Question Can't connect 4 monitors to a 3090Ti GPU

9-Ball

Distinguished
Apr 20, 2010
111
0
18,690
I have an odd issue with a Gigabyte 3090Ti connecting to 4 monitors.

My DP1 port connects to a main screen Samsung Neo G9 49" at 240hz, without issue.

DP2 is connected to a secondary LG 24" monitor at 144hz, with no issue.

DP3 is connected to a 22" basic 60hz touchscreen monitor - still working.

As soon as I plug a second basic 22" 60hz DP or HDMI monitor into the remaining HDMI slot, it will display but I lose signal to the main 49" immediately. I've tried multiple cables, different DP, HDMI, adapters, etc, but no luck.

I can't find obvious known issues with using the 4 output ports on this card. From what I read at Nvidia and elsewhere, all four ports should be able to run this without problems. I'm wondering if there are video settings or other issues with cables/refresh rates that might explain it, or is it likely a hardware fault with the GPU or HDMI port? It feels like the old days of juggling DVI adapters and port limitations.

I've also just ordered a DP-to-2x-DP adapter, which from the description should be able to run the two small 22"/60hz screens off one DP, removing HDMI from the mix.

Any ideas or advice?
 
Solution
Well, I got it working, I think, but I'm not altogether sure why, which is frustrating.

I switched the DP version on the 2560x1440 second screen to DP 1.2 from 1.4. I connected that screen and one of the 1920 side screens to a DP 2x splitter and input those to the 3090ti port 2. The Samsung Neo G9 remains on DP 1.4, port 1, at 5440x1440 and 240hz. Those three all work normally, as of writing.

Then I connected the second 1920 side screen via a DP to USB-c adapter to the mobo's USB-c port, intending to use the 12900k's integrated graphics. It turned on immediately and appears to be working fine at 60hz.

I'm 99.99% sure I already tried exactly this setup and it didn't work, but now it does. Posting just in case it helps someone else...
If you're running an Intel processor with an iGPU, you could hook one of the panels to the motherboard's display output ports. Got a link to the GPU you're working with? As for your GPU, you could try and use DDU to uninstall all of your current GPU drivers on the platform, then manually reinstall the latest Nvidia driver in an elevated command, i.e, Right click installer>Run as Admnistrator.
 
Quick addendum, since my only goal here is to get the 3rd and 4th screens running. My mobo is https://us.msi.com/Motherboard/MEG-Z690-ACE which has these Thunderbolt 4 ports listed:

2x USB Type-C up to 40G, Charging support up to 5V/3A, 15W
2x Mini DisplayPort input ports

With a 12900k, including integrated UHD 770 graphics, which displays correctly in Device Manager.

Those Thunderbolt/USB-c ports can be seen here [https://asset.msi.com/resize/image/....png62405b38c58fe0f07fcef2367d8a9ba1/1024.png] to the right of the 4x USB slots in the middle.

Is there a way to utilize these for the 2x small screens and keep the main screens on the 3090? I see some videos with an adapter running from a DP port on the GPU to the mobo port but I'm not sure what that is or how it should be arranged. Again, advice much appreciated.
 
If you're running an Intel processor with an iGPU, you could hook one of the panels to the motherboard's display output ports. Got a link to the GPU you're working with? As for your GPU, you could try and use DDU to uninstall all of your current GPU drivers on the platform, then manually reinstall the latest Nvidia driver in an elevated command, i.e, Right click installer>Run as Admnistrator.
Thanks for the quick reply - see my addendum. I just had the same thought.
 
AFAIK there is a pixel rate limit to each internal port inside the GPU, and if a display requires more than this limit, the GPU will combine two internal ports to drive the display, which will reduce your maximum number of displays from 4 to 3. Try lowering the refresh rate/resolution on the 49" to confirm this.

