Exotic Bug/Bandwidth Limitation with Intel Graphics when maxing out attached external displays

Hi,

I have found a very strange and exotic bug with my Intel GPU and external monitors. In short and generalized as much as I am currently aware:

When attaching a “high-bandwidth” external monitor, while the GPU is already driving the maximum supported displays, that “high-bandwidth” display gets limited in bandwidth. But another order of enabling/connecting the display has everything working as expected.

Definitions/Details:

PC: I have a Dell XPS 15 (9500), so Comet Lake Graphics, which supports Displayport HBR2 and 3 screens at the same time (up to the full HBR2 bandwidth). So 4xUHD@60Hz 10Bit is technically possible, if you have dual channel memory that is fast enough.

High-Bandwidth monitor, for me, with this GPU means, everything that gets close to the full HBR2 bandwidth. I could reproduce the problem with UHD@60Hz displays (HDMI 2.0, still some headroom to HBR2) , UWQHD@120Hz (basically maxing out HBR2) and 3840x1600@85Hz (also still within HDMI 2.0 bandwidth).

Bandwidth limited: The monitors do not achieve the same bandwidth (approx. Resolution x color depth x Refresh Rate) they normally do on the same machine. In practice this means for example UHD@60 displays become limited to UHD@30, the UWQHD@120Hz display becomes limited to 6bit color depth with 85Hz or similar.

I have now found multiple monitor setups, in which it essentially depends on the order in which I plug them in to get them working correctly.

Why am I writing this:

  • Maybe screeming into the void lessens my anger a little
  • documentation for other people running into this problem. It surely is rare, because I am not aware of many machines that offer enough ports of an Intel iGPU to even connect enough displays to run into the problem the same way I have
  • Maybe some of you have enough equipment & time on your hands to try to reproduce this with different HW. I am especially intrested in other notebooks, PCs. This might light a fire under Intel’s ass to fix that problem (which I am betting is in the driver). Or just provide them more examples with which to reproduce the issue. They told me, they are still procuring the required HW. As UHD Graphics between Skylake and Comet Lake are relatively similar, maybe more generations of iGPU are affected. As it stands I cannot imagine, how Dell could have caused this bug only on my specific machine. Also interesting, but even more difficult to test, would be if the same issue exists with Intel Xe Graphics in Tiger Lake / Rocket Lake. But as these support HBR3 and for Tiger Lake 4 displays simultaneously, I think it is probable, that one would need 5 displays, with the high-bandwidth one coming close to HBR3 bandwidth to even try to test this.

Steps to reproduce:

  1. Own/Have access to a device with most of the display outputs of the IGPU exposed, preferebly via Displayport.
  2. connect & enable 3 monitors (the max supported count). 3xFHD@60 worked for me. Internal Display of a notebook counts also towards this. I could daisy chain some of them or connect the displays, each to their own displayport, it did not affect the results.
  3. connect a “high-bandwidth” display. Because the GPU can only drive 3 displays at the same time it will remain dark for now. Then, disable any one of the 3 already enabled displays and turn the “high-bandwidth” display on. It should now be visibly limited. Windows/Intel driver should not offer all the refresh rates expected, or silently downgrade to 6bit. This can be easily verified via the new “Advanced Display” menu in windows 10/11.
  4. (alternatively to 2.): With Windows remembering a past display configuration with the 3 external displays, when attaching the “high-bandwidth” display, windows automatically disables 1 display and enables the “high-bandwidth” display all in one go. Just as previously setup.

Steps to workaround:

  • Rebooting the machine solved this for me from any limited combination
  • Linux (Fedora 34, Kernel 5.14.9) does not have this problem.
  • before step 2/3, disable in windows or just disconnect any other monitor, such that the GPU is not already driving the maximum amount of displays

I do not know, whether non-high-bandwidth displays get limited as well, and I just cannot see it (because they do not need the full bandwidth), or if they are just not affected. All I can say is, my Dell U2520D always shows an HBR2 connection and always has enough bandwidth left, to supply another daisy-chained display (both WQHD@60, so together, close to full HBR2 bandwidth).

