Suggestion for frameworks’ next device:
A thunderbolt/USB-4/5 dock with an upgradable mobile eGPU. Then people could have thin and light laptops for on the go, a good USB-C dock at home for external monitors/peripherals etc, but have an external mobile GPU (RTX 4000’s or 5000’s or equivalent AMD) in it for a gaming/LLM/3D Modeling powerhouse setup. Most importantly, the mGPU should be upgradable, which is why I think framework is one of the only companies that might make this happen!
I like the idea, though I’ll probably not use one.
I couldn’t afford to buy new dGPU since 2017-2018. Luckily I could afford the money, I just don’t have the use case, so for me it would be a waste.
By upgradeable do you mean you can exchange it with different GPUs? I’m a little confused, how would this differ from current eGPU enclosures? Or would it just be framework’s take on it? I’d buy it I suppose, but not sure I quite understand the idea.
This would be really good if it could use the GPU modules from the FW16 laptop…
Ooooh, is that what he meant? Yeah ok I can see that.
I guess the idea to add OcuLink adapter to the FW 16 (via expansion shell) was not picked up by Framework ?
Actually, now I did as well 
I’m not sure of the technical details, but I remember that OcuLink was faster than USB4.
Would that matter for gaming? Or professional software?
Would that matter for gaming? Or professional software?
Yes to both, but given no current Framework devices feature Occulink but they do feature USB4, I’d figure it would be a sensible feature to start with.
If I remember correctly, USB4 is about 2/3rds the bandwidth of Occulink… but Occulink is beset by other problems (quality and length of Occulink cables affecting performance and reliability) as well as the lack of hot plug capability… so despite the losses, USB4 might be a better option anyway.
Oculink should be 64 GB/s, if I remember this correctly, so faster than USB4 (40 GB/s ?) .
I am not so sure myself though on how much that matters. For gaming, the question would be how much it matters vs. 5070 over expansion bay. 64 GB/s vs. 384 GB/s is significant bottleneck, so I guess, the difference is notable for, lets say, to casual gamers. And to have eGPU with OCulink vs. adding dGPU - would it be pricewise also similar (?). The cost saving is the factor, where it could make sense, if it is significantly in favor for eGPU (i.e. eGPU with cheaper AMD graphics or second hand NVIDIA).
I am not a gamer though.
384GB/s is the memory bandwidth on the GPU and not what the card itself runs at… if it’s running in a PCIE 4.0 slot, even at x16 it’s limited to 64GB/s between the system and the GPU itself… a few modern GPUs don’t even bother with x16 and run at x8 … so don’t need all that bandwidth to begin with.
I think I remember the rough consensus from YouTubers seemed to be: don’t bother with anything more than a x070 class card in an EGPU as you get rapidly diminishing returns.
But I could be wrong.
Yeah, that is another topic. And quite important for Framework perspective I guess, as it would increase the count of support tickets. Also, thunderbolt 5 vs. Oculink is closer for being comparable, definitely bandwidth wise. So it would maybe be best to wait until that can be offered.
There’s a lot of mixed messaging in this thread as well.
The OP asked for a ‘USB-C dock’ … and, thanks to the buffoons in the Tech industry, this can means all manner of things.
People assume that a USB-C port is a USB-C port and they’re all made alike… but this is far from true.
Similarly, Thunderbolt 4 and USB 4 are not the same thing (although often have very similar features / capabilities) … and AMD devices do not have Thunderbolt (thanks Intel)… but they often do have USB 4.0.
See here for a decent overview and apologies if you are already aware of this and I’m covering old ground: Thunderbolt 4 vs USB4: What's the difference? | Tom's Guide
So, I’d expect it to be some time before Thunderbolt 5 makes it to a lot of PC laptops.
Thats OK. My head is all over the place today, I guess.
Yes, i meant a mobile eGPU. Agree it would be fantastic to use the same board/GPU setup from FW16 for decreased cost to produce and more ways to recycle boards!
Yeah, USB 5.0/Thunderbolt. I agree Oculink is faster, but very niche, so I would just be happy to increase gaming performance on any laptop using a external eGPU.
Also, include an NVMe slot inside the dock, then it can be a game library/backup storage solution in addition do a mGPU dock.
the big dawg talked about making an egpu housing for fw16 graphics modules in this vid fyi:
it would be so nice to have since i am not a fan of the fw16, but to have dedicated graphics w my fw13 would be peak af
dawgs you can just use real desktop GPUs and connect it over USB4
i do it with an RTX 4090
the only reason to use a mobile GPU in your eGPU is so that you don’t landfill it later
they’re otherwise worse in every metric including price
Except power efficiency and size but otherwise yeah, using a mobile gpu in an egpu has quite a niche application.
As much as I like things to be module, I’m not really sure what value an external enclosure for Frameworks GPU module would really be. Yeah it would be something you could put your old/unused modules in so that they can be used outside of the laptop, but that’s really the only value you’d get out of it; it can’t be used independently like a mainboard so it’s just another external thing to hookup.
Personally if I were looking for a small portable eGPU, I’d pickup something like the GPD G1, the GMKtec AD-GP1, the ONEXGPU, or similar. They all have a 7600M XT gpu which is similar in performance to the 7700S. Their footprints are much smaller than you’d be able to make an enclosure for the Framework GPU, and the GPD G1 even has the power supply built-in. When it comes time to upgrade, well just buy another one, as all it is is the GPU anyway.
That’s not to say that an enclosure for expansion bay modules wouldn’t be useful, just that the value for eGPUs isn’t really there. I would like one though just to play with.
Mini PCs are becoming more popular, so external enclosures for a framework GPU module could be a viable second life for them.
These Mini PCs could for example become a NAS running Plex or Jellyfin, which can sometimes require realtime transcoding depending on the client device (almost all smart TVs, especially if you use “non-standard” containers and encodings like MKV and AV1).
Since these Mini PCs almost never have a PCIe slot, but sometimes have USB4/Thunderbolt/oculink, you can make them useful by plugging in a GPU. No matter how rudimentary it is, it should be viable for realtime transcoding.