When a display is connected to the GPU and is set to DSC mode, the GPU may use two internal heads to drive the display when the pixel rate needed to drive the display mode exceeds the GPU’s single head limit. This may affect your display topology when using multiple monitors. For example if two displays with support for DSC are connected to a single GeForce GPU, all 4 internal heads will be utilized and you will not be able to use a third monitor with the GPU at the same time.
 
Last edited:
AFAIK there is a pixel rate limit to each internal port inside the GPU, and if a display requires more than this limit, the GPU will combine two internal ports to drive the display, which will reduce your maximum number of displays from 4 to 3. Try lowering the refresh rate/resolution on the 49" to confirm this.

Same results with the 49" at 120 and the 24" at 85hz. Not being able to run the 49" at the 240 rather defeats the object.

Any idea if the DP to 2x DP adapter on one port might impact this?
 
According to spec documentation, the 3090 max combined output frequency is 1695 MHz, which would suggest the 240/144/60/60 screens should be compatible, unless I am misunderstanding that?
 
According to spec documentation, the 3090 max combined output frequency is 1695 MHz, which would suggest the 240/144/60/60 screens should be compatible, unless I am misunderstanding that?
That's the GPU clock frequency, which is not related to the pixel rate limits for display output. I don't think NVIDIA documents the limit.
Same results with the 49" at 120 and the 24" at 85hz. Not being able to run the 49" at the 240 rather defeats the object.
Of course, I'm just trying to confirm the source of the issue. Can you try setting it to 60 Hz just to be sure?
 
I can set the 49" and 24" screens to 60 and the first small side panel to 59, and the same happens when I connect the second 22" HDMI port screen. Appears to behave identically regardless of refresh rate, based on that.
 
I have an ASUS ROG Maximus XIII Extreme and an MSI RTX3080 graphics board. Not vital for my flight simulation setup, but trying to understand what is possible:

52991991048_5b79ebb7a1_c.jpg


As you can see in the foto I have running with no problem 7 monitors. 2x using the Thunderbolt 4 I/F of the motherboard, the 2 on top. 3 monitors on the MSI RTX 3080 are connected to the dispülay ports of the RTX3080. The monitor on the left side which you can only see on the backside of it is connected to the HDI port of the motherboard. I did not expect to have my system to support this, but it does. The center monitor is a 32" Samsung Odyssey G9 running at 240 MHz and a resolution of 3840x2160. The 2 monitors on the side, are 28", 60 Hz, 3840x2160 resolution. the 2 24" monitors on top are running at 60Hz and have a resolution of 1920x1080. The 24" monitor al the far left runs at 60Hz and 1680x1050.

The 7th monitor on the far right, where on the foto is an empty space is a 19" and connected to the HDMI port of the RTX 3080. The monitor is detected by the settings of the W11 Pro but is not active. I repeat I am impressed that my system can run 6 monitors at the top resolution and refresh rate of the monitors.
 
I have an ASUS ROG Maximus XIII Extreme and an MSI RTX3080 graphics board. Not vital for my flight simulation setup, but trying to understand what is possible:

52991991048_5b79ebb7a1_c.jpg


As you can see in the foto I have running with no problem 7 monitors. 2x using the Thunderbolt 4 I/F of the motherboard, the 2 on top. 3 monitors on the MSI RTX 3080 are connected to the dispülay ports of the RTX3080. The monitor on the left side which you can only see on the backside of it is connected to the HDI port of the motherboard. I did not expect to have my system to support this, but it does. The center monitor is a 32" Samsung Odyssey G9 running at 240 MHz and a resolution of 3840x2160. The 2 monitors on the side, are 28", 60 Hz, 3840x2160 resolution. the 2 24" monitors on top are running at 60Hz and have a resolution of 1920x1080. The 24" monitor al the far left runs at 60Hz and 1680x1050.

The 7th monitor on the far right, where on the foto is an empty space is a 19" and connected to the HDMI port of the RTX 3080. The monitor is detected by the settings of the W11 Pro but is not active. I repeat I am impressed that my system can run 6 monitors at the top resolution and refresh rate of the monitors.
Interesting - thank you. Could you show and/or explain how you have the 3080 connected to your motherboard in order to activate those ports, or are they simply running off the mobo with a CPU with integrated video? If the latter, what cables are you using to connect to the mobo and what mobo is it? Much appreciated if you can elaborate. It looks like my board should do the same thing, and I only need to run 4 screens, ultimately.
 
