NVIDIA 670 2GB for 3 screens

Hi all,

I'm currently using a 6870 for the 3 screens I run at different resolutions (keep buying new screens over the years). Due to the need for CUDA, I'm looking to upgrade to a 670 however I'm concerned that it being 2GB, I'll run out of VRAM once I upgrade all the monitors to 1920 x 1080.

Is anyone running a 670 with 3 screens and what are you views?

Thanks!

 

I have a pair of the 4GB 680's across 3x 1920x1200 + 150 bezel correction total and if I max out the settings and up the AA I can easily trip over the 2GB mark. If I turn off Multiview, or turn off HDR and down the AA it drops to between 900MB and 1700MB depending on the track.

I am a IQ gannet though, I don't know what your priorities are, but for the extra few pounds/dollars/bucks/yen I'd go for 4GB versions of the 670 or 680 for 3 screens AND high detail.

 

Mr pix

Great setup, what sorts of fps are you getting at these settings? (also mention track no. Of AI's etc )

Also what CPU/MHz are you running..

 

@Demondriver. I just recently bought a 670 to replace my 6870 CF. I also use 3 monitors with softth. It made a huge difference with my system. I've got AMD Phenom 4 @ 3,6 Ghz and 8 gig ram. First of al anti-aliasing made me want to go back to nvidia. Ati can't coop with AA. Always had to use Supersampling eating my preformance like candy to have decent graphics. CF isn't well suported so mainly I used 1 GPU. I can run rF2 on high settings with 20+ cars on 3 screens. For sure I experience some jaggies on specific parts of a track but you notice only when you are paying attention to it. And even then it's hard to notice. I think this is just the ISI engine and isn't because your system is lacking. If you want to know more just PM me.

 

My i7 920 CPU runs at 4.3Ghz.

With the following visual setup:
- all display settings maxxed out
- FXAA OFF
- AA @ level 3
- Cockpit Mirrors ON
- Virtual Mirrors OFF

Race setup
- Megane @ Mills
- 23 opponents (30 visible)
- start position 12 (midfield)
- 14:00hrs
- Clear Weather
- Cockpit Mirrors ON
- Virtual Mirrors OFF
- single lap
- AI control

FRAPS benchmark running from last red light to finish line

Multiview ON, HDR OFF
Min FPS = 47
Max FPS = 77
Average = 61


Multiview OFF, HDR ON:
Min FPS = 63
Max FPS = 80
Average = 71

Multiview ON, HDR ON:
Min FPS = 25
Max FPS = 30
Average = 27

I choose to run Multiview ON, HDR OFF and it's as smooth as butter, but I do turn on Adaptive vsync and Triple Buffer in the NVCP (have no idea if triple buffer makes a difference with adaptive vsync).

Hope this helps.

 

Yes it will, I would recommend getting the EVGA 4GB Superclocked card as you will need the 4GB VRAM for Surround View

 

I run 24" triples on a reference 670 2GB with no problems, everything turned up full 20 cars on track That is before optimization...

 

yes he do , look :http://www.gamecritics.com/triple-buffering-improve-your-pc-gaming-performance-for-free

 

Hmmm.... or in my case... no he don't!

Thanks, a very interesting, if not 5 years old, article.

The article you pointed me too is referring to vsync, not adaptive vsync.

The article states that Triple buffering only comes into play below the refresh rate of the monitor... adaptive vsync turns off below the refresh rate of your monitor, thus it will therefore have no effect.

It also states that triple buffering only works on Open GL games and requires a 3rd party app to be used (D3DOverrider) to make it work with DirectX games (such as this one).

...and furthermore, triple buffering does not work with SLI/Crossfire (that well) due to the way the frames are split.

So five years ago when the article was written, triple buffering could not have helped me.

.... but you HAVE helped me, I'm going turn it off as it's pointless me using it. Thank you. I will however continue to use Adaptive vSync as it seems to keep things smooth.

 

OUPS!

