Optimal settings for ATI users with FSX - Great Results!

[Morph 9]

High AA settings effect performance in clouds. I had the 6950. Great card. I was able to push my weather harder when I went to 7950 while maintaining higher overall fps. So the card helps in some areas. But your right. The vast majority of FSX is CPU. You can drop the fps bringing that NGX into KLAX in a bad overcast situation. All the card in the world does not seem to fix those. Well. The 1000 dollar titan but that's it.

 

C.

haha yeah, I wonder when we will start seeing these threads popping up "Just bought new rig nvidia GTX TITAN, low fps!!!" and all the drama that it will cause.

[dazz 69]

I found this yesterday over at SimForum (thanks Peter for sharing)

 

http://www.simforums.com/forums/downsampling-and-fsx-jaggies-with-nvidia-card_topic45878.html

 

This may enable ATI users to have combined SSAA + MSAA (SS by downsampling, MS via CCC or RadeonPro) I tried it but couldn't get it to work on my T230H, but may be worth a try for you guys.

Besides the possibility to have combined SS + MS it might also allow for more SS granularity. With standard CCC settings you can only have 2x (1x2), 4x (2x2) or 8x SSAA, while downsampling you can do 1.25x1.25, 1.5x1.5 etc

[DaveCT2003 1,820]

 

 

But apparently those with NGX can tell.

 

 

Just an FYI here about the NGX.  Initially PMDG recommended TextureMaxLoad=4096 because of the would increase the graphics quality of the flight deck.  Then it was discovered that it only affected textures outside the aircraft (clouds, ground, buildings, etc). They tested at 1024 and there was no perceivable change in the aircraft.

 

Note also that running 1024 may/may not affect your frame rates depending on your computer, however for those that run combined scenery such as REX HD Textures, Manhattan X and FSFT JFK at the same time will suffer "TRUE" OOM errors as the 4096 increases the amount of memory consumed by 4096 textures is greater than that at 1046.  Kind of a no brainier there, but testing and monitoring memory usage proved this to be accurate.  I'll also say that the amount of additional memory used would likely be trivial under circumstances where that much high end scenery was loading all at the same it.

 

Frame Rate Stutters:  I haven't tested this myself, but there are some reports that 4096 resulted in mico-stutters for users of earlier model ATI and NVIDIA cards. Though I have not tested, I've not suffered this problem with a Radeon 5830 (note that this card's processor doesn't run much higher than 860MHZ but it has a higher number of graphics processors than some later model ATI cards, so some with later model ATI cards might still see micro-stutters at 4096.

 

 

 

Dave

[cvearl 60]

I found this yesterday over at SimForum (thanks Peter for sharing)http://www.simforums.com/forums/downsampling-and-fsx-jaggies-with-nvidia-card_topic45878.html

This may enable ATI users to have combined SSAA + MSAA (SS by downsampling, MS via CCC or RadeonPro) I tried it but couldn't get it to work on my T230H, but may be worth a try for you guys.

Besides the possibility to have combined SS + MS it might also allow for more SS granularity. With standard CCC settings you can only have 2x (1x2), 4x (2x2) or 8x SSAA, while downsampling you can do 1.25x1.25, 1.5x1.5 etc

Hey Dario. I'll try this. Would love a combined mode. Sold my 660ti last week. Got a 7970 GHz card. Got it cheap on a price beat. $397! Pretty happy. I am now able to run 8xSSAA most of the time now. At slightly steadier fps than my 660ti using 2xSGSS under stress tests. Except of course that wierd cloud test you sent me. At 8xSSAA, it was struggling at about 16fps. However, KSEA runway in JS41 with building storms theme I am in the 25+fps zone. My 660ti was more like 22. Tested them back and fourth a lot. Similar results landing KSFF in the JS41.

 

I'll try this other method when I get home. Camping this weekend.

 

C.

[cvearl 60]

Hey Dario. I'll try this. Would love a combined mode. Sold my 660ti last week. Got a 7970 GHz card. Got it cheap on a price beat. $397! Pretty happy. I am now able to run 8xSSAA most of the time now. At slightly steadier fps than my 660ti using 2xSGSS under stress tests. Except of course that wierd cloud test you sent me. At 8xSSAA, it was struggling at about 16fps. However, KSEA runway in JS41 with building storms theme I am in the 25+fps zone. My 660ti was more like 22. Tested them back and fourth a lot. Similar results landing KSFF in the JS41.

