i m wondering how many external monitor i can connect to the laptop both when using iGPU (work) and dGPU (gaming)
could it drive 2 external monitor plus the laptop screen?
Are there special constraint when rendering done on the dGPU (resolution, refresh rate, number of display connected to expansion card,…)?
does the usb-c connector on the GPU expansion bay support Power delivery?
Thanks for the answer.
To provide some additional context, applications are typically rendered on one GPU, though Nvidia SLI and AMD Crossfire/dual graphics can trick certain apps into rendering on two GPUs, provided they have the necessary support. Additionally, there is multiGPU support with DirectX and Vulkan, showing both the GPU and the app determine which load goes on which card. This is commonly used in machine learning programs, while some games might use one GPU for physics and another for graphics.
some laptop run applications on the dGPU while the display is handled by the iGPU (see the test at the end of the article for Asus Zephyrus G14 The process involves computing a new frame on the dGPU, which is then copied to the iGPU and sent to the display.
I want to confirm that we have this dGPU driving display on expansion cards (i m not sure if it’s a pure software implementation or it need hardware/bios support).
for example would it be possible to run a game on 2 external monitor plus laptop screen.
From my reading, I think that driving expansion cards from dGPU would have to be done in software, if it is even possible at all. The only direct from hardware links out of the dGPU are the USB-C port on the rear, which we don’t yet know the capabilities of, and the DisplayPort backfeed on the expansion connector, which is wired directly to the internal screen.
If the iGPU and dGPU can run at the same time, I think (though I’m far from being an expert) it would be possible to drive the equivalent of 8x1080p displays - DP1.2+ MST should be good for 4x1080p equivalent.
I’ve been looking into getting some new monitors, and I’ve kinda fallen in love with 2x21:9 stacked vertically with a 16:18 on each side. If my math checks out, those 16:18s are 2x1080p each, and the 21:9s are ~1.3x1080p each. There’s plenty of room left for overhead
Connect the 21:9s to the dGPU, the 16:18s to the iGPU, profit?
Correct me if I’m wrong - I don’t want to end up with a thousand dollars worth of blank screens.
IIRC both GPUs support 4 monitors (8 total). The internal display only counts towards those limits when it is active.
Due to only having 3 expansion card slots on the mainboard that support external display output and only 1 external display output on the expansion bay a MST hub will be needed. An MST hub allows a single DisplayPort output to operate multiple displays.
The expansion card slots on the laptop connects directly to the iGPU whereas the output on the graphics module connects directly to the dGPU.
When connected to the iGPU (through an expansion card slot) it will still use the dGPU for graphics intensive loads, however it has to route the output through the iGPU which incurs a performance penalty (~10%).
IIRC Framework has said that it supports display output and USB 2.0. I doubt it supports anything else (or they would have mentioned it).
Even more should be possible.
The iGPU supports DP2.0 UHBR10 MST and I think the dGPU supports the same.
That has a bandwidth of 40 Gbps per GPU.
4x1080p60Hz only inherently requires 11.94 Gbps excluding overhead.
Overhead is mainly in the form of blanking (bandwidth that is deliberately wasted to improve display compatibility) and error correction (bandwidth that is used to correct any corruption that occurs).
Including standard blanking and error correction 4x1080p60Hz requires 17.1 Gbps. It’s also possible to use reduced blanking which only requires 13.2 Gbps.
That leaves plenty of bandwidth left over.
I think that might be able to all work on a single port, although I’m not sure what docking stations would support that.
What you listed requires 19.9 Gbps of bandwidth assuming all displays are 60 Hz with 8 bpc color and no compression. Adding error correction and blanking brings that to 29 Gbps with standard blanking or 21.9 Gbps with reduced blanking.
That’s still plenty of bandwidth left over.
So it is theoretically possible for all that to be driven by a single cable using DisplayPort 2.0/2.1 UHBR10 MST.
Even if you step back to DisplayPort 1.4 which only has 32.4 Gbps of bandwidth and spends more bandwidth on error correction it’s still theoretically possible using reduced blanking (which will still require 26.6 Gbps including error correction).
I don’t see any reason it wouldn’t work. Theoretically you could run all of them off of just one of the GPUs if you wanted a single cable solution (although if you want that single cable solution to also do power delivery and/or higher bandwidth peripherals than USB 2.0 you’ll need to connect it to the mainboard port instead of the port on the dGPU, which hurts GPU performance slightly).
That’s far better news than I could’ve hoped for! The 21:9s I’m looking at are 2560x1080@200Hz, but can be fixed to 60 and 120 if needed (probably 144 also but I can’t find reference for it). So if I’ve worked it out correctly I’m about 1Gbps over the limit with standard blanking if I run at 120Hz. With reduced blanking though it comes in at around 31Gbps total - that seems like I should be able to push it to 144Hz on both if I wanted to?
