GTX-200 series (GTX-280) exhaustive testing for ROF

[TX-Gunslinger]

ROF 1920 X 1200 fullscreen
Landscape/Grass/Trees/Shadows/Reflection/Textures all HIGH
Landscape Mesh Quality = HIGH
Forest Distance = Far
Sound counts = 64
Interface sound checked
Sound quality = High

NVIDEA @ stock clocks
8X AA
8X AF
AA Gama correction = On
AA Mode = Override Application
AA Setting = 8X
AA Transparancy = Multisampling
Max pre-rendered frame = 3
Single display performance mode
Texture filter - AF sample Optimization = OFF
Texture filter - Negative LOD Bias = Clamp
Texture filter - Quality = High Quality
Texture filger - Trilinear Optimization = OFF
Threaded optimization = auto
Triple buffering = Off
Vertical sync = off

        CPU
Test    GHZ     Frames, Time (ms), Min, Max, Avg
01   3.40     7248,    179115,  25,  59, 40.466
02   4.05     7817,    179018,  26,  63, 43.666

Max pre-rendered frames set from 3 to 0
03   4.05     7658,    178951,  27,  62, 42.794

Max pre-render = 0 changed from High Quality to Performance Textures
04   4.05     7754,    179191,  27,  64, 43.272

AA Transparency - Multi-sampling to Super-sampling (HUGE!)
05   4.05     5072,    179141,  20,  36, 28.313

Now with AA Transparency OFF
06      4.05     7848,    179100,  27,  66, 43.819

One sanity check with AA Trans back to Multisampling [should match test 04]
07     4.05     8000,    179369,  27,  66, 44.601 (test04/0 low/+2 high/+1 avg)

2nd sanity check - Multisampling - Textures = High Quality [should match test 03]
08   4.05     7632,    179264,  26,  64, 42.574 (test03/-1 low/+2 high/0 avg)

OK we are sane.....within the margin of error

Now let's finish with some AA and AF iterations from the Nvidea control panel
All tests to this point have been in 8X AA and 8X AF with "Overide application setting"

Test 09 - AA to 8xQ
09   4.05   7176,    179084,  25,  59, 40.071 (test08/-1 low/-5 high/-2 avg)

AA to 16 [near identical to 8XAA] [On 280 if you run 8 then run 16]
10   4.05     7677,    179776,  28,  63, 42.703  (test08/+2 low/-1 high/0 avg)

AA to 16Q
11   4.05   6997,    179038,  25,  54, 39.081 (test08/-1 low/-10 fps high/- 3.4 avg)

AA to 4X
12   4.05      8208,    179043,  28,  70, 45.844 (test08/+2 low/+6 high/+2 avg)

AF to 16X [test12 is new baseline:previous tests used 8X AF]
13   4.05   7954,    179176,  28,  67, 44.392 (test12/0 low/-3 high/-1.5 avg)

AF to 4X with AA at 4X [this says just keep AF at 8X]
14      4.05    8189,    179072,  28,  71, 45.730 ...(test12/0 low/1 high/0 avg)

Current best baseline @ 1920 X 1200
4X AA 8X AF Multi-sampling - High Quality - Texture Filtering OFF

15 Texture filtering AF changed to ON (with Trilinear OPt OFF)[test 15 is new baseline]
15   4.05   7998,    179230,  27,  68, 44.624 (test12/-1 low/-2 high/-1 avg)

16 Texture filtering AF to ON - Trilinear Opt ON
16   4.05     8316,    179131,  29,  71, 46.424 (test15/+2 low/+2 high/+2 avg)

17 Once more at 8X AA [test 16 is baseline now]
17   4.05   7773,    179104,  27,  65, 43.399 (test16/-2 low/-6 high/-3 avg)avg)

18 Last at 16X AA[Wow - again 16X AA performs very well compared to 8X]
18   4.05   7714,    179249,  28,  62, 43.035 (test16/-1 low -9 high - 3.4 avg)

19 16 X AA + 16 X AF
19   4.05     7607,    179148,  27,  58, 42.462 (test16/-2 low -13 high - 4 avg)

Nvidea settings tests hold at 19 baseline.

Go on to ROF graphics and audio settings at test 19 Nvidea settings.

S~ Gunny

[TX-Gunslinger]

I continued testing and tweaking today.  In the previous post I believe I developed a good ROF baseline to start examining the impact of Videocard overclocking and then finally ROF graphics and sound settings.  Here's the starting baseline (not exactly the final numbers I'm sure I'll use but close).