 

I'll try this other method when I get home. Camping this weekend.

 

C.

 

Sadly the current driver and downsample tool are not compatible with eachother. I am not willing to drop the current driver as it has performance enhancements for many games that I recently got stuck playing like Bioshock Infinite and Crysis3. When I am not flying that is.

 

Currently in FSX, the 7970 at 1Ghz flys 8xSSAA locked at 40 (still experimenting) for all but the worst weather or most dense scenery locales like KSEA or CYVR where I end up in the mid twenties but still smooth. I am waiting for the next gen AMD or a price drop on the 780so I will likely stick with this 6 to 12 months.

 

C.

[TechguyMaxC 516]

One quick observation about cloud performance on Nvidia: on my GTX 680 there is no performance drop (locked @ 30FPS via Inspector) when running in heavy overcast with unlimited draw distance and max. cloud density, @ 2560x1440 in addition to 4xSS and 4xSGSS via Inspector.  This is in DX10 mode.  In this scenario one of the primary bottlenecks is going to be GPU memory bandwidth.  When comparing AMD to Nvidia at the 7950 vs. 660ti price point, this is one area where AMD has a significant advantage due to their 384-bit memory interface vs. the 660ti's meager 192-bit interface.  Both cards run their memory at about the same speed (allowing for some small difference due to AIB partner factory overclocks) so this translates to the 7950 having approximately twice the memory bandwidth of the 660ti.  

 

Now, prices being what they are and performance across a wide range of 3d applications also being what they are, I believe it is totally fair to compare these cards at this point in time.  However, given the original prices for AMD's cards it is clear that they never meant for the 7950 to sell for a mere $299 like the 660ti.  In fact, when the 7950 launched it was selling for approximately $459 on average as seen here.  So what happens when we use an NV card with more memory bandwidth?  Exactly the results I've stated above - max'd out settings and super high AA levels that virtually eliminate all forms of aliasing, including specular (wave tops), texture (runway lines), and polygon (all objects including autogen).  

 

All that being said, I'm not here to rag on AMD cards, I think they offer excellent value and are now finally an option for FSX users that wish to have higher than normal levels of detail.  If anything, I think Nvidia's cards are over-priced.  The 660ti should not be a $299 card, it should be a $199-$249 card.  The 680 should be $299-$349 and the 670 shouldn't exist (or the 660ti should be the 670).  And of course now with the 7 series on the way the prices are going up at the high end and staying the same elsewhere, with performance increases ranging from small (say 10%) to huge (up to 50%).  I've got a 780 on the way to replace my 680.  I'm going to be water cooling the 780 just like I did with the 680.  I don't expect serious performance increases at the settings I run now (maybe 1-2 FPS), but I do expect to be able to enable 8x SGSS with no performance hit at 2560x1440.  We've seen what Titan can do for FSX (care of Word Not Allowed's blog) and the 780 will do the same (when overclocked).  

 

I'm not advocating everyone here go out and buy a $650 graphics card, it is absolutely overkill (and over-priced) for FSX especially.  Again though, with a 384-bit memory bus (compared to 256-bit on GTX 670/680) it ought to unlock that last level of performance that was just out of reach before due to this bottleneck.  I look forward to testing this myself over the next few days, once my 780 arrives (Tuesday).  I may re-load just to get a clean performance comparison, we'll see.  

[WotanUK 289]

 

 

This is in DX10 mode

DX10 is known to be a significant performance improvement in itself at the cost of compatibility.

 

My 7970 is running FSX at 5760x1080 (Eyefinity) at around 20fps using EGLL-xtreme.  20% MyTrafficX.

 

The CPU is more important for FSX, i gained an extra 10fps by overclocking (again) to 4.0Ghz, i gained very little moving from the 6970 to the 7970.

 

Running 4xSSAA.

[cvearl 60]

Roger that. DX10 mode in FSX is definitely better FPS with some sacrifices. Some of which have been addressed with some community made patches to get rid of things like flashing runways and the likes. I had tried DX10 mode briefly and I ran away. While ya I seemed to have crazy high FPS on my little 6950 2GB card, it was at the expense of decent levels of Anti-Aliasing. Now, to be honest, I did not try super hard but the best I was able to muster was the rough equivalent to 2xSSAA. Scouring threads on the subject from people that were using DX10 heavily and looking at their screenshots confirmed they were getting no better.

