Anyway, if you're getting that frame rate, it's not going to go any lower by reducing settings, so either the AA modes are not working, or those GTX680 are so much faster than Fermi

Hi Dario,

I only just discovered your benchmark, and I am away at the moment so I cant help with any testing but just want to say that Hasses results compared to your results just confirms that the GPU benchmark is fully shader bound.

 

GTX680 has 1536 shaders running @ core clock. GTX580 has 512 shaders running @ 2 x core clock.

 

The "shader power" difference between your benchmarks are:

(1536*1110)/(512*2*900)=1.85

 

The FPS difference between your benchmarks are:

31.867/17.613=1.81

 

Almost perfect scaling!

 

4x Sparse Grid Supersampling is obviously very shader heavy in clouds and the GTX680 has a lot more shading power to offer. Seems like Kepler is a good buy for 4x SGSS...

Hi Dario,

I only just discovered your benchmark, and I am away at the moment so I cant help with any testing but just want to say that Hasses results compared to your results just confirms that the GPU benchmark is fully shader bound.

 

GTX680 has 1536 shaders running @ core clock. GTX580 has 512 shaders running @ 2 x core clock.

 

The "shader power" difference between your benchmarks are:

(1536*1110)/(512*2*900)=1.85

 

The FPS difference between your benchmarks are:

31.867/17.613=1.81

 

Almost perfect scaling!

 

4x Sparse Grid Supersampling is obviously very shader heavy in clouds and the GTX680 has a lot more shading power to offer. Seems like Kepler is a good buy for 4x SGSS...

 

yeah man, great analytic thinking of yours there once again. that makes complete sense.

I may get me a 670, would be great to have a few other regulars chip in *wink wink* Word Not Allowed *wink wink* Stephen *wink wink* Techguy, HLJames, Alain... :P

I wonder if dual GPU can be of benefit in this case? Anyone with a 590 or 690 or any one running SLI on x58, x79 or z77 with ivy?

I try a liitle OC now or do you want someting else, a do what you want

Hasse,

 

You've clearly shown that the test is shader bound and that the GTX680 is a lot faster than an overclocked GTX580 in this benchmark.

 

Dario has showed that the test is not affected by memorybandwidth on the 580. Your 680 does have a lot more shader power but it does not have any more memory bandwidth.

 

It would be interesting to see if lower memorybandwidth starts to be an issue with this increased shader to memory bandwidth ratio. Would you mind downclocking your memory and test?

 

As for the GTX670 vs GTX680. I would expect to see around 17% difference at stock speed at this benchmark.

 

And comparing a GTX580 @ 900Mhz vs a stock GTX670 I would expect to see around 55% difference.

 

I may get me a 670, would be great to have a few other regulars chip in *wink wink* Word Not Allowed *wink wink* Stephen *wink wink* Techguy, HLJames, Alain... :P

 

You forgot to add me!

You forgot to add me!

 

i did, sorry about that Ben. also Corey, even though he's not so active anymore. where the hell are you Corey BTW?

Lars is a regular now too, a great techie contributor... there's TechGuy too, the list goes on and on... so take 5 minutes and run the test guys! even if it's to show that it's worthless

If we could get a couple of consistent results on the same GPU that would be something

So it's time to recap here, and define a set of FSX and Inspector settings for the test:

Is this really what we should be using?

 

Am I correct in saying that your result that I'm quoting below uses these settings but WideViewAspect=True (True or False doesn't seem to do much in this case actually)

 

In that case we have no results posted using these "official" settings

This is my score with Scenery & Traffic down to minimal

 

Frames, Time (ms), Min, Max, Avg

6652, 300000, 20, 28, 22.173

 

I've ran the benchmark using the above settings and I'm getting numbers I don't think look particularly GPU limited.

My stock GTX470 (448 shaders @ 2*607Mhz) gives this result:

 

Frames, Time (ms), Min, Max, Avg

5915, 300000, 17, 23, 19.717

 

I'm sure that the GTX580 @900Mhz is more than 12% faster in everything GPU related compared to a stock GTX470.

 

 

But I have also run the test using the same settings you and Hasse used for your previous results. (FSXmark11 but at 1920x1080x32)

And they make perfect sense:

 

First GTX470 @ stock:

Frames, Time (ms), Min, Max, Avg

3494, 300000, 10, 13, 11.647

 

Followed by GTX470 @ 810Mhz core clock

Frames, Time (ms), Min, Max, Avg

4272, 300000, 11, 16, 14.240

 

Here you can see that it doesn't scale particularly well with the core clock on my GTX470. But I know since previously that FSX will try its hardest to maintain at least 10 FPS and the GTX470 @stock is really stuggeling to do that.

 

But with it overclocked to 810Mhz on the core I don't think FSX is making any uncontrollable sacrifices to try to not go below 10FPS.