 

Interesting,, I ran your test with a single 680 and got a similar result.
Phenom II X4 965 3.4GHz
Gigabyte GF GTX 680 OC EDITION

All settings the same as yours including 150 bezel correction, except I have triple buffer turned off.
Multiview ON, HDR ON:
min 23
max 49
avg 32

 

For sli you have to do two things.
Forst AFR2 by driver or nvidia inspector.
Disable both reflection maps in the graphics settings of rf2.

Right now the reflection destroy the sli benefit, you can measure gpu load with and without those both reflection maps.
There is a huge performance difference between.

I am using everything maxed out, except of both reflection maps
1920*1080 @ 120 HZ, 4XSGSSAA, 16XAF @ ~180FPS, but i am using adaptive vsync @ 120 FPS.

Triple buffer should be disabled
max pre rendered frames should be set to 1

My spec
Intel Core I7 Ivy Bridge, 3770k @ 4500 Mhz
16GB DDR3 - 2133 Mhz
2x GTX 680 SLI 2 GB
2x IIyama 27" 120 HZ
1x LG 30" 60Hz
1x Samsung 26" 60 HZ

not playing in multi monitor, i prefer single monitor and 120 fps.

 

Thats better results than mine makes me feel bad.

Going to try your recommendations now... 2 questions though...

1. Why force to AFR2 when rFactor 2 has a SLI profile?
2. Doesn't it look like a 1990's game with reflections off?

 

There is a profile in the newest beta drivers, nv is using afr2 too, with it, i just want to keep sure, that you re using the correct sli mode. Disable SLI AA as well.
Without reflection its not as nice as with it, but the performance gain is impressive.
I have no idea how much the devs can improve this behaviour, maybe someone of them will response by theirself.

 

Thanks, yes using the latest beta driver for these cards.

Just finished testing and logging the results. I see no improvement with Multiview on with your recommendations, maybe a frame or two per second either way to be fair... but with Multiview OFF I see approx 15%-20% improvement overall with reflections off, but it was very playable over 60fps with them on, so no real benefit for me on this test.

Results show original FPS then with refelections off in brackets:


Multiview ON, HDR OFF
Min FPS = 47 (40)
Max FPS = 77 (83)
Average = 61 (59)


Multiview OFF, HDR ON:
Min FPS = 63 (71)
Max FPS = 80 (104)
Average = 71 (88)

Multiview ON, HDR ON:
Min FPS = 25 (25)
Max FPS = 30 (31)
Average = 27 (28)

Since running the original test, I have setled on HDR ON and Multiview OFF, FXAA ON, ingame AA OFF and am happier with the playability on all tracks.

Multiview seems broken at the moment as it breaks SLI (GPUs dip to approx 60%).

 

I think you also control the curiosity with some fanatec wheel and problaby you gonna be arrested for using this to play rFactor2 without permission

 

Haha, no just G27, which i prefer over the fanatec, except of the volume the ffb creates....

@MrPix: Hmm, maybe i should show you my nvidia Inspector-Profile?
For me there is a huge gap in between.

BTW: You should not use fxaa, which costs a lot of performance.
Use 4x Multisample + 4xSGSSAA + 16X HQAF and a negative LOD of -0,750 for good image quality.
Dont use any sync from rf, just hdr on, your resolution, hz, no vsync.

 

Thanks for the advice, .... but you have a resolution of 1920*1080 = 2073600 pixels, a little over 2MP, my resolution is 5910x1200 = 7092000, a little over 7MP, so our performance results should be very different and to top that, you are not testing Multiview at all as this requires multiple screens.

I would be interested in trying your nvinspector profile if you can share it, just to see if there is a big difference.... to be honest, I doubt there will be.

 

I know that your resolution is a lot higher .
Maybe you re running out of mem?

NvInspectorFile:
View attachment rf2.7z

 

I'm no where near running out of memory, less than 2GB in use of the 4GB available (my cards are the 4GB versions).

Thanks for the profile, I'll test it later and post the results when I get back, just off to a mates.