Tested Configurations:

Following a table of monitor-combinations I tested

PC iGPU Auxiliary Monitors High-Bandwidth Monitor (max possible settings with PC) Result
Dell XPS 15 (9500) Comet Lake UHD Integrated Display (UHD+@60), Dell U2520D (WQHD@60) via DP, U2515H (WQHD@60) via daisy-chain off U2520 or directly attached Alienware AW3821DW (3840x1600@85) via HDMI/DP AW38 limited depending on order
Dell XPS 15 (9500) Comet Lake UHD Integrated Display (UHD+@60), Dell U2520D (WQHD@60) via DP, U2515H (WQHD@60) Sony XF90 (UHD@60) via HDMI XF90 limited depending on order
Dell XPS 15 (9500) Comet Lake UHD Integrated Display (UHD+@60), 2x Dell U2417H (FHD@60) via DP (daisy-chained or direct) Alienware AW3418DW (UWQHD@120) via HDMI/DP AW34 limited depending on order
Dell XPS 15 (9500) Comet Lake UHD Samsung XL2370LED (FHD@60) via HDMI, Dell U2520D, U2515H (WQHD@60) via DP, daisy-chained Alienware AW3821DW (3840x1600@85) via DP AW38 limited depending on order
Dell XPS 15 (9500) Comet Lake UHD Integrated Display (UHD+@60) Alienware Alienware AW3821DW (3840x1600@85) via DP, Sony XF90 (UHD@60) via HDMI, Dell U2520D at UHD@60 (via HDMI in) last high-bandwidth display limited
Dell XPS 15 (9500) Comet Lake UHD Integrated Display (UHD+@60), Samsung XL2370 (FHD@60), U2520D, U2515H (WQHD@60) works
Dell XPS 15 (9500) Comet Lake UHD NEC E231W (FHD@60) via VGA-adapter, 2x Dell U2417H (FHD@60) Alienware AW3418DW (UWQHD@120) AW34 limited depending on order
Lenovo ThinkCentre M710t Kaby Lake HD 630 NEC E231W (FHD@60) via native VGA, 2x Delll U2417H (FHD@60) Alienware AW3418DW (UWQHD@120) works

Monitor Limitations

Monitor Native Resolution/Max Bandwidth Max Bandwidth with HBR2 GPU Limited Bandwidth with HBR2 GPU
Integrated Display (XPS 15 9500) 3840x2400@60 8b same as native never limited
Sony XF90 HDMI 2.0, UHD@60, 8b RGB same as native UHD@30 8b RGB
Dell U2520D WQHD@60 10b via DP, UHD@60 8b RGB via HDMI same as native UHD@30 8b RGB via HDMI, not limited via DP
Alienware AW3418DW HBR2, UWQHD@120 8b same as native UWQHD@85 6b RGB
Alienware AW3821DW HBR3, 3840x1600@120 10b /@144 8b 3840x1600@85 8b 3840x1600@60 6b via DP, 3840x1600@30 8b via HDMI

Why do I care?

I have Thunderbolt-Dockingstations for my 2 monitor setups and wanted to simply drive the same 3 external displays I am also using with my desktop PCs. As with that dockingstation, I leave the displays plugged in permanently and attach the notebook with a single cable, I cannot really control the order in which the displays are detected/recognized/enabled. Thus most of the combinations of plugging the monitors into the dock end up limited. There are only very few combinations (using additional adapters that should not be needed), that reliably work at full bandwidth. For the longest time, I believed the docks to be at fault, until I reproduced the same limitations without any docks.
Also, by now, I have invested considerable time (and some equipment) into this and I’d just like to now that it lead to an improvement. Do I need another reason to want such a crazy bug fixed by the ones who caused it?

Thanks for reading this far.
If anybody has any further suggestions how to troubleshoot / narrow down the cause even more, I’d be happy to try. Any recommendations for an affordable displayport sniffer? If anybody wants to attempt to reproduce this, I’d be happy to help and further fill out the tables.

