I am trying to reduce a current installation of 7 PCs networked together to run a simulator application running on a video wall of 6x4 1080p displays.
Each machine runs 4 displays. One machine acts as the simulator and sends data to the 6 graphics PCs to render and output to the video wall.
Task is to reduce this to a single (or max 2) PCs. I think we can probably get sufficient performance for rendering the unity application on a 4080/4090 or A series equivalent however the bigger issue seems to get getting the video to 24 displays due to the max 4 display limit per GPU.
I hoped to use the MST Feature of the displays to essentially run 6x 4K displays (6 physical connections over a couple of GPUs with each of the 6 displays daisy chained to 3 others) but apparently this isn’t possible despite the overall resolution being the same? Frustratingly I can’t see why it is any different but seems to be a software limitation?
It looks like NVIDIA used to make an 8 port card (NVS 810) but this is now discontinued without a replacement?
Any ideas? Outputting to a video wall processor seems to be around 3x the price of just running 7 mid-spec machines. We want to optimise our setup but can’t justify that premium.
Happy to run multiple GPUs if needed, possibly combined with some external hardware but PCI bandwidth (and physical slot count) may become a limitation if it’s more than 3-4cards.
Any ideas?
Thanks
Hello and welcome @james350 to the NVIDIA developer forums.
At the moment there are no current gen GPUs with more than 4 Display outputs I am afraid. So that will be one limiting factor.
MST and DP daisy-chaining adds additional limitations with respect to bandwidth, 1 main and 3 daisy-chained Displays will not necessarily support the maximum resolution the GPU can provide.
I think you might need to go with 6 GPUs in total, which would make up for 2 PCs. Unless you have synchronization figured out you might also require a Quadro Sync card.
Thanks @MarkusHoHo , I have literally just been looking at this as you responded. Would the following work?
-2x suitably powered Quadro GPUs, potentially joined with a Quadro Sync 2 card
-Each GPU outputting a 4K signal to a splitter like the DataPath Fx4 (so PC sees 6x connected 4K displays and doesnt breach the 4 display limit)
-The splitters each converting the 4K signal out to 4x 1080P displays
-Use Mosaic to create the 6x4K display video wall (and possibly deal with sync timing using the datapath boxes?
As far as I can tell from the specs it should work, yes.
You should probably verify with DataPath first if your use-case is covered by their HW. You should ask if you can do some trial run first with GPUs you already own, just to be on the safe side.
And if you manage to try it out I would be very curious to hear back from you with your success story.
Thanks!
Sorry, late to the game, but the above is the path to go, if you can’t go with more pixels per screen/projector…
If you need 24 outs, you have to have 6 GPUs! (NVS810 was actually 2x GPUs on a single PCB, single PCIe slot…)
Splitters will take a high res input, show as a single sink to the GPU, so we can drive 4 of them per GPU [our limit is in HW, 4 scanout units per each GPU only…].
What happens behind the splitter, we don’t see, don’t care. To sync the 6 high res outs to the splitter, it needs a sync card. Splitters need to maintain stable latency when buffering the frames to split them up into multiple smaller outputs, otherwise they will break the sync. Only disadvantage is the (typically 1 frame) added latency the splitter must add, because he needs to process on frame level, can’t work on line or even pixel level…