on both sides of the PCB, that may end up being quite a bit thicker since modern gddr6/7 is clocked hot enough to need active cooling, but maybe possible without breaking backwards compatibility, I’m not sure.
Yes that was my thought. I don’t know how much of a height difference it would make but I feel like if the Graphics Module Cover was taller I wouldn’t think it would be an issue. From what I can tell the VRAM is already on the “top” (relative to the laptop sitting on a table) so it should be easy to add more on the bottom. That would make the module cover thicker and they may need a way to get heat pipes or something to cool it, but it could be done.
