Looking for XG Mobile-like, direct PCIe gen4 interface, eGPU capability for the Framework 16

Framework Laptop 16
#1

Looking at the Asus XG Mobile, I can’t help, but wish for something similar from the framework 16.

Considering that the framework 16 expansion module offers 8 lanes of PCIe gen4, I think that this is an amazing opportunity for designing a competitive eGPU enclosure standard.

I know that this isn’t for everyone, but I am looking to ditch my current 5 year old desktop PC, and use the GPU inside (intel arc A770) with my framework 16 for the next 1-2 years before buying a new desktop. Lastly, I am probably just dreaming, and will have to buy a thunderbolt 4 enclosure to realize this vision; I am allowed to dream, maybe it will happen! :slight_smile:

1 Like
#2

Oculink

We have discussed it in this same forum several time. Use the search and you’ll see.

1 Like
#3

@Name2 link? I only found this, which is for the FW13.

#4

Oculink is a standard for carrying PCIe over external cables, and hasn’t been implemented in any laptops I know of. Its possible for an expansion bay module to expose an oculink connection, or even an XG Mobile connection if Asus were to make the module or open source the interface.

#5

@Nich_Trimble that is not what I am requesting here. I want to bypass the expansion bays altogether and connect through the new PCIe 8x connector that is under the FW16.

This is should a PCIe 8x connector with minimal overhead compared to what is offered through the expansion bays.

I am not sure what problem Oculink is trying to solve when we already have thunderbolt 4 / USB 4. I have to look into it to understand, thanks for pointing it out anyway. Update: After looking into Oculink it seems that it only able to carry 4 pcie lanes, and not the 8 pcie lanes that are available in the FW16.

#6

@Can

Yes, Oculink only carries 4 PCI lines, but the 63Gbps that Oculink can reach is enough to saturate any modern GPU, so there is no iaaue there. Thunderbolt 4/USB 4 is limited to 40Gbps and that can give lacking performance, specially in certain games.

You could try running a PCI riser cable, the problem with that is the PCI connection is not made for connecting and disconnecting regularly and could lead to physical failure under regular use. Plus is a more inconvenient solution.

If you want to connect a GPU Oculink is enough.

#7

The expansion bay is the slot at the back of the FW16 with an 8x PCIe interface. Those lanes could be broken out to connect to a GPU somehow; other threads have talked about connecting a full desktop GPU to it with a simple slot. AFAIK, two Oculink cables could be used to bring all 8 lanes somewhere.

1 Like
#8

I’d rather add a block of battery to fill this expansion bay, and use the PCIe lanes for my eGPU.

@Name2, the fewer lanes means also latency; correct me if I am wrong here. Are you guys saying that Oculink has way less overhead than TB4?

1 Like
#9

Occulink is a connector, but doing pcie over occulink has no overhead compared to the lots of overhead of thunderbolt, plus the possibility of going pcie4 which allmost doubles the speed. You do pay for that bay it being just a naked pcie link, no hotplug (at least for now) and no usb or dasy chaining or any of the stuff you get from tb4.

It’s basically a slightly fancier version of plugging a gpu into an m.2 slot that people have allready been doing for years.

3 Likes
#10

Both should be possible in one module, given enough engineering work.

#11

I don’t see the point. Framework 16 will already feature a built in upgradeable GPU. eGPU just seems like overkill.

#12

With an eGPU you can minimize the fanvolume of the Notebook to have a fast but silent environment.

Most people me included already have an eGPU

#13

eGPU is indeed a valid use case, I’d like to connect a desktop 4090 for “serious work”, and get rid of the current desktop computer.

Notably GPD has launched similar system to XG Mobile using OCuLink. Framework should however target OCuLink 2 if feasible as its a newer(2016) and faster standard.

1 Like
#14

@Perrin_Trapp Desktop GPUs by far outperform “same” Notebook GPUs these days - unfortunately. A Notebook 4090 has nothing to do with a Desktop 4090, it’s not even close. So a fast connection like oculink would allow much higher performance when connecting Desktop GPU.

#15

@Philam you are so mistaken… These videos paint a very clear picture:

#16

@Can

Nope.

I knew both of these videos before, and they paint a picture where oculink would loose about 10% of performance compared to a desktop.

So, sorry to say, but there is simply no laptop GPU that comes 10% close to their “same” desktop equivalents.

Here just one video: RTX 4090 Laptop vs Desktop - It’s not even close… - YouTube

We USED to have far closer desktop and laptop performance, back in the Pascal days when all stars aligned, but these times are over. Here some info on the much celebrated GTX 1060 mentioning a potential up to only -1% performance loss to desktops (ah, the days…): Mobile NVIDIA GeForce GTX 1060 (Laptop) - Benchmarks and Specs - NotebookCheck.net Tech

I believe high-end notebook GPUs have these fake names now simply because there are people that do not know. Notebook 4090 obviously does not deserve the name.

#17

You seem idealogically driven so I’ll keep this discussion short so we can move on. Fact as food for though: An RTX 4090 is often bottlenecked by a CPU when playing games in 4k provided you have upscaling such DLSS or FFX.

#18

I had never heard this before, plus intuitively seems this would be very game dependent. I’m not saying you are wrong, but do you have any links where i can dig deeper?

#19

There is plethora of videos out there.