Summary:  Whatever you do - DON'T BUY A GTX-295 for NON-SLI APPLICATIONS!

Note: WQHL 196.21 with SLI patch

These are 1920 X 1200 tests starting with 16 X AA and 16X AF.  They only scale down because 295 was performing so poorly.

Rise of Flight 1.009 Retail Release

ROF-Reltek audio

Effects On - MultiSampling Texture Filtering
Frames, Time (ms), Min, Max, Avg
  7411,    179139,  28,  52, 41.370 18:15
  7397,    179099,  28,  52, 41.301 18:19
  7389,    179070,  28,  52, 41.263 18:23
  7356,    178803,  27,  52, 41.140 18:28

Effects Off - Multisampling
Frames, Time (ms), Min, Max, Avg
  7330,    179128,  26,  56, 40.920
  7280,    178848,  27,  56, 40.705
  7315,    179121,  26,  56, 40.838
  7258,    178894,  26,  56, 40.572

Effects On - SuperSampling Texture Filtering
Frames, Time (ms), Min, Max, Avg
  7436,    179183,  27,  52, 41.499

Effects Off - Supersampling
Frames, Time (ms), Min, Max, Avg
  4221,    179076,  17,  28, 23.571
  4216,    178840,  17,  28, 23.574

GTX 295 - Effects Off - 16X x 16X - High Quality - Texture Filtering:OFF *can't select - ROF Terrain rendering : high - Supersampling
Frames, Time (ms), Min, Max, Avg
  3204,    188589,  10,  23, 16.989

Next change to Multisampling
Frames, Time (ms), Min, Max, Avg
  4347,    186424,  13,  38, 23.318
  4365,    186807,  13,  39, 23.366

Textures to Quality
Frames, Time (ms), Min, Max, Avg
  4391,    187443,  13,  39, 23.426

ROF Landscape to Medium
Frames, Time (ms), Min, Max, Avg
  4410,    188116,  13,  41, 23.443
  4367,    186754,  13,  39, 23.384

Multi-GPU to Single
Frames, Time (ms), Min, Max, Avg
  5020,    179321,  17,  43, 27.994

AA to 8X AF to 8X
Frames, Time (ms), Min, Max, Avg
  5247,    179900,  17,  47, 29.166

IL2-46 : UltraPack 2.0  

GTX 280
Nvidea
Water=2 - Supersampling 16XQ AA 16 AF sample opt on
Frames, Time (ms), Min, Max, Avg
 11656,    150031,  36, 132, 77.691
 11665,    150117,  36, 131, 77.706

Water=4 - Supersampling 16XQ AA 16 AF Sample opt on
Frames, Time (ms), Min, Max, Avg
 10274,    149781,  31, 123, 68.593
 10327,    150473,  31, 124, 68.630

Water=2 - Supersampling 16X AA 16 AF sample opt on
Frames, Time (ms), Min, Max, Avg
 13885,    149933,  36, 210, 92.608

GTX 295 :


Notes - can't have High qual textures and Texture compression optimization (trilinear optim - or Aniso sample opt)


- Multi-GPU mode = recommended Display/Mixed GPU - Single Display
Water=4 - Supersampling 16XQ 16XAF sample opt off *not allowed
Frames, Time (ms), Min, Max, Avg
  7976,    150140,  17, 121, 53.124

Multi-GPU mode = -alternate frame mode 2 - Display/Mixed GPU - Compatiblity
Frames, Time (ms), Min, Max, Avg
  7492,    149739,  20,  91, 50.034

Multi GPU mode = Nvidea recommended
Water=4 - Supersampling 16xQ 16AF
Frames, Time (ms), Min, Max, Avg
  7072,    150534,  20,  91, 46.979

Multi-GPU mode = Nvid recommended - Display/Mixed GPU - Single display
Water=4 = Supersampling 16X AA 16X AF
Frames, Time (ms), Min, Max, Avg
  8093,    150118,  19, 126, 53.911


Multi-GPU mode = Nvid recommended - Display/Mixed GPU - Single display
Water=2 = Supersampling 16X AA 16X AF TEXTURE Filtering QUALITY = High Quality
Frames, Time (ms), Min, Max, Avg
  9182,    149981,  23, 141, 61.221
  9224,    149982,  23, 145, 61.501

ATI compare test2 (must be repeated on GTX 280)..Texture filtering qual = Quality (all prior tests were high quality) - Aniso sample and Trilinear filtering are ON
Multi-GPU mode = Nvid recommended - Display/Mixed GPU - Single display
Water=2 = Supersampling 16X AA 16X AF -
Frames, Time (ms), Min, Max, Avg
  9229,    150132,  23, 146, 61.473
  9224,    150030,  23, 144, 61.481
  9226,    150067,  23, 144, 61.479

S~

Gunny

I'm going to give Bevo back the 295 this morning to see if he can replicate in his system.

Couple of things that bother me in the testing:

When in SLI mode - time counted in fraps is skewed.  I was very careful to start and stop fraps at the points indicated in the tracks:

a. Start Benchmark and Stop Benchmark Subtitles in Rise of Flight
b. At 00:03 seconds and 02:33 seconds in Il2

Can't for the life of me understand how the times shifts (millisecond counter) on the 295 between Single GPU and Multi-GPU modes.

There are many posts over at ROF forums of folks getting rid of there 295's for other cards - I can see why now.

At least you can gain an understanding of their frustartion by these numbers. I have total bandwidth detected by GPU-Z for both cards which I'll post.   GTX-295 MUST be in SLI to have better performance than a 280/285.

