Some 7870 Overclocking GPU and Memory performance scaling

[Frakk]

For anyone who is interested....

This will show what you might expect with your actual BF3 gaming performance, i say might because every card is different, but its still a good guide.

Methodology used, Campaign: Going Hunting. It has a good few minutes where you have no control of the game, it plays out as a story, for that reason its consistent so the performance gain measurements are about as accurate as can be, only that part of the game was used to get the results.

Example Vid

Settings are Ultra Preset @ 1920 x 1080P

Because the stock reference 7870 is 1000Mhz / 1200Mhz thats what we will start with.

First GPU clock only

1000Mhz / 1200Mhz
Min 41 - Max 77 - Avg 53.875

1100Mhz / 1200Mhz
Min 43 - Max 80 - Avg 56.983 \ GPU overclock = 10% Avg FPS Performance increase = 6%

1150Mhz / 1200Mhz
Min 43 - Max 82 - Avg 57.856 \ GPU overclock = 15% Avg FPS Performance increase = 8%

1200Mhz / 1200Mhz
Min 45 - Max 84 - Avg 59.397 \ GPU overclock = 20% Avg FPS Performance increase = 10%


Now the Memory only, with the 20% GPU overclock.

1200Mhz / 1300Mhz
min 46 - Max 87 - Avg 60.992 \ Mem overclock = 8% Avg FPS Performance increase = 3.5%

1200Mhz / 1400Mhz
min 48 - Max 88 - Avg 62.584 \ Mem overclock = 16% Avg FPS Performance increase = 5%

1200Mhz / 1500Mhz
min 50 - Max 92 - Avg 64.486 \ Mem overclock = 25% Avg FPS Performance increase = 8%

With a 20% GPU and 25% Mem overclock total scaling is 19.4%


As requested @ 1680 x 1050

1000Mhz / 1200Mhz
Min 46 - Max 87 - Avg 60.989

1100Mhz / 1200Mhz
Min 49 - Max 97 - Avg 64.454 \ GPU overclock = 10% Avg FPS Performance increase = 6%

1200Mhz / 1200Mhz
Min 50 - Max 96 - Avg 66.717 \ GPU overclock = 20% Avg FPS Performance increase = 10%


Now the Memory only, with the 20% GPU overclock.

1200Mhz / 1300Mhz
min 52 - Max 98 - Avg 68.967 \ Mem overclock = 8% Avg FPS Performance increase = 3%

1200Mhz / 1400Mhz
min 53 - Max 102 - Avg 70.577\ Mem overclock = 16% Avg FPS Performance increase = 6%

1200Mhz / 1500Mhz
min 55 - Max 105 - Avg 72.758 \ Mem overclock = 25% Avg FPS Performance increase = 9%


With a 20% GPU and 25% Mem overclock total scaling is 19.4%

It's the same.

-----------------------------------------------------------------------


Test again using the OCF methodology

Example Vid -

Operation Semper Fidelis, Ultra Preset @ 1920 x 1080P

1000Mhz / 1200Mhz
Min 42 - Max 83 - Avg 57.099

1100Mhz / 1200Mhz
Min 42 - Max 86 - Avg 59.857 \ GPU overclock = 10% Avg FPS Performance increase = 5%

1200Mhz / 1200Mhz
Min 46 - Max 89 - Avg 61.841 \ GPU overclock = 20% Avg FPS Performance increase = 8.5%


Now the Memory only, with the 20% GPU overclock.

1200Mhz / 1300Mhz
Min 50 - Max 92 - Avg 64.553 \ Mem overclock = 8% Avg FPS Performance increase = 4.5%

1200Mhz / 1400Mhz
Min 49 - Max 92 - Avg 65.253 \ Mem overclock = 16% Avg FPS Performance increase = 5.5%

1200Mhz / 1500Mhz
Min 50 - Max 96 - Avg 67.847 \ Mem overclock = 25% Avg FPS Performance increase = 8.5%


With a 20% GPU and 25% Mem overclock total scaling is 19%

-----------------------------------------------------------------------


Some Multiplayer (48 - 64 Player's) BF3 benches for refrance, All at Ultra Preset 1920 x 1080P (Multiplayer will vary greatly, this is by now way a comparison bench, just rough reference)

GPU: 1200 / 1500

Metro: Min 42 / Max 104 / Avg 80.015 FPS

*more to come*

-----------------------------------------------------------------------

I thought i might add the canned Avp bench with this.

