qubitI didn't say it's got the exact same rendering engine. It's a development of the original one, so all the SLI/CF issues should have been ironed out. Also, multicard setups have been around for the last 15 years, so it's hardly something new and on top of that, the game's usual price is very expensive. Given all this, you'd expect SLI to work properly. There's no excuse for this poor show in my book.

i cant expect multigpu to work 'properly', or should i say 'max performantly', based on the way modern rendering is done or optimized & its complexity (such as deferred, layers get rendered out of order rather than objects that get lit in sequence, data has to be duplicated back & forth between gpus, frame timing has to be dealt with, so inefficient...), also there are the limitations of dx11 where devs cant make their own sli like dx12/vulkan

i dont know what low level optimization tricks are doable on pc, but when you look at what was done on later ps3 games, they started doing lighting on the cpu in parallel to other things on the gpu, at which point a theoretical sli would be impossible (i know it's a console, but gran turismo had experiments of using multiple ps3s to render multiple screens or corners of a high resolution screen, aka SFR rather than AFR)

nvidia removed more than 2 gpu & low end sli support in their cards, so that's not a good sign

still dont get why the companies havent attempted stitching gpus the way intel made its first quads, it would require more uncore? parts, but wouldnt it be cheaper & less of a hassle than making a new huge chip? the only thing close is cutting down big ones, a 960 appears to be half a 980 in specs if i remember

price shouldnt be a factor, in fact i'd expect cheap games to have more multigpu with the assumption they use older/simpler engines

i'd say there are plenty of grey area excuses, the question is how much we value the difference in rendering techniques or if we should even care when single gpus perform so well on their current games compared to the past

I think in the end they're just not bothered supporting SLI, because SLI is just not that popular, leading to a downward spiral of popularity and support. That low level SLI support removal you mentioned is very telling of this. @Mussels's thread is an example of yet another person who's given up on it due to problems. I abandoned it a few months ago when I bought my GTX 1080 and haven't looked back. Ironically in my case, the 780 Ti SLI ran out of memory before running out of GPU performance. I could easily max the VRAM in a recent Call of Duty game at 1080p and lose performance from just that,

When you say "stitching together" I think you might mean "ganged together"? This is an idea I had a few years ago and described on TPU. That it hasn't been done as a commercial product is perhaps due to cost and technical issues.

The idea is to design the GPUs so that they can sit next to each other on the same motherboard and be connected directly together, with the two working in tandem as one double-wide GPU. This would have a single pool of RAM too, so none of this wasting of memory and shuffling of data and microstutter nonsense. On top of that, the performance gain would be a perfect 2x just as if the two were a single GPU that's twice as big. In fact, there would be no need to switch off the connection between them as it would have no downsides.

Working fine here.

Tried to use GFE to do this update, crashed GFE for me for some odd reason when it started to setup for install. Tried it again and it worked fine...Not sure what caused it.

As far as the driver goes, works great for me other than that hiccup trying to install it.

qubitI think in the end they're just not bothered supporting SLI, because SLI is just not that popular, leading to a downward spiral of popularity and support. That low level SLI support removal you mentioned is very telling of this. @Mussels's thread is an example of yet another person who's given up on it due to problems. I abandoned it a few months ago when I bought my GTX 1080 and haven't looked back. Ironically in my case, the 780 Ti SLI ran out of memory before running out of GPU performance. I could easily max the VRAM in a recent Call of Duty game at 1080p and lose performance from just that,

When you say "stitching together" I think you might mean "ganged together"? This is an idea I had a few years ago and described on TPU. That it hasn't been done as a commercial product is perhaps due to cost and technical issues.

The idea is to design the GPUs so that they can sit next to each other on the same motherboard and be connected directly together, with the two working in tandem as one double-wide GPU. This would have a single pool of RAM too, so none of this wasting of memory and shuffling of data and microstutter nonsense. On top of that, the performance gain would be a perfect 2x just as if the two were a single GPU that's twice as big. In fact, there would be no need to switch off the connection between them as it would have no downsides.

It's more so SLI is in a tricky transition state currently. High end is pushing substantial increases in pixel density and AA which is currently cripplingly difficult with multi-card setups using the current DX11 APIs. 3+ cards will run you into godray and AA issues unless you dedicate a card to SLI AA. AFR is just a poor solution overall. SFR will fix a hell of a lot.

With DX11 transitioning to DX12, you also have limited options on the driver side to expand support yourself. You can no longer 'force' AFR magic tricks to make it work. Compound that with UE4's PFR reliance destroying AFR support as well, its a really tricky position to be in.

VR SLI is thriving, UE4 has phenomenal support (i.e. Serious Sam VR), but VR is in a stagnation until some real advances happen.

Dropping to 2 card SLI means Nvidia is able to expand to extreme resolutions via HB SLI bridge until NVLINK comes through the pipeline, but the cost of this is aggregating both bridges, which cripples 3/4 card setups. Given the market situation, it's an easy call on Nvidias part.

3/4 card support will inevitable continue to exist/return once NVLINK hits and DX12 SFR becomes standard as it's natively designed to scale unlike AFR/PFR.

Quick note about 3DMark, Firestrike and TimeSpy. And driver 376.33 scores.
=================================================

Nothing amazing but in the benchmarks above this driver consistently scores about 1.8 to 2% better (graphics tests). I have re-run many times as the gap is small, but found it consistent.

Prior to this I was using 373.06. I have years of old test scores lol!

EVGA GTX 1080 FTW (Both out the box and OC speeds multi tested).
I know you may think these increases are too low to warrant attention, but the point is I'm happy to get a few, just a few extra fps in games.

Could be system specific, I don't know, but due to repeated re-runs, and playing games the difference is positive, all be it small.

BTW - Deactivated all new "call home," tasks" via CCleaner. A total of five (nvupdatedaily/coreupdate, NvTmMon - 3 similar, and updater on logon).

I know two of them are needed if running for GFE, Shadow play or whatever, but my system is single gpu always clean install, always only physX and the driver itself.
NO negative side affects are many hours of PC use with all this crap disabled. Wouldn't mind if they announced it, and they are harmless, but just putting them there annoys me. Like MS loves to do.

Finally - Very good drivers (376.33) for me at least. (z77 based system. W10 64 anniversary, Pascal 1080)

EDIT: Verified Schd Tasks were disabled by using the Task Sheduler of windows 10, and checking
Task manager. Quite useful the little old free CCleaner!