Gtx 1080 TI and VR performance

If you also want to add more stress on your system, [for more bullet proof settings] you can test at sunset or sunrise. Check your azimuth to make sure you have longest shadows.

Sent from my Nexus 5X using Tapatalk

 

The more stuff you have on the screen, the more draw calls (DIP in AC's render stats) are generated. Simply put, this is the CPU preparing the scene for the GPU to render. If you ever hear of rendering being CPU-limited, the likelihood is that it can't process this stuff quickly enough to keep the GPU busy. Being serious sim business, AC differs slightly in this respect in that the physics can bog down the CPU as well, particularly with lots of AI.

Why this is bad for VR is because regular DX11 means the scene has to be generated independently for each eye, which doubles the number of draw calls. I'm not sure whether this means AC is actually CPU bound in VR. It could be that no matter how fast your processor is there's simply too much data going back and forth between CPU and GPU for the PCI bus to cope. Haven't seen that tested anywhere, but it would explain why VR drops frames even when neither the graphics card or processor appear to be stressed.

This is why I mentioned throwing money at graphics hardware is slightly futile when it looks like the bottleneck is elsewhere. Sure, you get a slight increase in image quality, but the potential performance benefit a Ti supposedly has over a regular 1080 is seriously compromised in VR.

 

Any kind soul can do a TLDR for me? I own a 980 (non Ti) and the 1080Ti is tempting, but I am still unsure if it's worth the upgrade or not.

My CPU is a 6700K@4600.

Thanks!

Enviado desde mi Nexus 6P mediante Tapatalk

 

You can't have a better gpu than the GTX 1080Ti, it's fast, silent, not that hot. For VR it is more than you need until you want to max out all parameters but it is still ridiculously good.

 

Are you using VR or a flat screen display?

 

Sorry! Thought it was obvious given the thread topic. Yes, I am Oculus Rift user. Maybe I should stop slacking and put my specs in the signature.

Right now I am quite happy with my current settings, but you always can do with more bells and whistles.

Enviado desde mi Nexus 6P mediante Tapatalk

 

I went from a 1080 to a 1080TI and noticed a significant improvement. I think you would see a big jump in overall performance.

I would expect that in most cases you have a pretty decent display with what you run but that most of the time you are in ASW meaning that many (up to 50% depending on conditions of the frames) are synthesized not real frames. The 1080TI would change that until most of the frames would be real frames and few would be synthesized. This impacts the smoothness of the display and in coordination of control inputs, and head movements with what is happening on screen.

 

I may be in the minority here, but I'm contemplating upgrading to this card; however, I am waiting towards May's end to see what AMD reveals with Vega. Not that it will provide a better VR experience, but more of seeing if Nvidia is forced to reduced prices. I believe the 1080 & TI are overpriced at the moment. All speculations aside of course.

Sent from my Nexus 5X using Tapatalk

 

Just got my gigabyte aorus 1080ti oc....came from a 980 ti....WOW...worth the upgrade..oh yea...the 980ti wasnt terrible...but now..everything on full( I dont use pp) and pd at 1.8...runs butter smooth...no hiccups..

Another good test was Areofly fs flight sim...able to bump settings to high and it looks amazing...rins smooth

 

CV1 users, what about the cpu, does an upgrade to an i7 from an i5 make the sim w/ CV1 run better? Couldn't find a link (aside from Oculus' min specs) saying how the cpu is utilized.

 

guys i have made a tiny test for my pleasure which aimed to find how many mixed class opponents can hold my system on full Nords while driving remains comfort
so at first i have set 59 opponents (including me there were 60 cars on track) and got a constant 99% cpu occupancy. However i could play with the main issue - due this 99%, the ffb became weak (on my wheel and motion platform) and the overall impression was like i'm moving through water, i mean like the time stretches so it was looking like a lil bit slooow motion.
ok, i reduced opponents to 47-53 - much better, but 99% still popped up and while it happened the expression was like i described above.
and finally, the amount ~43 opponents seems play well enough for me and my devices
please note, that my system is not oc at all, all stock as in sig, cpu is usually around 70% while i play, fps on grid ~45 (so no pure fps obviously but it was assumed anyway)
at first i wanted to mix all gte cars two pieces each, all gt3 cars two pieces each (including some mods) and gt4/cup cars two pieces each. but as you see, i had to reduce some of them in favour to playability - some pairs of gte, gt3 and gt4/cup has gone, additionally on this video there is no any m235i - cause i remember, guys, that someone was talking it is some kind of a hog, so i excluded it somehow from the test here. in future i will find out does it actually has any impact on me
no fcat graphics, diagrams and such, sorry - not so much time to collect and analyze all of it, besides i was testing for an immersive experience (which i most interested in) and not a tech scope