The MSI RTX3080 is just plugged into the slot of the motherboard. The 3 monitors, 1x 32" and 2x 28" are just plugged into the data port connectors of the graphics card. The 2x 28" monitors are connected to the Thunderbolt 4 connectors of the motherboard. The 24" monitor is connected to the HDMI connector of the motherboard. So the only connector left in my PC is the one the 4th one left on the graphics card. That is the one W11 Pro can detect is connected to the graphics card but not activated. Here is the link to the motherboard.
 
Can you tell me what cables/adapters are connecting the 28" screens to the mobo Thunderbolt 4 ports? Are these DP or HDMI to T4, or using adapters? Did you have to do anything in bios to enable those ports?

It looks like you're not connecting the GPU card to the mobo except via it's PCIE slot, i.e. no cable out the back of the GPU and into the mobo, which is how I understood this could or should work, so I'm confused.

I've connected my 2 side screens with DP-to-Mini-DP cables to the mobo ports, but no output. I suspect my DP to mini-DP cables aren't the same as DP to T4/usb-c.
 
For the Thunderbolt 4 connectors, I use a cable from the C-USB connector to the display port. I always make sure I buy the highest quality cables I can get hold of.
Just to make sure I understand correctly, this means you have one GPU displayport filled by a cable that runs to the C-USB on the mobo, and then the monitors connect to the Thunderbolt4 out ports on the mobo? If so, that fits what I understood for my mobo too, and the cables I ordered last night should make that work, I think.
 
I tend to like earlier posts about maximum bit rates (pixels, fill rate, etc.). You might just be exceeding capacity. Physical monitor size won't matter, but higher resolutions use that performance up because it requires more VRAM.

Something you might consider just for getting more information: Boot to a live Linux distribution, e.g., KUbuntu is good for this. Monitor "`dmesg --follow`". Then plug in that last monitor, and see what shows up. Also, there will be a log, "`/var/log/Xorg.*.log`" (there might be multiple logs for this case, so the wildcard, '*'), and you could copy those to a thumb drive or something and post them.
 
I tend to like earlier posts about maximum bit rates (pixels, fill rate, etc.). You might just be exceeding capacity. Physical monitor size won't matter, but higher resolutions use that performance up because it requires more VRAM.

Something you might consider just for getting more information: Boot to a live Linux distribution, e.g., KUbuntu is good for this. Monitor "`dmesg --follow`". Then plug in that last monitor, and see what shows up. Also, there will be a log, "`/var/log/Xorg.*.log`" (there might be multiple logs for this case, so the wildcard, '*'), and you could copy those to a thumb drive or something and post them.
Thanks for the ideas. That said, the GPU has a fill rate of 211 gigapixels per second. A single 4k screen at 60hz is just under 5 million pixels/sec. Ten such screens connected, hypothetically, wouldn't even come close to that cap. It's not a pixel fill issue, given there are tens of thousands of people using 4 and more screens off a single 3090 ti's 4 ports, at far higher res/refresh rates than I'm trying to do. Installing Linux solely for the purpose of diagnosing this seems a bit ... well, Linuxy? 😉
 
Thanks for the ideas. That said, the GPU has a fill rate of 211 gigapixels per second. A single 4k screen at 60hz is just under 5 million pixels/sec. Ten such screens connected, hypothetically, wouldn't even come close to that cap. It's not a pixel fill issue, given there are tens of thousands of people using 4 and more screens off a single 3090 ti's 4 ports, at far higher res/refresh rates than I'm trying to do. Installing Linux solely for the purpose of diagnosing this seems a bit ... well, Linuxy? 😉
Pixel fillrate isn't related here either, this is about the limitations of the display outputs. The GPU may render 211 Gpx/s, which is a global limitation of the GPU rendering capabilities. This is separate from the display output capabilities, since the GPU may render many more pixels than are actually displayed, during anti-aliasing operations for example. In addition, each display interface will have its own limitation on the output pixel rate which is specific to that interface. The DisplayPort interface on Kepler GK104 for example was limited to 540 Mpx/s per port. This is independent of the GPU pixel fillrate (which was 33 Gpx/s on Kepler).

