![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
B^gD^ddy~CF~
Limp Gawd
- Joined
- Nov 5, 2006
- Messages
- 172
Ok I did this to test which is actually better in Battlefield 3 the 8.921.2 RC11 driver or the 12.2 Pre-Certified driver.
Test Setup
OS
Windows 7 x64
Hardware
Case
Thermaltake Armor Series VA8003BWS Black Full Tower Case w/ 25CM Fan
Cooling
Noctua NH-U12P SE1366 120mm SSO CPU Cooler
3 x Thermaltake DuraMax 12 AF0060 120mm Case Fan
4 x Thermaltake TT-9025A-2B 90mm Case Fan
Motherboard
GIGABYTE GA-EX58-UD5
Processor
Intel Core i7-920 BX80601920
Memory
2 x OCZ Platinum 6GB (3 x 2GB) 240-Pin DDR3 SDRAM DDR3 1600 OCZ3P1600LV6GK
Storage
Pioneer Black Blu-Ray Burner BDR-203BKS
PLEXTOR 24X DVD Writer Black PX-880SA-26
Intel X25-M SSDSA2MH080G1 SSD
6 x WD Caviar Black WD1001FALS 1TB
WD Caviar Black WD2001FASS 2TB
Power
BFG Tech EX-1200 BFGR1200WEXPSU
GPU
XFX FX-797A-TNFC Radeon HD 7970
SAPPHIRE 21197-00-40G Radeon HD 7970
Displays
BenQ G2400WD
Lenovo ThinkVision L2440x
SAMSUNG 2443BW
Connectivity/Peripherals/Sound
SIIG SuperSpeed USB 3.0 Bay Hub+Host Kit Model JU-H40411-S1
Creative Labs GH0100 Sound Blaster World of Warcraft Wireless Headset
Logitech X-530 70 Watts 5.1 Speaker
GRIFFIN - radio SHARK 2
Philips Webcam SPC1300NC
Logitech G13 USB Advanced Gameboard
HP Scanjet G3110
Fragpedal Quad
APC XS 1500 UPC
CPU-Z Report Belarc Advisor Report
Software Used
HWiNFO64 3.93-1530
Fraps 3.4.7
AfterBurner 2.2.0 Beta 14
BF3 Latest Version
Software Configuration
HWiNFO64- Set to scan interval 5000 ms(5 seconds)
MSI AfterBurner - Set to fan speed update period 5000 ms(5 seconds), hardware polling period 5000 ms(5 seconds)
Fraps - Standard benchmarking mode with FPS, Frame Time, MinMaxAvg checked
Drivers Used
8.921.2 RC11 with 12.1 Cap 3
12.2 Pre-Certified Driver(Actually reported as 12.3) with 12.1 Cap 3
The 12.2 drivers had the atipdlxx.dll from RC11 copied to to "C:\Windows\SysWOW64", and atipdl64.dll from RC11 copied to "C:\Windows\System32" while in safe mode after the initial installation. These files are present in the RC11 driver extraction location which is normally "C:\AMD\AMD_Radeon_HD7900_Win7_64\Packages\Drivers\Display\W76A_INF\B132075\" where they will have the extension .dl_ so that they have to be renamed either before copying or after they are copied to the correct folder. If this was not done then there were clock irregularities, difficulties with some programs, and other odd behavior.
Both drivers were installed with the following process. Any previous Catalysts were fully removed via the Catalyst uninstall manager. The system was booted into safe mode where it had a driver cleaning with Driver Sweeper 3.2 from Phyxion.net(this is the latest version vs the one from Guru3d). The system was rebooted and then the driver was installed on a "clean" installation.
Testing Methodology
HWiNFO64, FRAPS, AfterBurner started, Origin launched, BF3 Campaign launched, single player game in Operation Swordbreaker started. The difficulty was set to hard. If at any point during the test I died then the test was repeated until I didn't die. The test starts when you step out of the troop carrier and gain control of your character. At that point I pause the game, alt+tab out, start logging for HWinfo64 and MSI AfterBurner. After those two are started I alt+tab back into BF3 and un-pause the game. After un-pausing I immediately start Fraps's benchmarking mode. The test ends when you take out the sniper with the AT rocket and the "saving game" notice appears on the screen. At that point I shut down FRAPS's benchmarking mode, alt+tab out and kill logging in HWiNFO64 and AfterBurner as well. All except the Eyefinity resolution runs were launched from a mirrored display desktop to avoid the BF3 Crossfire scaling bug which occurs when the game is launched from an Eyefinity desktop but played at a single monitor resolution.
Note On Time Graphs
There are 8 saves in the portion of the game that I play through for the benchmark, but there are 9 long frames that occur in the same spots every time. I initially thought there were 9 auto-saves occurring within the portion of the game I was testing but the more frame time graphs I piled through the more I began to wonder. So I recorded a testing session from beginning to end and counted them up. There were only 8. Taking note of where they occurred and how far they were apart. My best guess is that the first major time frame delay is not from an auto-save. I can only guess that new textures or assets are being loaded for the rest of the map at this point. I have fired off an e-mail to one of the engine developers at DICE to see if he can give me any further insight. You can see and feel the stutter on that first frame time spike. If this occurs in multiplayer at a bad point your dead it's as simple as that. I think this might only occur on large maps in multiplayer where the card can't hold all of the assets at once even with 3Gb of ram, and consequently new assets have to be loaded in on the fly.