Realistically, I would be fine with running 60Hz on all screens (120Hz on the main display seems achievable), and turning off all but the main display when I care about high frames. Day to day I do programming, read text, and watch YouTube, all of which is fine at 60Hz. And if it comes to it, I can run those side screens off the iGPU and let the ultrawides eat all the bandwidth they need.
Monitor models for reference. I’m not at all tied to these models, just these aspect ratios, size (ish), and budget.
21:9 - ROG Strix XG309CM
16:18 - LG 28MQ780-B
My current thinking is to splash on a high quality MST Hub and connect the monitors to that via DisplayPort. I’m not worried about the PD aspect, I can cable manage that problem away.
The 16:18s I’m looking at support being used as two distinct 1080p displays via two HDMI connections. Say I split for the four physical monitors with an MST hub then split a DP to 2xHDMI, cable nightmare aside is there any reason the FW16 couldn’t do that? It would be slightly more convenient from a window tiling perspective.
The peripherals thing though, I can connect USB 2.0 devices through the dGPU? I have seen some MST hubs with USB ports on them, but I had just assumed that’s because the expected input was USB 3.2/USB 4 (the ports are marked USB 3.X), I didn’t realize DP alt could do USB data - like i said I’m far from being an expert!
That would be really useful for just hooking up the low data/low power stuff, though I have no idea what i would do with the expansion card slots. Power in, ethernet, what else is there? Do I just rig up a bunch of M.2 drives? Starting to realize just how awesome the FW16 is gonna be.
Here’s a tool I use to calculate the bandwidth requirements of a display. The column titled “CVT” is for standard blanking, the columns titled “CVT-RB” and “CVT-RBv2” are for reduced blanking. The row titled “Peak BW” is how much bandwidth it requires in the cable.
It doesn’t calculate bandwidth consumed by error correction, although if you scroll down it will list what percent of the available bandwidth it will consume under various standards. Look at the line labeled “DisplayPort UBR10 (10 GHz)” for DisplayPort 2.0/2.1 or “DisplayPort HBR3 (8.1 GHz)” for DisplayPort 1.4. The rows under “DisplayPort Type-C Alt Mode” are different (See below).
As long as the percent of the bandwidth consumed adds up to less than 100% you should be good.
Many displays will run CVT (standard blanking by default), however I’ve never had an issue with forcing CVT-RB (reduced blanking) by adding a custom resolution in GPU settings or using CRU (Custom Resolution Utility, allows for customization of things like blanking).
The Strix XG309CM matches the description you gave above of “the 21:9s are ~1.3x1080p”, however the LG 28MQ780-B is actually similar to 2x1440p rather than 2x1080p like you mentioned, which increases bandwidth requirements quite a bit.
USB-C has 4 “SuperSpeed” lanes plus a USB 2.0 lane. SuperSpeed lanes can be used for multiple purposes.
Typically I see people use “USB-C DP alt mode” to refer to a mode where 2 of the SuperSpeed lanes are being used for DisplayPort (with crippled bandwidth) and 2 of the SuperSpeed lanes are being used for USB 3.
The Framework Expansion Module however is intended to use a different mode where all 4 of the SuperSpeed lanes are being used for DisplayPort (full bandwidth) and the only other data lane available is the USB 2.0 lane.
Many of the USB-C docks expect the first mode and do not support allocating more than 2 SuperSpeed lanes being used for DisplayPort, so when buying a dock you need to make sure it supports the latter mode.
The dual display variants of that adapter are basically the single displayport variant with a built in dual port MST hub (and DisplayPort to HDMI adapters for the last variant). I think it should also be possible to take the single displayport variant and connect an external MST hub like this model.
Those adapters use the USB 2.0 capabilities to offer two USB 2.0 ports (great for peripherals) and a ethernet port (that is limited to the 480 Mbps of USB 2.0, which is not super fast for ethernet but still faster than most internet connections). Unfortunately I can’t find any DisplayPort 2.0 capable hardware for this so you may be limited to DisplayPort 1.4 bandwidth.
DisplayPort does support a feature known as “Display Stream Compression”, which can compress the data to allow only a third as much bandwidth to be consumed by the display, which can help you to fit everything into as few cables as possible but hurts quality slightly.
Personally I don’t have much need for high bandwidth peripherals so I would use the aforementioned CableMatters adapter connected to the dGPU and connect the 180w power brick to the laptop, which would allow for a lot of display bandwidth, 180w charging, and basic USB 2.0 peripherals all over 2 cables.