You can also see this NVIDIA whitepaper as another example, the Maxwell GTX 980 has a pixel fillrate of 77 Gpx/s (page 8) but the maximum output pixel rate on DisplayPort is 1045 Mpx/s (page 12).


That limit is per port. But there also seems to be a global output pixel rate combined across all interfaces which is not published. People can and have run into these limits trying to run multiple high-resolution high-refresh rate displays.


But this one is a sort of separate issue as it has to do with the number of independent displays allowed, which is normally 4 on NVIDIA GPUs. As mentioned in the NVIDIA support article, when the pixel rate exceeds a certain (unspecified) limit then it will use two internal heads instead of one, consuming an extra count of allowed displays.

People can and have run into this issue as well:

But why lowering to 60 Hz wouldn't re-enable it, is the mystery to me. Perhaps some other limitation I'm not aware of. Not sure.
 
Thanks for the ideas. That said, the GPU has a fill rate of 211 gigapixels per second. A single 4k screen at 60hz is just under 5 million pixels/sec. Ten such screens connected, hypothetically, wouldn't even come close to that cap. It's not a pixel fill issue, given there are tens of thousands of people using 4 and more screens off a single 3090 ti's 4 ports, at far higher res/refresh rates than I'm trying to do. Installing Linux solely for the purpose of diagnosing this seems a bit ... well, Linuxy? 😉
You are right about fill rates, but I am also wondering about VRAM associated with this. I don't know, but maybe it just does not have enough VRAM. As an experiment, could you temporarily reduce the resolution of the highest resolution monitor by some significant amount, and then try the other monitor?

A live DVD (or thumb drive) never gets installed unless you tell it to do so. It runs in RAM (it takes a long time to boot). There wouldn't be any modification of the system, but the X server for the GPU would create logs (most likely describing the hardware). I do actually prefer the NVIDIA driver instead of the default driver, and this would give you much better information than the generic driver. Installing the NVIDIA driver would also not modify the system since it is entirely in RAM if you don't tell it to install (and it will be obvious before any install, you just have to select the live boot and not install).

Do note that all of the *NIX o/s variants (of which Linux is one) separate the GUI software (and text shell software) from the actual operating system. Unlike Windows, the GUI is not "mandatory" and invasive. It is a simple matter to go back and forth between a purely text login and a GUI login (or both simultaneously). On Ubuntu (which KUbuntu is one flavor) this turns off the GUI but leaves text login:
sudo systemctl isolate multi-user.target

This turns on the GUI from text mode:
sudo systemctl isolate graphical.target

This gives you the ability to install GUI drivers and reboot the GUI without rebooting the entire system. Pretty useful on a live DVD or thumb drive. The default log file from NVIDIA has a lot of good information, but you could configure for even more logging ("make it sing and dance").

Most of what you see will be a result of the "mode pool". That's the driver reading the EDID (basically plug-n-play broadcast from the monitors of their capabilities) and producing an intersection of monitor capabilities and GPU capabilities, and then picking a mode within the mode pool. If you don't enable verbose mode, you get a lot of information about what is selected on each monitor (each has an EDID). If there is an error, then this is also logged. If you enable verbose EDID logging, then you get the driver itself explaining every single monitor mode and why that mode is rejected (or accepted).

The point though is that even without any special edits, if you live boot, you get good information even from the generic driver. If you want to install the NVIDIA driver, it gets a lot better (and you don't have to reboot if the GPU driver is for that kernel; you would likely have to leave the GUI, graphical.target to multi-user.target, but that is a trivial thing to type on the command line.

