GPU Expansion Card USB-C Port Specs

Hey there everyone! I noticed that the new GPU Expansion Card has a rear-facing USB-C port, which is presumably intended for DP alt mode, but I was wondering if there are any official specs for it, what else it can do, or what limitations it has. All I know is I don’t think it can charge the laptop due to the power limitation of the FXBeam interposer.

Thanks in advance!

5 Likes

That’s a great question but I don’t think Framework has shared that info anywhere, we’re just going to have to wait. I’m guessing at latest they’ll cover it in their expansion bay deep dive.

It’s a native USB-C display out.

3 Likes

What ports does that thing have, anyway?

*asks the guy who already pre-ordered it

3 Likes

The specs for the 7700S list a USBC port with DP Alt Mode. I was wondering if anyone knew anything about the necessity of this port? Will it be required for high-res freesync displays or can the mainboard USB4 handle all of that (and if the mainboard can, what advantage does the GPU USB port have when the module is connected to the laptop)? Trying to plan my setup for when the FW16 arrives; I want to have a 4k freesync primary monitor and a lower-res secondary monitor, and it would be great if all of that could be passed through a USB4 dock via a single USB-C port without losing anything.

1 Like

Somebody already asked at GPU Expansion Card USB-C Port Specs, but not much is known. It is directly connected to GPU, so there can be benefit in using that port to connect hi-res monitor.

TLDR at bottom.

The mainboard USB4 should be able to handle all of that.

The big difference is that the port on the graphics module is connected to the dGPU (the dedicated GPU in the graphics module) whereas the ports on the mainboard are connected to the iGPU (the GPU integrated into the CPU).

When using the graphics module and using a port on the mainboard the dGPU will still do most of the work, however the iGPU will handle the final outputting to the display.

The iGPU does support outputting at up to 8k 60Hz or 4k 240Hz and up to 4 displays and freesync, so the mainboard ports should be plenty capable of driving displays. However the additional step of having the iGPU do the outputting does result in a slight performance penalty, which is why the graphics module does have a port.

The USB-C ports on the mainboard should be fully capable of running all that you want, however with an ~10% performance penalty due to the iGPU handling the final outputting to the display.

The USB-C port in the graphics module on the other hand should get full display performance, however is severely limited in other capabilities. The port has 480 Mbps of other bandwidth (for purposes other than displays) and it has not been announced if it will support charging through that port (I doubt it will).

Although that last part (charging support) may not matter much because afaik there are not yet any docks that support the new charging standard that Framework is using, so for now using it in a dock would limit the charging speed (and you could even lose charge while using it and docked). So I personally would suggest using a separate cable for charging anyway.

So what you could do is use the graphics module’s port for displays and peripherals (the limited bandwidth for non-display purposes is still plenty for basic peripherals and even is double the speed of the average internet connection so it can be useful for ethernet as well) and use a port on the laptop for charging.

Although beware that many docks attempt to reallocate half of the bandwidth away from displays and too USB instead, however since the graphics module does not support that the bandwidth is just wasted (which can result in displays being limited to lower settings due to inadequate bandwidth). So if you do want to use the port on the graphics module I suggest this dual display or this single display dock from Cable Matters as those allow full bandwidth to the display.

TLDR:

Mainboard ports have an ~10% reduction in graphics module performance.

Graphics module ports have reduced bandwidth for purposes other than displays and it is unknown if charging will be supported (I doubt it).

So to get full capabilities a two cable solution will be needed, although even if there weren’t those limitations it would likely still be beneficial anyway due to the lack of docking stations currently available with support for the full charging power of the Framework 16.

7 Likes

That is wonderfully thorough, thank you! I think I’ll drive the 4k freesync with the GPU port and everything else through a dock. One more cable for that kind of performance gain is trivial.

Agreed.

I am a little bit disappointed that the mainboard ports aren’t on a mux like the internal display (a mux would allow the ports to be switched between the iGPU and dGPU as needed), however I understand that it is likely not worth the effort for two big reasons.

The modularity of the Framework 16 likely makes implementing a mux much more difficult. Having more things connected to a mux would require more connections between the mainboard and expansion bay. Furthermore the modularity of the expansion cards means more potential ports that customers could want mux output through. Both of those add additional complexity, points of failure, and cost just to reduce from needing two cables to get full performance to only needing one.

Another roadblock for adding a mux is the CPU. Since the CPU contains the USB4 controller (so the capabilities of the USB4 ports are largely dictated by the CPU) and IIRC AMD doesn’t support a mux on USB4 and Intel dropped support after 11th gen, which adds significant complications to fully supporting a MUX on the USB4 ports, partial support (ie. Mux only in some modes of display over USB4 as there are multiple) may be possible, although IDK if any laptops do support it.

2 Likes

You could only Mux raw DP Alt mode but not anything that is actually USB4.

I know of no device that actually does that (as it would be really confusing) and most likely departs greatly from any sample configs from AMD and Intel.
Some notebooks simply ignore the integrated USB4 controllers and have just added external TB controllers, like the XPS 17 line with Maple Ridge controllers for each side (precisely so they are muxeable).

But those provide a worse implementation than modern integrated controllers, because they have much more limited PCIe connectivity, will use more power and cannot use Windows’ USB4 drivers which have more control over what is going on. And so far we have seen no AMD notebook with Intel TB controllers. Even the desktops are not certified from Intel and often warn of problematic hot plugging etc.

Yeah, IMO mux on USB4/Thunderbolt was effectively dead (ie. Having downsides that make it not really worth doing) the moment that Intel stopped having DP inputs on their CPUs for the internal TB4 controllers to use for mux purposes (I can’t find any documentation confirming it, but I’ve seen a few comments indicating that 11th gen H-series was the last to have that).

Do you have any test data for the performance difference numbers? I did a quick test comparing port #4 to the graphics module port and I found a 10% improvement to FPS in Starfield, going from ~34 FPS on the rear USB to ~38 FPS on port #4.

Incidentally, I did see significantly better performance when using the integrated display (~50 FPS) than when using an external display.

This wasn’t a comprehensive test, but I wanted to check if some of this info is still true or if others are seeing similar performance differences.

Ubuntu 22.04 LTS with Wayland
Starfield run with DRI_PRIME=1 option in Steam