Update 1:

Intel reports that they cannot reproduce this with a reference board and UHD-Monitors (specs not further qualified).
I found a Lenovo Desktop with 2 Displayports + VGA on 7th gen Intel which also does not have the same bug, even when using the exact same displays as with my notebook. Results have been added to the table above.

2 Likes

Hi

It shouldn’t suprise anyone running of an x86 based IGPU that the bandwith gets limited, with the amount of monitors connected. At higher resolutions., Even more so.

That problem is probably gone with the amount of bandwith offered by latest Apple silicone, m1. I do wonder how many displays are possible, how much power it would use and in what kind of resolutions. All running at the same time on that incredible m1 chip.

Not trying to sell anyone on Apple, but benchmarks on the m1 does speak for itself. With the power it offers at the scale it works at, if not revolutionary is incredibly impressive. Even more so than the performance jump in x86 computing (Example Ryzen)

Otherwise a pretty comprehensive read, thanks :+1:

That should actually have basically nothing to do with the processor being x86. Display-Pipelines are dedicated HW inside the GPU. They should only be limited by their own throughput, maximum bandwidth of the outputs and the bandwidth of the memory from which the pixel-contents need to be read.

A simplified calculation shows us that driving a UHD@120Hz 10Bit Display would require around
3840x2160x30x120/8 Bytes/s which comes out at around 3.7 GB/s.

Now, granted Apple has dramatically increased the memory bandwidth of their M1X SoCs and Intel reused their old GPUs for far too long (still limited to HBR2 up until Comet Lake). But as you can see, even with dual channel DDR4 (My system has 45.8GB/s of memory bandwidth per Intels spec), supplying that kind of bandwidth to 4 monitors is not a problem. Although this might notably cost CPU performance, as that bandwidth is no longer available to the CPU. On the other hand, memory accesses from frame buffers are very straight forward and easy to prefetch and pipeline, so are likely to be less expensive than any CPU-Process requiring that much bandwidth to more random locations.

Also, the first M1 processor supported only 2 displays, even bound to particular ports, which was a large setback from the 3-4 displays on any displayport connector supported by Intel and AMD and previous Macs at the time. Also, as I tried to describe, my notebook can drive 3 UHD@60 displays just fine from a HW perspective. Its the software/driver that imposes arbitrary limits on top of that.

1 Like

Update

Intel reports that they cannot reproduce this with a reference board and UHD-Monitors (specs not further qualified).
I found a Lenovo Desktop with 2 Displayports + VGA on 7th gen Intel which also does not have the same bug, even when using the exact same displays as with my notebook. Results have been added to the table above.

I’m just here to report that i commend you for all that work, like you said looks silly to invest that amount of time on it, but indeed i do things like that also and keep digging is the only way.

I’m totally stupid for drivers, so i will type something that maybe makes no sense to even test.

If you have 2 monitors and connect a bunch of external USB components and use them… the monitors throtlle? Basically what i mean is to check if is not power limitations, because i have no clue about 90% of the technical stuff you just wrote and sometimes that can help narrowing down an issue, someone with a different point of attack LOL

Sure, spitballing ideas is half of why I typed all of it up here.

So power limitations, mhh. I cannot imagine this being a problem, because all of the behavior stays the same, with power supply or without and also when using the official Dell Dock with 180W power supply to drive all other peripherals. I would have expected some change there if it had something to with power.

But also, with it not being reproducible on other systems, I do not have many educated guesses left. I considered signal integrity being a problem, like too many high-frequency signals making the connections unstable, but then I would have expected some variance with which ports I am using. And I tried to use all sorts of combinations of the 3 USB-C ports on my device and no change…

My reasoning for it basically boils down to Dell not wanting to fry anything and get sued so they limit the power draw, i worked for Mobile companies and they where really anal about those types of protection, and the Thinkpad doesn’t give a fuck.

Also you could try to disable DDC/CI on the monitors, why? idk, takes 2 seconds just one less thing to communicate.

Outside of those things i legit have no clue

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.