Regardless of o/s, you can find drivers here:
https://www.nvidia.com/Download/index.aspx?lang=en-us

Incidentally, once you have a bootable thumb drive or DVD, it is a great rescue tool. I'd recommend KUbuntu 22.04. which is the "long term service", or LTS. If you can get past the effort to create the DVD or thumb drive, and don't mind the long boot time, it is a great tool for a lot of things.
 
This might be a stupid question (I know) but on reading numerous articles and posts about Maximum Digital Resolution caps on the 3090 cards, I get very mixed messages. Officially the MDR for the card is:

Multi-monitor:Up to 4 displays
Maximum digital resolution:7680 x 4320 @60 Hz
Maximum DP resolution:7680 x 4320 @120 Hz
Maximum HDMI resolution:4096 x 2160 @ 60 Hz

Does this mean per port, or total? In other words, a 5120x1440+2560x1440+1920x1080+1920x1080 setup equals 11,520x5.040, such that it would exceed that total for a combined total output, but not if it's per screen. I see dozens of posts saying it's per port/screen, so all 4 should work, but a few that seem perhaps more detailed stating it's a combined total, which might explain the issue, except I can connect 3 of the 4 without issue, which exceeds the cap. Soft cap? Nonsense cap? Plot thickens.

I can get three screens running with the 2 big ones to the 3090 and one via DP to USB-c into the mobo. Stuck with the 4th.

Any thoughts or experience on whether a USB-c splitter/hub like this [https://sidetrak.com/products/sidet...utm_campaigntype=PMax&utm_sku=&utm_productid={productid}&utm_keyword=&utm_device=c&utm_persona=&utm_productcategory=&utm_adtype=&utm_adname=&utm_adgroupname=&utm_targeting=Smart&trueroas=17964116548&gclid=Cj0KCQjw5f2lBhCkARIsAHeTvlhFKWZehbEZyzbBrSpqVMcSNYTqz_N6p4FBleGDX9Qw7yWhj0S2w5YaAuT7EALw_wcB] would allow two DP-to-usb-c cables and run both little screens?
 
This might be a stupid question (I know) but on reading numerous articles and posts about Maximum Digital Resolution caps on the 3090 cards, I get very mixed messages. Officially the MDR for the card is:

Multi-monitor:Up to 4 displays
Maximum digital resolution:7680 x 4320 @60 Hz
Maximum DP resolution:7680 x 4320 @120 Hz
Maximum HDMI resolution:4096 x 2160 @ 60 Hz

Does this mean per port, or total? In other words, a 5120x1440+2560x1440+1920x1080+1920x1080 setup equals 11,520x5.040, such that it would exceed that total for a combined total output, but not if it's per screen. I see dozens of posts saying it's per port/screen, so all 4 should work, but a few that seem perhaps more detailed stating it's a combined total, which might explain the issue, except I can connect 3 of the 4 without issue, which exceeds the cap. Soft cap? Nonsense cap? Plot thickens.

I can get three screens running with the 2 big ones to the 3090 and one via DP to USB-c into the mobo. Stuck with the 4th.

Any thoughts or experience on whether a USB-c splitter/hub like this [https://sidetrak.com/products/sidet...utm_campaigntype=PMax&utm_sku=&utm_productid={productid}&utm_keyword=&utm_device=c&utm_persona=&utm_productcategory=&utm_adtype=&utm_adname=&utm_adgroupname=&utm_targeting=Smart&trueroas=17964116548&gclid=Cj0KCQjw5f2lBhCkARIsAHeTvlhFKWZehbEZyzbBrSpqVMcSNYTqz_N6p4FBleGDX9Qw7yWhj0S2w5YaAuT7EALw_wcB] would allow two DP-to-usb-c cables and run both little screens?
It's per port, and there is no such actual resolution limit at all. They just give those as examples of formats that are close to the bandwidth limits of the ports. There is no cap on the resolution itself. It's simply a matter of bandwidth. DisplayPort HBR3 runs up to 25.92 Gbit/s, anything that falls within that limit is supported on each DP output (subject to other limitations I described above).