Quality wise I have tested ATI and nVidia somewhat extensively in the AA side of things. I have come to understand it goes like this. This is all DX9 BTW... I have no personal data on DX10 mode which I plan to try again soon.

Understand that DX10 AA and DX9 MSAA (Mulitsample) cannot fix gauges as they are transparent textures. This is why in Youtube videos you see some guy gets crazy high FPS but for some reason you cannot read his GPS fonts or they look like a can of glowing night crawlers wriggling about. That is MSAA and no fps hit. SSAA cleans up those textures and makes them sharp. See my previous screenshots.

As a side bar... Word Not Allowed guides from AVSIM discuss this as well. All his tests and guides are about DX9 and NOT DX10 for the simple reason that, as far as I know, DX10 cannot do more than 2xAA in FSX. The guages shimmer and the nVidia Inspector settings don't do anything. Now I only tried this briefly and maybe I did it wrong. If I did I apologize. But in the guides, the holy grail is elimination of shimmers from gauges and autogen objects. The common nVidia setting for this is 8xS which is combined AA mode 4xMS+2xSSAA on edges (nice) and 2xSSAA on transparencies like Gauges (Also nice). Wish ATI did combined mode. This combined mode looks slightly sharper on nVidia versus ATI 4xSSAA (Which is really nice still and miles ahead of AA in DX10 mode or basic MSAA at any level). 8xS with 2xSGSS on nVidia performance hit is similar to 4xSSAA on ATI except that nVidia can take the cloud density slider to MAX and ATI needs one notch lower at HIGH for the same performance (again somewhat negligible IMO as I rarely fly in deep overcast). This is due to how ATI tries to apply AA to the clouds I think. I went back to ATI simply because I wanted a 7970 for a few other games (like Dirt racing game) and also for additional test I want to try in DX10 mode now that I see ATI's tool (Radeon Pro) has MLAA and MSAA and FSAA post processing effects that can be injected. If they clean up gauges and external model paint lines like 4xSSAA, then I might be a DX10 convert. Anyway here are the AA modes in DX9 in order of worst to best.

No AA (barf)
ATI and nVidia any MSAA sample level alone (wriggly guages and shimmery Autogen)
ATI 2xSSAA
nVidia 4xS with 2xSGSS
ATI 4xSSAA
nVidia 8xS with 2xSGSS
ATI 8xSSAA (with heavy overcast, fps tank into the teens)
nVidia 8xS with 4xSGSS (Even the Titan struggles)

NOTE: last two settings listed above are a tie quality wise in my experience.
 

Google Word Not Allowed's Titan test. See chart for FSX AA Test Clouds... Titan gets 20 fps with 4xSGSS enabled. I tested 8xSSAA with my 7970 and I get about the same result as the GTX580 in that chart using his test (he supplies these if you want to try) and I get bad stutter. 8xSSAA with full cloud cover is not doable with high Weather Sliders.

Since the 6950 card, I cannot live with less than 4xSSAA. If I owned a 680 I would use 8xS with 2xSGSS as my go to AA setting for all occasions like a few of my friends use. But even highly overclocked 680's and 4.6 Ghz CPU's, in DX9 mode.... Sitting on runway 6L at KSEA in the PMDG JS41 (heavier fps plane than the NGX) with the Building Storms theme in FSX weather, they average 24 fps and this is on a single monitor 1920x1080. My 4.2 GHz 2600k with a 1Ghz 7970 3GB card gets the same using 4xSSAA (not quite as sharp as 8xS 2xSGSS but close).

Moral of the story? DX9 in FSX with high AA mode is HEAVY even on the higher end systems even with only a single screen. Period.

Now I look forward to some time off work soon. I want to use some of that time off to test DX10 mode again with the new RadeonPro tool with injected AA modes not available in DX9 and see what is the max sharpness I can render. If it is as good as 4xSSAA and doubles my framerates in all weather? It might be the new FSX grail. But I am not holding out much hope. Nope I think the new Grail is GTX780 using 8xS and 2xSGSS with max quality AF set in inspector.

Charles.

[TechguyMaxC 516]

I hate to say it but Word Not Allowed's doing something wrong if he's only getting 20 FPS @ 4x SGSS with a Titan, because I can do that now with my 680 @ night in the NGX sitting on the runway @ FSDT KJFK looking towards the terminal and Manhattan X with UT2 @ > 50% AI traffic.  He says only 14 FPS for a 680 but the absolute lowest I see in that scenario is 20 (22 is the average).  