Setting Medium, (Canned default settings)

Resolution: 1920 x 1080
Texture Quality: 2
Shadow Quality: 3
Anisotropic Filtering: 16
SSAO: ON
Vertical Sync: OFF
DX11 Tessellation: ON
DX11 Advanced Shadows: ON
DX11 MSAA Samples: 1

Clock only

1000Mhz / 1200Mhz = 66.8 FPS
1100Mhz / 1200Mhz = 70.5 FPS
1200Mhz / 1200Mhz = 73.9 FPS

Memory only

1200Mhz / 1300Mhz = 76.4 FPS
1200Mhz / 1400Mhz = 78.6 FPS
1200Mhz / 1500Mhz = 80.5 FPS

AvP D3D11 Benchmark Report
==========================
**************************************************
* Report Created: 2012-09-01 @ 16:06:58
**************************************************
* Executable Build: V1.03, Apr 19 2010
**************************************************
*DX11 Hardware Detected*
Using Default Video Settings:
Resolution: 1920 x 1080
Texture Quality: 2
Shadow Quality: 3
Anisotropic Filtering: 16
SSAO: ON
Vertical Sync: OFF
DX11 Tessellation: ON
DX11 Advanced Shadows: ON
DX11 MSAA Samples: 1

Benchmark Summary:
Number of frames: 8444
Average Frame Time: 12.4ms
Average FPS: 80.5

Settings high (same as OCF Settings)

Resolution: 1920 x 1080
Texture Quality: 3
Shadow Quality: 3
Anisotropic Filtering: 16
SSAO: ON
Vertical Sync: OFF
DX11 Tessellation: ON
DX11 Advanced Shadows: ON
DX11 MSAA Samples: 4

Clock only

1000Mhz / 1200Mhz = 47.2 FPS
1100Mhz / 1200Mhz = 49.5 FPS
1200Mhz / 1200Mhz = 51.2 FPS


Memory only

1200Mhz / 1300Mhz = 53.4 FPS
1200Mhz / 1400Mhz = 54.7 FPS
1200Mhz / 1500Mhz = 56.9 FPS

AvP D3D11 Benchmark Report
==========================
**************************************************
* Report Created: 2012-09-01 @ 17:33:15
**************************************************
* Executable Build: V1.03, Apr 19 2010
**************************************************
*DX11 Hardware Detected*
Using Video Settings from file <ocfconfig.txt>:
Resolution: 1920 x 1080
Texture Quality: 3
Shadow Quality: 3
Anisotropic Filtering: 16
SSAO: ON
Vertical Sync: OFF
DX11 Tessellation: ON
DX11 Advanced Shadows: ON
DX11 MSAA Samples: 4

Benchmark Summary:
Number of frames: 5972
Average Frame Time: 17.6ms
Average FPS: 56.9

 

[scrambler]

Got you to test your mem overclock for real world stuff? : D

 

[Frakk]

Got you to test your mem overclock for real world stuff? : D

Yes you did

I was a little surprised at the result, and the card seems really happy running 1500Mhz Memory, you probably wouldn't notice it with your fast 384bit Memory bus but the texture fill with maxed AA just seems to be in the zone on screen, before you could just catch it filling in GFX as you were going along on the biggest maps, just very slight... not any more.

 

[scrambler]

I noticed the frames : P

I do however still notice a line of "detail" in BF3 that really bugs the crap out of me. Nothing I do gets rid of it.

 

[Frakk]

I noticed the frames : P

I do however still notice a line of "detail" in BF3 that really bugs the crap out of me. Nothing I do gets rid of it.

just out of interest what is it?

 

[EarthDog]

Thanks for the testing! Did you use a cutscene meaning you have ZERO control over the game, its just a movie essentially?

just out of interest what is it?

I think what he was trying to say is he see's a distinct line where details are better (closer) and worse (farther) in the scene... right scrambler? Can you post a screenshot (in a thread of your own since this is about FPS) and we can take a look?

 

[scrambler]

The line of detail? It's a line that I see that literally "details the world" as I move forward. Not even questionable that it is there. On the side closes to me I see full detail and on the other side I see blank textures. Nothing else. Buildings, vehicles but never players. Players are always shown in full detail. Stock, overclock, stock CPU, overclocked CPU, nothing. It sits there and mocks me.

Even weirder is that on some maps, it can be across the world, where I only see it in a heli or jet. On other, Seine Crossing as an example, it is only like 20 feet.

 

[Frakk]

@ ED, i copied what Toms Hardware do http://www.tomshardware.com/reviews/battlefield-3-graphics-performance,3063-2.html

I went further than that thought, just the first few seconds doesn't push the card enough, the sequence is about 4 mints long in total before it hands control of the game to you. (there is one short part in that where you check flaps) i don't think 5 seconds of looking left and right will have any effect overall, but i wanted to get some more demanding flying about and large detailed scenery in with the bench too.

I dont think your math is correct. Not ONE example showed more than 10% so how does it improve 19.4% overall?

I tested the GPU and memory separately, if you look again. if you calculate the stock Avg FPS, add 19.4% to it what you will find is a match for the Avg FPS at the highest overall clock.

 

[EarthDog]

Yep. I needed my morning Joe before I do math. Its correct.

As far as cutscenes, its as good as it gets... but its not good. They said this about their own testing:

The video clip above demonstrates the short sequence used for benchmarking. It's not the most demanding segment of Battlefield 3 I've seen thus far, nor is it necessarily the most representative of single-player game play (and certainly not multi-player). However, after a couple of hundred runs, I know it's consistent.