For example back in the day, the GTX 580 stated "maximum resolution 2560×1600", but you can run 3840×2160 on it just fine if the refresh rate is lowered to 30 Hz to make the total bandwidth the same. The GTX 780 states "maximum 4096×2160" but people ran 5120×2880 displays on them just fine, or connected multiple 4K displays. It's not a combined limit. Some people just assume that it is, don't test it, don't search the internet to see if anyone else has tested it, and then start giving advice to other people about it, because internet :)

But, it's always possible there is something different about RTX 3000, you never know what random undocumented limitations NVIDIA hasn't told us about with each generations, there may be some global combined pixel count limit (I doubt it would be 7680×4320, but maybe something higher). You're welcome to try lowering the resolution to see if that helps.
 
Last edited:
It's per port, and there is no such actual resolution limit at all. They just give those as examples of formats that are close to the bandwidth limits of the ports. There is no cap on the resolution itself. It's simply a matter of bandwidth. DisplayPort HBR3 runs up to 25.92 Gbit/s, anything that falls within that limit is supported on each DP output (subject to other limitations I described above).

For example back in the day, the GTX 580 stated "maximum resolution 2560×1600", but you can run 3840×2160 on it just fine if the refresh rate is lowered to 30 Hz to make the total bandwidth the same. The GTX 780 states "maximum 4096×2160" but people ran 5120×2880 displays on them just fine, or connected multiple 4K displays. It's not a combined limit. Some people just assume that it is, don't test it, don't search the internet to see if anyone else has tested it, and then start giving advice to other people about it, because internet :)

But, it's always possible there is something different about RTX 3000, you never know what random undocumented limitations NVIDIA hasn't told us about with each generations, there may be some global combined pixel count limit (I doubt it would be 7680×4320, but maybe something higher). You're welcome to try lowering the resolution to see if that helps.
Thanks for the detailed info. That fits more with what my understanding was before going down this rabbit hole. I've had 3-6 screens connected to one or two GPUs since the early days of dedicated GPUs, sometimes with issues between cables and ports, but nothing like this.

Unfortunately, that means I'm back to square one. I see dozens of users with 3090's or even the same 3090ti I have with 4 screen connected, including an ultrawide and ultimately higher resolutions/refresh rates than any of my totals. I can plug every screen into every port without issue. Yet, something prevents more than any three working together, such that it seems the fourth port on my 3090ti is disabled when a fourth cable is plugged in. That port works fine on its own with any screen. Starting to suspect a mechanical problem with the card, till I did that last test to see all ports work normally.

That leads me to your point about some hidden technical limitation that's not documented, yet I still see others running more demanding res+ref setups on the same card and lesser CPUs, so that doesn't add up, either. I also can't just plug three into the GPU and one into the mobo's USB-c - same behavior, anything works into the usb-c mobo port for integrated graphics output, but if that's the fourth screen added, nothing. Looking at DP hubs/splitters or USB monitors as an alternative, but not knowing what's going on here makes me feel the same issue may arise, caused by something else I haven't figured out yet.
 
Well, I got it working, I think, but I'm not altogether sure why, which is frustrating.

I switched the DP version on the 2560x1440 second screen to DP 1.2 from 1.4. I connected that screen and one of the 1920 side screens to a DP 2x splitter and input those to the 3090ti port 2. The Samsung Neo G9 remains on DP 1.4, port 1, at 5440x1440 and 240hz. Those three all work normally, as of writing.

Then I connected the second 1920 side screen via a DP to USB-c adapter to the mobo's USB-c port, intending to use the 12900k's integrated graphics. It turned on immediately and appears to be working fine at 60hz.

I'm 99.99% sure I already tried exactly this setup and it didn't work, but now it does. Posting just in case it helps someone else out in the future. I remain perplexed.
 
Solution