 

So if we use this 810Mhz number and compare it to your and Hasses previous numbers we get the following:

 

GTX580 @900 Vs GTX470 @810

Shader difference

(512*2*900)/(448*2*810)=1.27

 

The FPS difference

17.613/14.240=1.24

 

Good scaling with the shaders.

 

 

GTX680 @1110 Vs GTX470 @810

Shader difference

(1536*1110)/(448*2*810)=2.35

The FPS difference

31.867/14.240=2.24

 

Also good scaling with the shaders.

 

 

GTX680 @1110 Vs GTX470 @810

Shader difference

(1536*1110)/(448*2*810)=2.35

The FPS difference

31.867/14.240=2.24

 

 

This is quite an eye opener! I would've never guessed that the shader count would scale so well!

I am really glad to see that the new Kepler cards are performing so well in high AA scenarios.

 

Thanks so much Saab and Dazz for all of the testing. It is extremely helpful for me to see these numbers before I make any solid purchasing decisions.

Westman, did you test the GTX 670 in the end? I'm interested because currently my only performance issue with FSX is with high cloud cover and rain dropping my FPS below what I deem acceptable; probably a combination of my high resolution and AA (although not running high levels of AA, it isn't needed with my pixel density). So at the moment my only concern is to remedy the FPS in this situation - and if the shader scaling is also true for the GTX 670 and GTX 680, it may be worth getting the GTX 680.

 

Thanks.

Westman, did you test the GTX 670 in the end? I'm interested because currently my only performance issue with FSX is with high cloud cover and rain dropping my FPS below what I deem acceptable; probably a combination of my high resolution and AA (although not running high levels of AA, it isn't needed with my pixel density). So at the moment my only concern is to remedy the FPS in this situation - and if the shader scaling is also true for the GTX 670 and GTX 680, it may be worth getting the GTX 680.

 

Thanks.

What card do you have at the moment and what resolution do you run. High SGSS is obviously very shader heavy. That does however not have to be the case using other modes. I do remember doing tests quite a while ago with my old 8800GTS512. With that card you could adjust the core, memory and shader clock independently. An that card was easily the limiting factor using only in game AA and 1920x1200. But in that case it was not responding anything to shader clock. Only to core and memory clock. That made me believe it was the ROPs that were limiting then.

What card do you have at the moment and what resolution do you run. High SGSS is obviously very shader heavy. That does however not have to be the case using other modes. I do remember doing tests quite a while ago with my old 8800GTS512. With that card you could adjust the core, memory and shader clock independently. An that card was easily the limiting factor using only in game AA and 1920x1200. But in that case it was not responding anything to shader clock. Only to core and memory clock. That made me believe it was the ROPs that were limiting then.

 

GTX 560 Ti 1GB. I did have a GTX 580, but it had to be RMA'd. I run at 2560x1440 and use 4xS and 2x SGSS.

Hi

Today i run a couple of tests 670 vs 680 it

is no extreeme overclocking no fiddeling with voltages.

 

PC: 3770K @4.8 ram 2666mhz

 

Cards:

EVGA 670 OC 4gb

KFA2 680 OC 4gb

 

EVGA 670 OC 4gb Factory deafault.

Frames Time (ms) Min Max Avg

7675 300000 23 30 25.583

 

EVGA 670 OC 4gb Nvida default 670. (915-980)

Frames Time (ms) Min Max Avg

7358 300000 22 29 24.527

 

EVGA 670 OC 4gb 1006-1086 (Nvidia deafault 680)

Frames Time (ms) Min Max Avg

7773 300000 23 31 25.910

 

EVGA 670 OC 4gb 1186-1166 6600

Frames Time (ms) Min Max Avg

8465 300000 25 33 28.217

 

 

KFA2 680 OC 4gb Factory deafault.

Frames Time (ms) Min Max Avg

9486 300000 29 36 31.620

 

KFA2 680 OC 4gb Nvida default 680. (1006-1070)

Frames Time (ms) Min Max Avg

8729 300000 26 33 29.097

 

KFA2 680 OC 4gb 1184-1250 6600

Frames Time (ms) Min Max Avg

10019 300000 30 38 33.397

 

KFA2 680 OC 4gb 1266-1330 6600

Frames Time (ms) Min Max Avg

10562 300000 32 40 35.207

 

Kind Regards Hasse

 

 

 

 

 

 

 

 

Hmm, seeing as my only performance issue at the moment is with heavily overcast clouds (and rain) - I'm wondering if I should be going for the 670/680 over the 580.

Please, can à 690 owner do à test on factory deafault clocks.

This is not CPU limited you get the same at 3.5 as 5.0ghz .

Its not à competition

 

Or 580 ,570,680 SLI

 

Kind Regards / Hasse

Frames Time (ms) Min Max Avg

6909 300000 20 27 23.03

 

GTX580 @ 850MHZ

 

Running FSX default 1024 Clouds