That makes this system precise, but not necessarily accurate, so its validity could certainly be called into question. The challenge in a game like this, which doesn't offer the timedemo functionality purportedly made available to IHVs, but not press, is that nailing both precision and accuracy then becomes very difficult due to Fraps-based testing of actual sequences. While not perfect, what we have here is, at the very least, great for relativistic comparisons.

Not much you can do about it really. Though through our editing team's testing of manual run throughs in SP BF3, the results were surprisingly repeatable across reviewers and do depict a much more accurate (as in real game play FPS) result than a cutscene.

Anyhoo, thanks again for the testing. Good to see this scaling info Im curious to see, if you have the time, if it scales better/worse at a lower resolution (1680x1050) or higher if you have the monitor. Im also curious to see if the scaling is the same for 7950/7970...etc and on the NVidia line.

 

[Frakk]

All my monitors are 1920 x 1080 max, but yes i can test in 1680x1050, and will later

 

[Frakk]

As requested @ 1680 x 1050

1000Mhz / 1200Mhz
Min 46 - Max 87 - Avg 60.989

1100Mhz / 1200Mhz
Min 49 - Max 97 - Avg 64.454 \ GPU overclock = 10% Avg FPS Performance increase = 6%

1200Mhz / 1200Mhz
Min 50 - Max 96 - Avg 66.717 \ GPU overclock = 20% Avg FPS Performance increase = 10%


Now the Memory only, with the 20% GPU overclock.

1200Mhz / 1300Mhz
min 52 - Max 98 - Avg 68.967 \ Mem overclock = 8% Avg FPS Performance increase = 3%

1200Mhz / 1400Mhz
min 53 - Max 102 - Avg 70.577\ Mem overclock = 16% Avg FPS Performance increase = 6%

1200Mhz / 1500Mhz
min 55 - Max 105 - Avg 72.758 \ Mem overclock = 25% Avg FPS Performance increase = 9%


With a 20% GPU and 25% Mem overclock total scaling is 19.4%

It's the same.

Do some nvidia scaling for me if you have the time, please.

 

[EarthDog]

Do some nvidia scaling for me if you have the time, please.

It will be a while. Im buried in reviews/life...perhaps its better to get it from someone else, sorry. Not to mention, I have no idea where that cutscene even is as i do not touch SP outside of the first scene/level for our testing. If I have to play the SP game for a significant period of time, its not happening from me (unless I can get a savegame file or something?)

Thank you for the lower res testing.

 

[Frakk]

No its ok, i understand. Don't know about the save game file.... "It will be a while" -> if you don't mind telling me your exact methodology for that testing i'm willing to replicate it if you don't still have one of those 7870's you reviewed earlier.

 

[EarthDog]

I still have one, but its FS and there is little to no chance I can get to that kind of testing before its sold for the same reasons listed above.

The method is posted on the Front Page. What we test, while not the best method either, is better than a movie cutscene. This area is, for the most part canned in that there are little scenery changes and little in the way of options to move except through the train. To be honest, I wish we would have went deeper in the game to find a similar 'half canned' scene later in the SP that better represents the FPS one would get. Our results, due to the indoor scene are always notably higher than other sites. That said, I would imagine scaling to be in the ballpark as your findings here... but we will see.

 

[Frakk]

Ok, i will re-run the tests using that methodology. Later, i'm all tested out for now

I agree they are probably the same levels of scaling, its just interesting to compare the 2 different architectures in different ways.

 

[EarthDog]

I actually believe that Nvidia's would be LESS than AMD as far as scaling in overclocking. No clue why I feel that way though..

 

[Ivy]

Didnt expect the RAM to scale that good at above 1300 MHz. Prehaps the pushed core was helpful on that. It definitely will be a hard battle for the 660 ti.

 

[Frakk]

[Edit] I had to redo the test because i forgot i had FXA-Injector v1.3 BETA installed, it reduces FPS, i have Adjusted the results accordingly, what has changed is higher FPS across the board, and slight differences in Scaling.

 

[Frakk]

Added Operation Semper Fidelis @ 1920 x 1080P to Post #1 This time without FXA-Injector ^^^^^ installed.

Will add @ 1680 x 1050 later, and maybe AVP and the actual FPS hit that FXA-Injector v1.3 BETA has, it seems to be about 2-3 FPS

 

[amddork]

What temperatures are you experiencing at the 1200mhz/1500mhz level and is it going to need water cooling? I have the Asus 7870 ghz edition, so this thread directly correlates to my situation, also my current psu that is not going to get me anywhere overclocking I was wondering what you recommend?

Will be running

965BE(Maybe a 8150)
Asus Crosshair V 990FX
2 x 8gb Corsair Vengeance 2133mhz
Asus 7870 ghz edition
1x SSD
2x WD Caviar black in RAID 0
1x BluRay combo drive
6x 120mm Fans

My watercooling has it's own psu in a seperate place.