SLI framerate tests

[GSalden 4,679]

For my home cockpit , based on the Ifly 737 NGX , I wanted to see if I could use an SLI setup to get better fps.

 

Some tests , done about a year ago and posted here at Avsim , showed that ,when using the external view, the fps with almost all tested aircraft went up by 30-90%.

 

For my home cockpit I used 2 pc's :

- pc 1 : K2600@4.8 Ghz / 8 gb 2133 mhz / 4x SSD / 2x GTX680 SLI -- W7 / FSX / Ifly server ( no VC )

- pc 2 : I-860@3.6 Ghz / 4 gb 1899 mhz / GTX470 -- W7 / Ifly client

 

Setting : All sliders maxed , full overcast, except AI 30% / car traffic 0% / server clocked back to 2600 mhz

 

 

Test 1 --

- 2d cockpitview without any gauges visible / Aerosoft MAP Schiphol , view at Gate D

- Single GTX680 - 19.1 fps

- SLI GTX680. - 26.1 fps

 

Test 2 --

- Spotplaneview / Aerosoft MAP Schiphol, view at Gate D

- Single GTX680 - 10.1 fps

- SLI GTX680. - 17.3 fps

 

Test 3 --

- Spotplane view / City view Amsterdam at 2500 ft with Schiphol in the background

- Single GTX680 - 9.9 fps

- SLI GTX680. - 13.8 fps

 

 

This clearly shows that SLI can work in FSX when no gauges are shown and the VC is off.

[N1125Y 22]

in SLI how maney Monitors can be connected at one time ?

Did you run the test in Full Screen or win mode ?

 

Cheers

 

Yair

[GSalden 4,679]

Test were done with 3 32" hd tv's, in 5850x1080 resolution and in fullscreen using Nvidia 2d surround.

 

Two tv's connected to graphicscard 1 ( dvi + hdmi ) and the third tv to graphicscard 2 ( hdmi ).

[ArjenVdv 190]

You can get way higher FPS with that system. At least in the mid 20's in the NGX in your situation (without SLI). You running an FPS limiter? Are you running BP=0?

 

Interesting find btw. It might be that it just helps with multiple monitors...

[SAAB340 28]

- pc 1 : K2600@4.8 Ghz / 8 gb 2133 mhz / 4x SSD / 2x GTX680 SLI -- W7 / FSX / Ifly server ( no VC )

 

Yes, there are scenarios where SLI is helpful in FSX.

 

Did you have both GPUs plugged in but SLI not activated in the control panel when you did the non SLI test?

 

The reason I'm asking is because the 2600K will only provide 8x PCIe 2.0 lanes to each card, and that will be the limiting factor before you hit the CPU limit in many scenarios if you use BP=0 and in most scenarios if you use bufferpools. That's the case for 1920x1200 at least.

[virtuali 4,421]

5850x1080 resolution and in fullscreen using Nvidia 2d surround.

That's why you are getting a visible benefit using SLI: at such high resolution, you are limited by fill rate and SLI will always give you more fill rate in any case. It's likely that with a single monitor at the standard 1920x1080 resolution, the benefit wouldn't be nearly as large.

 

Single monitor users will also get improvement from using SLI, if they like to use very high quality antialiasing settings. Since antialiasing is another thing that consumes a lot of fill rate, two cards in SLI will help a lot. Someone not using anti-aliasing (or using cheaper / lesser quality modes) might not see much improvement either.

[GSalden 4,679]

You can get way higher FPS with that system. At least in the mid 20's in the NGX in your situation (without SLI). You running an FPS limiter? Are you running BP=0?

 

Interesting find btw. It might be that it just helps with multiple monitors...

Like I wrote, I ran the tests with the cpu clocked at 2.6 Ghz to see what SLI was capable of .

 

At 4.8 Ghz my fps at Schiphol with the high settings showed a framerate of 27-38.

Where with my single Gtx680 is was 15-22.

And I added a lot of houses, farms and trees for the Schiphol area myself...

 

As I wrote testes were done with a single GTX680 vs 2x GTX680 in SLI.

 

Many OC testes with the 2600K ( pci 2 ) vs 3770 ( pci 3) showed that the 2600K was higher overclockable and the fps where higher.

My mobo also supports pci 3 and when using a pci 3 slot the data transfer rate is higher.

 

Regarding running a graphics card in pci 2 vs pci 3 gives no performance gain... 

 

 

 

[SAAB340 28]

My mobo also supports pci 3 and when using a pci 3 slot the data transfer rate is higher.