my graphics settings are like this:
AF: 16x
AA: 4x
World: max
Shadow: max
Smoke: low
Show smoke in mirrors: on
PP quality: ultra
Glare: ultra
Dof: ultra
Saturation: 90
Blur: max
Crepuscular rays, Heat shimmering, FAAA: on
Mirror resolution: high (custom 2048 + wider fov)
HQ mirror: off
Reflection quality: medium
Reflection rendering: medium

as you can see my cpu is a bottleneck now (being at stock though) but i'm not in despair cause obviously the game is not intendedd to run such amount of opponents in VR, besides my system can hold all 59 opponents in non-VR mode flawlessly
so i believe the best amount of opponents for me should be around 40 (my dream is to play with ~200 to get a proper Shcleife traffic but that is indeed a groundless dream)

before i did this test i was perfectly playing 30 opponents on full Schleife so maybe now i will add some more for fun

 

I played a few hours with 1.14 (Windows 10 1607) and while the CPU occupancy is much higher than before (I hit 95% where I was 65%), the frame rate is better/smoother than before.
Maybe core usage is more balanced now, but better AIs and better frame rate : win win !

 

I decided to sell my 1070 and sell 4 shares of Nvidia stock to upgrade to this 1080TI. I tried your settings in MP mode, but I did lower the smoke to low and bumped up the PD to 1.6(1.4 was to blurry for my taste). After the grid gets dispersed I was only using 50-60%GPU. Did you tune yours for 20 or 24 AI players? I'm not sure if participants have agreed on a standardized benchmarking track and car combo. Everyone seems to be using something different. Thanks.

Sent from my Nexus 5X using Tapatalk

 

something is bottlenecking with lots of cars in view, gpu and cpu never reach 100% so it must be a software problem not hardware

i have noticed cars with better optimized lods do perform better though

 

Scroll up and read post #202.

 

nvidia smp (or similar techs in future) should do wonders, it only renders a scene once instead of having to render it twice

 

It certainly should, yes. Unfortunately I can appreciate why it might not be particularly high on Stefano's to-do list.

It would be interesting to know why PP is so damn expensive in VR and whether that could be improved. Obvious all round benefit rather than just those with particular hardware.

 

i heard somewhere that pcars2 might use smp, dunno how true that is though

it would def be interesting to see to difference in vr between smp and normal

 

Only a few people have reported on actual data based results.

I have used the Lotus Exige S at Imola and run my tests for the first 60 secs of the race. I sought conditions that would put a significant but not overwhelming load on the system so I could discriminate between variables. I found that 20 total (19 AI and 1 driven), starting 20 and staying 20 gave that type of load.

Since I last reported results two things happened in my system.

1. AC was upgraded to 1.14
2. Windows was "upgraded" to Win 10 version 1703.

I tried to prevent the automatic upgrade of Win10 because of the reports by K_Soze regarding poor performance. Somehow it was installed without my agreement.

I ran a test of the new conditions. The combination of the two changes created a slight loss in performance. Average from 4 runs before and after the changes.

Before Win 1703 & AC 1.14 -- 17% synthesized frames
After Win 1703 & AC 1.14 -- 21% synthesized frames

I really wished I could have measured these changes independently but I think from the reports of K_Soze we can conclude that AC 1.14 is an improvement and Win 1703 degrades performance.

The performance of my system with anything but very full AI fields is 100% so I have been spending my time enjoying the 1.14 upgrades and not doing any further testing.

 

It's hard to say what limits the performance. Measuring performance of the system under high load is difficult as one needs to know what the instantaneous load is on the gpu and each of the core/threads in the cpu during a complete frame cycle. Measuring in a precise manner creates load on the system and interferes with the performance (sort of a variant on the Heisenburg Uncertainty Principal). Tools probably exist for this type of thing but who has them and how to use them is beyond my knowledge.

The actual limits have to be reached in the hardware when some software cycles can't run fast enough to complete it's work in time to write a new frame. The bottleneck exists in the hardware but its genesis starts in the software. We just don't know where it is actually occurring and I would be surprised if anyone actually does.

The AC physics/graphics engine is instrumental in producing high level performance. In my experience with various sims AC is steps ahead of most of its competitors but work to stream line the engine for VR would be greatly appreciated.