S~

Gunny

So, if your a simulations-first kind of guy - your top card from Nvidea will be the GTX-285 or a used GTX-280.

GTX 280 (EVGA)
(https://webspace.utexas.edu/joem/computer%20hardware/BIG%20Benchmark%20tests/GTX-280GPUZ-pg1.jpg)

GTX 295 (BFG)
(https://webspace.utexas.edu/joem/computer%20hardware/BIG%20Benchmark%20tests/GTX295-pg1-GPUZ.jpg)

CPU (Q9650 4.0 GHz)
(https://webspace.utexas.edu/joem/computer%20hardware/BIG%20Benchmark%20tests/GTX-280GPUZ-pg3%2BEverestCPUID.jpg)

S~


Gunny

Yes  very interesting Gunny thanks a lot for the great job.. like i was saying in your other post Il2 and RoF perform poorly on Alternate 1 and 2 SLI .
They have to be Splited.. i never seen yet something good from alternate 2

Hey Gozr.  Yes I tired alternate 1 and alternate 2 with no good results.

Eco predicted this over at SIMHQ or ROF forums a long time ago - but I can't find the thread to link.....

I can say this, if you need SLI = 2 X 260 or 2 X 280/5 is the way.  Not this way :)

S~

Gunny

I don't mean to be a bore, but this card will be here - at my house either today or tommorow:

(https://webspace.utexas.edu/joem/computer%20hardware/BIG%20Benchmark%20tests/3000_01.gif)

S~

Gunny

Summary of GPU-Z output:

Card         Bandwidth Shaders ROPs  Pixel Fillrate (GP)  Texture Fillrate   GPU Clock  VRam Clock  VRAM Config
GTX-295   112 GB/s    240      28       16.1                    36.9                576           1008         2 X 896 MB DDR3
GTX-280   141.7        240      32       19.3                    44.1                602           1107         1 X 1024 MB DDR3
5870 VPX  160          1600     32       69.6                    69.6                870           1250         1 X 1024 MB DDR5

To be fair - the reading here is 1/2 of the POTENTIAL capability of a GTX-295.  However this potential can only be realized when BOTH sides of the card come together in SLI.  No single board SLI for ROF or IL2.  It's possible by reverting to 190.38 or 182 - 184 series that something can be made to work - however, it's beyond me at this time.

I did try the EVGA SLI patch which is dependent upon drivers 196.21.  No joy.

EDIT:  Added VRAM configuration to table.  It was Gozr that first told me to ensure I had 1024 MB CONTIGUOUS VRAM.   He was right!

S~

Gunny

EVGA SLI patch is bollox it's just modifying some profiles that about it.

I don't waste my time @ SIMHQ..
I like the GDDR5 on ATI.

Let me know when you receive the ATI :)

Since i'm not running XP/Vista or Windows 7 as my main OS i discovered quite a while back  that the patches are just profiles something that is easy to do. But most of people wont touch this and just get the patches..

PS: My 280's ( SSC version ) are 650 Mhz from stock, they are set on cruise @ 670 mhz.. i have no reason for OC those more for Flight sims.


ATI 5870 256-bit memory interface
GTX 280 512-bit memory interface

I've been completely unimpressed with the 295 on a whole. My 285 is a stout performer and hasn't missed a lick since day one.

I've been watching the latest from ATI and they really look solid, but not enough (from everything I've seen) to warrant an upgrade. It was my intention to wait until Spring and look closely at Fermi thinking it would have already been released and the second set of (more solid) drivers would be out. Right now, it looks like it may be barely released by then.

I guess I'll still hide in the bushes and watch.

Hi TB :)

Your 285 is one of the two best cards in ROF.  I can tell you that.  The 285 and 5870 are so close, for ROF - I'm not sure you could really tell the difference between the two if you did not know.

As I try other applications, I might find something.   But not yet.

S~

Gunny

One important thing I think I've discovered in all of this "Graphics board testing".... is the Nvidea to ATI vocabularly.

Nvidea AA - ATI Equivalent
8X AA  =   8X AA "Box Filter" Samples: 8X
8XS AA =  8X AA "Edge Detect" Samples:12X
16X AA  = 8X AA "Wide-Tent" Samples:16X
16XS AA = 8X AA "Edge-detect" Samples:24X

For AA MODE - this is a very critical setting for both Nvidea and ATI, in that it's a huge hit to frames - but minimizes (NVidea) or elimantes (ATI) the terrain "shimmering" effect that Eco first found and documented - and everyone else has been unappy with.
Nvidea            -       ATI Equiv
Mutisampling         Multisampling
Supersampling       Adaptive Multisampling
No NVidea Equiv -  Supersampling

S~

Gunny

This is of course, one of the two main reasons that ROF Nvidea users are switching to 5870 and 5850, in numbers.

Second reason is bad experiences with GTX-295.

S~

Gunny

You may want to try some custom drivers for Nvidia that are based on getting a better quality or performance.. but the better quality is well worth the try from ForceWare Omega, DNA-Force or NGO NVIDIA Optimized Driver..

OK here some news..
NVIDIA Surround is coming and getting good. !!  Yeas !!

http://www.hardocp.com/article/2010/01/08/nvidia_3d_surround_multidisplay_gaming_editorial/ (http://www.hardocp.com/article/2010/01/08/nvidia_3d_surround_multidisplay_gaming_editorial/)

http://www.hardocp.com/news/2009/10/13/eyefinity_update (http://www.hardocp.com/news/2009/10/13/eyefinity_update)

Something i was telling about vertical views for Sims