TR Forums

I've been wondering about this for awhile now - as in: will it work?

There are several monitors that have this feature: picture by picture.  This is what it looks like when it's a "full screen" feature:

http://s7d2.scene7.com/is/image/Samsung ... nefit-jpg$


-what I'm wondering is: has anyone used this to "get around" the limitation of high resolution and mid-range card*s* (..in a more capable fashion than crossfire), particularly in a Freesync setup?

Ex. 2 RX 470's each driving one side of a UHD Monitor, or 1920 x 1440?



This is one of the more intriguing monitors that keep me thinking about this:

http://www.ebay.com/itm/QNIX-UHD3216R-R ... 7012&rt=nc

-though of course this would be 1920 x 2160.

From what I gather of your post, you're wanting one GPU dedicated to each side of the screen to increase performance of gaming across the entire display?

There are several logistical barriers that make it rather inefficient. For smooth frame times, you want each GPU to be doing roughly the same amount of work. Partitioning the display half and half doesn't mean that the GPU workload is divided evenly in half. Say an explosion happens on one side of the screen and one GPU will have to do all the extra work to render it while the other side continues on. nVidia actually did this before in the DX9 days with an SLI technique called split frame rendering (SFR) but dynamically adjusted the areas each GPU would be wholly responsible for rendering. The dynamic part of how the image was split lead to rather good scaling but it had some compatibility problems with some titles. Alternate frame rendering (AFR) won out as it had greater compatibility and worked with DX10 and DX11.

The other logistical issue is that splitting a displays in half logically requires the drivers to unify them virtually for an application. nVidia Surround or AMD's Eyefinity are necessary to expose the higher resolution of the full screen. Some games do no play nice with Surround/Eyefinity and there is a slight overhead vs. a single display of similar resolution (i.e. one 3840 x 2160 display gets slightly higher frame rates than four 1920 x 1080 monitors in Surround/Eyefinity).

Lastly is the monitors own internal logic for displaying images may not play nice. There is single panel that is being used and it may require each side to update synchronously. Thus if the frame for the left side arrives slightly earlier than the right side, the left side's frame is slightly delayed to render together with the right. This could cause pacing issues and/or image tearing. Worst case is that the display incorporates its own frame buffer to do PBP or PIP and thus everything is delayed on screen by one frame which is not optimal for gaming.

Best practice right now is to simply run the displays at its native resolution and then left SLI/Crossfire do AFR for load balancing. Split frame rendering can make a return with DX12 and Vulkan but that is entirely up to the game developer.

..the question persists: will it work?

I don't "buy into" the best practice argument (..at least not until someone has tried this out).

Sure, there will be games that won't like eyefinity, not because of eyefinity per se but rather because of the resolution - which would be a problem with the native high res. of the monitor regardless.

I do know that eyefinity works with freesync, what I'm wondering is will the monitor that supports freesync actually passthrough freesync to two separate displayport inputs simultaneously in a picture by picture mode?  Also, IF that works - will it in fact produce better performance than a crossfire solution?

..anyone?

I think eyefinity just renders to a single large virtual screen which it then sends sections of to the monitors. All the monitors need to be connected to the main card for eyefinity to work at all so it's crossfire or nothing to use any cards. It's not like running your desktop over several monitors on several cards. http://support.amd.com/en-us/search/faq/20

Eyefinity is a neat little trick but not actually that complicated and doesn't have anything to do with how the the rendering gets done so I don't think you can use it to get more performance.

On the freesync side of things I assume all the monitors in an eyefinity screen get set to the same refresh rate so it could work as the two halves of the pbp screen would always be in sync. 

The other problem with SFR is that many algos like SSAO look at other nearby pixels to figure out what a pixel looks like. If that's not accounted for, there will be a big discontinuity right down the middle of the screen. Sharing that data between cards on the fly is tricky for numerous reasons, so usually SFR setups have each card render more than half the screen (55-60%?) and do a post-process blend in the middle. Eyefinity doesn't usually have reason to handle such things, because the discontinuities in the rendering coincide with discontinuities between monitors anyway, so I'm not optimistic about this. (That's if Eyefinity even splits work that way in the first place, which isn't a thing I've heard of.)

AFR is better for framerates, but can have various problems with consistency (a smooth frametime graph doesn't guarantee a good experience), and it adds an extra frame of latency regardless. SFR isn't good for such big framerate boosts, but the end result is all but indistinguishable from that of a single more powerful card, so I do prefer it.

I use picture by picture on my 34" 3440-1440 panel when i watch tv " TV side has bars" and surf the web but it squeezes the PC screen down to fit so it looks a little funky, I don't know if you ran your game at 1720x1440 if it would look more natural or if it would even use half the screen properly. GL

..yeah, that's not a full screen picture by picture.

-several manufacturers do that, Phillips among them.

cheesyking wrote:

I think eyefinity just renders to a single large virtual screen which it then sends sections of to the monitors. 

That I didn't know (..I thank you), and interestingly - when I looked closer at eyefinity, it seems that it's primary goal is for offset adjustments to cope with bezels between monitors - something that Windows alone doesn't do.  ..and at least as far as eyefinity goes: that's a definitive no-go or crossfire only (when using more than one card). 

I'm assuming Nvidia surround has the same limitation? (..it seems likely.)

I remember Matrox being "big" into this (..a long time ago in a galaxy far, far away), I wonder if they pay Matrox royalties or just ripped them off?


What about Windows itself?  

CScottG wrote:

I remember Matrox being "big" into this (..a long time ago in a galaxy far, far away), I wonder if they pay Matrox royalties or just ripped them off?

Matrox was big in multi monitor support because it was a very niche area where they could sell specialised hardware after their 3D performance got left behind by ATI and nVidia, not really because they had any special sauce. Personally I wouldn't say anyone could rip them off or need to pay royalties because the basic idea of rendering a large virtual display and then splitting it off to multiple monitors is "bleeding obvious". I don't know how windows handles multi monitors but a single large display technique is how a lot of linux distros work. In fact on Linux some games where a dev ignored advice and didn't use an existing library like SDL can end up doing accidental eyefinity by rendering a single image across all monitors.

The bezel adjustment might possibly be considered special sauce but that doesn't seem like it would be that hard to do either.