 

Edit: I just downloaded and ran his clouds AA test file - ok, now I see why the FPS are so low, in fact, this is one situation where DX10 is actually a detriment to performance (at this time) becuase AA is applied to the clouds as well which is completely unnecessary since they're not solid polygonal objects but are actually volumetric.  I'll do some testing with this again later when I reload my O.S. & FSX for Titan.  

[cvearl 60]

I hate to say it but Word Not Allowed's doing something wrong if he's only getting 20 FPS @ 4x SGSS with a Titan, because I can do that now with my 680 @ night in the NGX sitting on the runway @ FSDT KJFK looking towards the terminal and Manhattan X with UT2 @ > 50% AI traffic. He says only 14 FPS for a 680 but the absolute lowest I see in that scenario is 20 (22 is the average).

 

Edit: I just downloaded and ran his clouds AA test file - ok, now I see why the FPS are so low, in fact, this is one situation where DX10 is actually a detriment to performance (at this time) becuase AA is applied to the clouds as well which is completely unnecessary since they're not solid polygonal objects but are actually volumetric. I'll do some testing with this again later when I reload my O.S. & FSX for Titan.

Go over to his blog and have a chat. He likes talking about this kind of thing.

 

And I'm sorry but I don't want to burst your bubble but you cannot enable SGSS in DX10 mode with FSX.

 

but I do not doubt that you're getting some good results are you able to show any screenshots anywhere do you have any posted somewhere around here?

 

C.

[TechguyMaxC 516]

Go over to his blog and have a chat. He likes talking about this kind of thing.

 

And I'm sorry but I don't want to burst your bubble but you cannot enable SGSS in DX10 mode with FSX.

 

but I do not doubt that you're getting some good results are you able to show any screenshots anywhere do you have any posted somewhere around here?

 

C.

 

So the anti-aliasing I'm seeing applied to the entire scene including autogen, waves, clouds, and textures isn't SGSS?  

 

I usually fly in photoreal areas now, here's a recent screenshot I posted:

 

[cvearl 60]

 

 

So the anti-aliasing I'm seeing applied to the entire scene including autogen, waves, clouds, and textures isn't SGSS?

I usually fly in photoreal areas now, here's a recent screenshot I posted:

This is definately getting my attention. Can I get a shot of a panel please? Preferably a Carenado plane with a Garmin GPS in it? The ground in this shot being photoreal is irrelavent to the conversation with regards to AA as is discussions surrounding framerates. You have to be over some dense autogen in ORBX scenery as that is 90% of what this discussion is about. ORBX autogen with cloudy scenarios with high AA modes. But ya the paint job lines definately look like SGSS applied to me so that is exciting. But Gauges are most important to me as I spend 95% of the flight sitting in the plane. I need to see a shot of the VC.

What would really get me excited is a shot from the VC looking out the front window at KSEA Runway 6L in any one of any of these planes.... Carenado B200, C90 or RA Duke (Turbine or Piston) or PMDG JS41 (NGX is lighter than these) at KSEA (ORBX only no photoreal) with Very dense autogen with Building Storms Theme set from within FSX with the garmins ON.

Again, this looks promising. It is starting to look like I need to revisit DX10!!! Can you also send a screenshot of your nVidia Inspector settings for FSX please? If you have SGSS working in DX10 mode in FSX with really high FPS whilst surrounded by clouds and mist (building storms) in the scnario I describe above with 30+ fps, I am a DX10 convert and I am buying me a GTX670 FTW real soon. Or a 680.

Thanks!

C.

[TechguyMaxC 516]

Here's the requested shot.  FPS in this scenario was between 24-30.  I haven't gotten around to flying the Duke yet so 1 Garmin is active but I haven't gone through a startup sequence.  

 

[cvearl 60]

 

Here's the requested shot.  FPS in this scenario was between 24-30.  I haven't gotten around to flying the Duke yet so 1 Garmin is active but I haven't gone through a startup sequence.  

 

 

 

If you have FSX set to DX10 Preview, dude... You are blowing my mind right now. What zoom is that please?

 

C.

[TechguyMaxC 516]

DX10 is most assuredly on. Zoom level is default. I'll post my Inspector settings for you in a bit.