ROF 1920 X 1200 fullscreen
Landscape/Grass/Trees/Shadows/Reflection/Textures all HIGH
Landscape Mesh Quality = HIGH
Forest Distance = Far
Sound counts = 64
Interface sound checked
Sound quality = High

NVIDEA @ stock clocks vanilla GTX-280 - Core 602 Shader 1296 Memory 1107 Fan 40%
16X AA
16X AF
AA Gama correction = On
AA Mode = Override Application
AA Setting = 16X
AA Transparancy = Multisampling
Max pre-rendered frame = 0
Single display performance mode
Texture filter - AF sample Optimization = ON
Texture filter - Negative LOD Bias = Clamp
Texture filter - Quality = High Quality
Texture filger - Trilinear Optimization = ON
Threaded optimization = auto
Triple buffering = Off
Vertical sync = off

[TX-Gunslinger]

Based on the new Nvidia baseline above - I re-sampled my recorded performance track:

4096 MBytes DDR2 PC2-6400 (400 MHz) Corsair CM2X2048-8500C5D
3000.8 MHz Bus 333.4 Rated FSB 1333.6 DRAM 400.1 Core 1.072V
                                                 End   Temp
Frames, Time (ms), Min, Max, Avg    Test  Time      C1 C2 C3 C4 GPU CPULOAD%
(Temps Idle)                                         11:51:45,36,36,33,34,47,2.0
  6633,    179487,   23,  54, 36.955 11:58
  6545,    179048,   23,  51, 36.554 12:02
  6534,    178812,   23,  53, 36.541 12:06
  6572,    179120,   22,  53, 36.690 12:13  12:13:20,43,42,40,40,75,76.3

Core 2 Quad Q9650 4005.1 MHz Bus 444.9 Rated FSB 1780.1 DRAM 534 Core 1.296V
GTX 280 Stock GPU Core 602 SHADER 1296 Memory 1107 Fan 40%

Frames, Time (ms), Min, Max, Avg    Test    Time     C1 C2 C3 C4 GPU CPULOAD%
(Very low load)                                      12:44:25,37,37,34,34,47,5.0
  7557,    179123,  27,  59, 42.189 12:50   12:49:50,50,47,44,45,69,75.1
  7452,    178920,  26,  59, 41.650 12:54   12:54:20,51,48,48,46,76,76.9
  7412,    178501,  27,  57, 41.524 12:58   12:58:00,52,48,47,46,77,74.9
  7461,    179170,  27,  60, 41.642 13:05   13:05:00,50,48,44,47,74,75.1
  7444,    178976,  27,  59, 41.592 13:10   13:10:10,51,48,45,47,75,75.2

[TX-Gunslinger]

The results between a stock clock @ 3.00 GHz and the 4.00 GHz Overclocked tests make you wonder why you spend so much time tweaking

+4 Min +6 Max +5 Avg is not to sneeze at.  This is what you should expect going in with all the trouble an CPU/RAM/Mobo Overclock will take.

Now let's see what we get out of a GTX-280 overclock.   Yes, for all of those that saw me post my Graphics problems the other day - we'll find out next if I fixed them....  If not - the OC should not go very far.

Again, I have a vanilla EVGA GTX-280 - GPU rev A2/Bios ver 62.00.1C.00.80 1GB GDDR3.  I did not buy the more expensive SSC (Superclocked) version nor the much more expensive FTW "factory selected" versions.

S~

Gunny

[TX-Gunslinger]

Wow, learned a lot here - namely that the Graphics Card overclock does not add a whole lot.  I have no idea why a simmer would blow any change on the "factory hand-selected graphics boards".

I was able to succesfully OC my 280 to FTW specs!  Saved some change there

My 602 MHz Core is now at 691 MHz - 1107 MHz GDDR3 is now 1215 Mhz and Shader Clock is 1458 Mhz.  I have an FTW  Let's see what good it does me

RESULTS OF 4 Consecutive Performance track playbacks:

Core 2 Quad Q9650 4005.1 MHz Bus 444.9 Rated FSB 1780.1 DRAM 534 Core 1.296V
GTX 280 OC'd GPU Core 691 SHADER 1458 Memory 1215 Fan 75%

Frames, Time (ms), Min, Max, Avg    Test    Time     C1 C2 C3 C4 GPU CPULOAD%
  7887,    179934,  28,  66, 43.833 21:18   21:18:00,50,47,44,46,60,74.0
  7659,    179600,  27,  66, 42.645 21:36   21:36:00,49,47,45,46,61,73.0
  7583,    178973,  27,  61, 42.370 21:40   21:40:05,51,47,45,46,61,73.7
  7603,    179252,  27,  62, 42.415 21:48   21:48:00,49,47,45,46,57,73.4

0 Min, +4 Max, +1 Avg

EDIT: I need to retest this  I can't decide which of the 2 Max pair are outliers....