 

Regarding running a graphics card in pci 2 vs pci 3 gives no performance gain...

 

The 2600K does not support PCIe3.0. You need all 3 of the CPU+MOBO+GPU to support PCIe3.0 otherwise you won't get it.

 

Is that table based on FSX?

Because FSX will easily saturate x8 PCIe 2.0 fully and loose a lot of FPS when it does.

[GSalden 4,679]

Quote from Tomshardware.co.uk :

A Pci-e 2.0 slot is more than capable of handleing the GTX 680 and will not restrain it at all.

There are currently no cards out that will use up all the bandwith of a Pci-e 2.0 slot.

I don't know for certian why the Pci-e 3.0 was relaesed now when there are no cards that can take advantage of the doubling of the bandwith but I guess in the near future there will be so maybe it's best to get the technology out there and working so when the cards are released that will take advantage of it the technolgy will be completely stable and ready to go.

 

Every modern mobo , even 2 years old has 16x pci2...

[jabloomf1230 2,445]

This clearly shows that SLI can work in FSX when no gauges are shown and the VC is off.

 

 

SLI does improve performance for both FSX and P3d, despite popular opinion that it doesn't. I initially had one GTX 580 and then added a second in SLI and the performance gains that I observed were similar (percent-wise) as to what you posted. But if an app is designed properly to work with SLI, the performance gains should be somewhere more than noticeable, but obviously less than double. Neither app is within that range. I should add here that the two GTX 580s didn't even come close to my present GTX 680.

 

So the choice of going to SLI for either app must be made on the basis of cost effectiveness and that yardstick is not met. According to posts of the official LM forums, P3d 2.0 also will not be optimized for SLI, but a later release will.

[GSalden 4,679]

SLI does improve performance for both FSX and P3d, despite popular opinion that it doesn't. I initially had one GTX 580 and then added a second in SLI and the performance gains that I observed were similar (percent-wise) as to what you posted. But if an app is designed properly to work with SLI, the performance gains should be somewhere more than noticeable, but obviously less than double. Neither app is within that range. I should add here that the two GTX 580s didn't even come close to my present GTX 680.

 

So the choice of going to SLI for either app must be made on the basis of cost effectiveness and that yardstick is not met. According to posts of the official LM forums, P3d 2.0 also will not be optimized for SLI, but a later release will.

Lucky me I have a flightsimfriend who has almost the same hardware as I have.

So I borrowed his card for the tests.

 

Now I have bought his Asus GTX680 Top II CU graphics card and he has bought an EVGA GTX780 Classified.

 

After the release of P3D v2 I first will wait to see how that developes itself in the flightsim world.

Changes are that my system can handle it with ease now that with v2 some load should be shifted from the cpu to the gpu.

[ArjenVdv 190]

I see the PCI multiplier influences performance. Where in the BIOS do I set my PCI multiplier and how do I know it's set optimally?

[GSalden 4,679]

What mobo do you have ?

[SAAB340 28]

It's good that it's shown when SLI helps in FSX. SLI isn't broken as such, it's just many scenarios when it doesn't help, or even makes things worse.

Quote from Tomshardware.co.uk :
A Pci-e 2.0 slot is more than capable of handleing the GTX 680 and will not restrain it at all.
There are currently no cards out that will use up all the bandwith of a Pci-e 2.0 slot.
I don't know for certian why the Pci-e 3.0 was relaesed now when there are no cards that can take advantage of the doubling of the bandwith but I guess in the near future there will be so maybe it's best to get the technology out there and working so when the cards are released that will take advantage of it the technolgy will be completely stable and ready to go.

Every modern mobo , even 2 years old has 16x pci2...

I've posted a topic looking at PCIe performance in FSX now.

http://forum.avsim.net/topic/425500-this-is-when-pcie-30-matters/

 

This is one FPS graph from that topic

 

It's very clear that FSX is able to fully use up the bandwidth x16 PCIe 2.0 has to offer. Using BP=0 helps a lot though.

That's why I was wondering if you used x8 or x16 with the single 680 results, and if you used BP=0?

[GSalden 4,679]

Very nicely done !

 

I tried the 3770 ( pci 3 ) from my other pc in my FSX server and did my EHAM and Amsterdam framerate tests again.

Running it at 4.3 Ghz ( max it would run ) gave me 6% less fps than the 2600k at 4.8 Ghz...

BP=0

Fps unlimited

 

Most people run FSX with BP=0 , fps=30 and Nvi with refreshrate at 1/2.

When running it like that pci2 and pci3 would perform the same according to your tests.