View groups set up and performance

[kolaf]

Hi,

I have just completed building a new computer (i7 6700k, GTX 1070 together with an old 560 for some secondary monitors). To this I have connected three new HD monitors to the 1070 card, and I have three different monitors on the other card for additional display purposes (panels, maps, et cetera).

This is my first experience with running P3d spanning multiple monitors.

I read through the excellent tutorial posted in this forum. I did not manage to use the automatic grid set up, probably because my three monitors do not follow sequential indexes from left to right (they are labelled 4, 6, and 1 from left to right)? Perhaps a good idea would be a field listing the monitors in order that should be part of the grid instead of just the starting index? Or have I misunderstood something.

Anyway, I set up three views following the tutorial, and the result looked pretty good. However, the performance was horrible. Admittedly, I run pretty high settings (since I think I deserve it with the new hardware ;-)), and for the default scenario I easily reach 80-90 FPS. With the three displays this drops below 16 in some cases are fluctuates between this and 35.

Disliking this quite a lot I decided to try to set up Nvidia surround on the three monitors. I configured an approximate bezel correction, fired up the simulator, and stretched the display to span all three monitors. My performance was immediately almost as high as with a single monitor, mostly lying around 60 FPS. Of course, using this method I get somewhat of a fishbowl effect (stretching) near the outer edges, but I think that is preferable to the abysmal performance.

So, to the question. Is this significant drop in performance expected? I realise that P3D now has to render three different views at the same time instead of just one big one, but since the total resolution is identical, should this be that big of an issue? Have I perhaps done something wrong with the configuration?

I noted when running the view group set up but neither my CPU nor my GPU was running full tilt, both were around 50%, or maybe even slightly below that. I tried to reduce some settings such as AA from 8x to 2x, but this did not have any significant performance impact.

In summary, I can live with the Nvidia surround solution, but I would very much like to understand the view groups better, and if possible, switch to that solution.

Thank you for your attention.

[Beau Hollis]

View groups do not currently provide any performance benefit of manually opening multiple views unless views are distributed across machines using the new multichannel feature. Running multiple 3d views does come at a hefty performance cost when the system is CPU bound. Your primary core utilization is likely high. You won't generally see CPU usage stay consistently high on all cores, unless you fly so fast that background jobs are constantly loading terrain. Features like AA, volume fog, and HDR will likely perform about the same in 3 small views as they do in one large view since most of the work is per-pixel. Turning up scenery density or using a detailed VC with hundreds of draw calls will not scale as well across multiple views.

Thank you for your feedback. Improving performance is something that we will continue to focus on as the platform progresses.

[Beau Hollis]

You can change the DisplayIDs for each display so that they are sequential, and then you will be able to use the grid setup. We do try to auto-populate the list by attempting to sort all the monitors into a grid based on the windows desktop layout. The size and placement of your 6 monitors may be tripping up our sorting.

[markm75]

It was my understanding that the 1070+ cards corrected for the fish eye distortion on the left and right screens for 3 screen setups.. i guess this was not the case for you? (compared to say the 980ti ).. i was considering the jump from 980ti to 1070 just for this alone, but maybe not.

[MHargrove]

The Pascal-based cards do not automatically do any perspective correction -- it's a feature that application software has to invoke.