S~

Gunny

[TX-Gunslinger]

Now let's move on to ROF settings in game.  I have a Creative Sound Blaster X-Fi pcie card installed.
We'll keep the GTX-280 Overclocked for the last few tests today.
Audio testing.  All tests so far have been with Audio at 64 counts and High Quality.

Let's drop to minimum with 32 sound counts and Low Quality:

Soundcounts to 32 Sound Quality to LOW
Frames, Time (ms), Min, Max, Avg
  7791,    179192,  28,  61, 43.479
  7705,    178628,  28,  61, 43.134
  7714,    179294,  27,  62, 43.024
  7695,    179168,  27,  62, 42.949

S~

Gunny

[TX-Gunslinger]

I hear alot to put your Landscape to medium as it will save frames.  First - it's important to see what the difference is between Medium and High Landscape settings in ROF - here's a picture:



I see a difference between these two.

Look at the performance!  You gain nothing by downgrading your Landscape on a GTX-280 - don't know if this scales to a 260 due to less GDDR3.

Soundcounts back to 64 Quality back to high
Change Landscape to Med
Frames, Time (ms), Min, Max, Avg
  7658,    179233,  26,  60, 42.727
  7556,    179187,  26,  61, 42.168
  7509,    178787,  26,  60, 41.1000
  7701,    182299,  27,  65, 42.244

S~

Gunny

[GOZR]

This is great Gunny.. lots of work.. Setting to high make texture more define and better looking on my side but using a GPU with 1 GB of Vram will work perfectly medium should be for 512 or a bit higher Vram.. but High setting on 512 mb Vram is a no go..i tested it. With the Simulators the GPU OC is not important really only Bird of Prey should get a performance gain. The Simulators love CPU's

[TX-Gunslinger]

It's a whole lot easier to test now with tracks    Good points Gozr.  I'm really glad I got a video card with 1 GB, and as I remember you told me to do that as I was considering a 260 initially, which only has 768.  I also thought I'd "push" my 280 just to make sure I was not having a VRAM problem.

S~

Gunny

[TX-Thunderbolt]

Good tests Gunny. While I haven't posted on it, I DO follow such things and appreciate the tedium.

Interesting on the landscape settings. I leave everything on my end set to the max on all settings and while my testing was considerably more rudimentary in comparison, I came to a similar conclusion (i.e. that the higher settings had no discernible negative impact...so I leave them up there).

In another month or so (maybe two) NV should release their Fermi based, next-gen GPU's. I've already set aside some cash and will probably be an early adopter unless preliminary tests show it to be a dud (a la the FX5900 Ultra). If it's as good as rumors state it should be a nice, if not remarkable, upgrade.

Thanks for posting this and for putting it in a public area as I frequently point people to these forums for just such purposes.

S!

[TX-Gunslinger]

Thanks TB.  Also as a summary point, 16X AA + 16X AF for 1GB GDDR3 or 5 200 series cards is probably the optimal setting.  It sure helps with ground texture "flickering".  I think Gozr get's a nod here - as I believe I can remember that he's preferred 16 over 8 AA for some time.

I don't have time to post everything today - but I do have pics soon for probably another thread.

My crashes came back the day before yesterday and I was able to clock on GPU-Z, what might be the world-record temperature for a video card that DID not completely fry. That temp was:

171.5 degrees C - !

This was not on my GPU - but on my VDDC temps.   I should just RMA it - but I'm too stubborn.  So, I opened it up and sure enough - the VRM chips Thermal Pads had almost evaporated - ugly pics.

Anyway, I replaced all the thermal grease on GPU and other parts ..... temps on everything are much lower - however VDDC under load will still climb to 110C.

Beware the really crappy thermal material on these cards....  many posts on the web.

As soon as I get a chance I'll post some pics....

More later,,

S~
Gunny


P.S. Takeaway is get a chopy of GPU-Z and check the temps that you cannot see any other way.......

[TX-TOAD]

The only thing I suggest is, that you select a track for testing, and share the track, so others test the same thing.

Sort like "The Black Death" track in IL2

[TX-Gunslinger]

Hey Toad - I missed your post here.  Just came back to it to check my settings for testing this weekend.

Rgr on Black death ..... and tracks.  I've a question to ask in the "R&D" forum - effects track release.

S~

Gunny