If you look at the graph below I've numbered where the auto-saves happen and long frame time outliers are. That first spike tends to be the only one that really affects game play noticeably mostly because you are on the move when it happens, but at 130-190 ms it's a heck of a pause. If you correlate this to the FPS over time it doesn't stand out because it happens in less than a second and then your back to your normal frame rate in the area.
1. This is not an auto-save. A big jerk occurs as you turn the corner into the building with the door that leads to the alleyway before the school.
2. This is the auto-save that occurs as the guy kicks the door in so you can run down the alleyway toward the school
3. This is the auto-save that occurs as your getting ready to go into the building before the firefight
4. This is the auto-save that occurs after you've dragged your buddy to safety during the first part of the firefight
5. This is the auto-save that occurs right before the guys with RPGs let loose in the second part of the firefight
6. This is the auto-save that occurs after the firefight as you enter the building to go up the stairwell
7. This is the auto-save that occurs right before you go through the door onto the roof
8. This is the auto-save that occurs right after you pick up the TOW anti-tank missile
9. This is the auto-save that occurs right after you take out the sniper with the TOW anti-tank missile
This is it until you get to the 5292x1050 test for 12.3 you'll see what I mean. It becomes very hard to tell exactly which ones are the auto-saves and which ones are just long frame times. As such I included all long frame times in calculations. The only time frames which are excluded from a calculation is if the frame time is 0 and I'm calculating minimum frame time or average frame time.
Now down to it
What can you say about that. Both drivers never drop below 88 frames per second according to the individual frames per second that are recorded. Both drivers trade blows as to which is more chaotic in frame rate. Interestingly FRAPS own MinMaxAvg frame rate data disagrees with this. I can only guess the MinMaxAvg data is being polled more than once per second whereas the FPS data is being polled only once per second.
12.3 has the higher minimum frame rate as well as the higher average but they are less than a frame apart for average frame rate
Click these to get huge original graphs! It appears as though RC11 has 12.3 beat here in terms of just having fewer random long frame times and having frame times that are just less chaotic overall
RC11 has 12.3 beat in terms of max frame time. While 12.3 has the lower minimum frame time and the lower average frame time but not by much.
RC11 is looking better and better as far as responsiveness goes. 99% of the frames from the RC11 are almost a full millisecond faster than 12.3.
It becomes pretty clear that 12.3 just has quite a few longer frame times than RC11 has. On that basis alone I would recommend RC11 over 12.3 if you're a competitive BF3 gamer. Even though 12.3 has a lower frame time for it's smallest auto-save frame time when you look at all of the numbers it goes back and forth on different auto-saves as to whether RC11 or 12.3 has the lowest frame time.
12.3 looks to have RC11 beat in terms of GPU utilization. At one point RC11 chokes and falls below 80 % where 12.3 never drops below 90%
I really don't know what to make of these numbers. Either BF3 is using so much memory that it is having to be streamed over the PCIE bus or AfterBurner has a weird way to calculating video memory usage. Either way it looks like 12.3 uses more memory than RC11 in BF3
Well I don't know if it's a good or a bad thing but RC11 total CPU usage is just higher on every account.
Well there's the reason for the low total CPU usage right there. On 12.3 the thread usage is not only low for each thread but also it's all over the place. This is especially true for the last three threads. The usage is still low overall though. I'm not sure what to make of this. You could argue it either way. You could be making fuller use of your hardware on RC11 or maybe 12.3 just have more optimizations for better efficiency.
Nothing to see here the difference are mostly due to whether I nailed the benchmark on the first try or whether I died and had to repeat it a few times.
OK now time for 5292x1050
12.3 just falls apart at this resolution. It starts low and stays there the only good thing about it is that's it has slightly less extreme swings in fame rate
Both drivers have minimum frame rates that I would classify as too low to play competitively. Never mind that multiplayer worst case scenarios are more demanding than the campaign worst case scenarios. RC11 does manage to get an average frame rate that is closer to playable. 12.3 is just dropping the ball....hard
Both drivers clearly pick up quite a bit more long frame times at this resolution. RC11 is about where you'd expect but 12.3 just gets purely chaotic. You can count the number of frames above 50 ms in RC11 just looking at the graph, I challenge you to do the same with 12.3
RC11 schools 12.3 on every metric of frame time
Now you really begin to see the true performance picture 99% of the frame times produced by RC11 are 26.581 ms faster than the frame times produced by 12.3.
Wow what can you you say about that. 12.3 is only 120 frames away from having ALL of its frame times above 20 ms with just over four times RC11's number of frames above 20 ms. 22% of 12.3's frame times are above 50 ms with thirty six times RC11's number of frames above 50 ms. There has to be a serious bug in the 12.3 drivers somewhere that is causing this king of performance devolution.
RC11 does a better job keeping the GPU's pegged than 12.3 at this resolution
Both drivers are being reported by AfterBurner to be going above the actual amount of VRAM on the cards again and RC11 is still lower in it's usage than 12.3. Maybe this is what's causing the horrible performance of 12.3?
at this resolution you are of course much more GPU limited. However RC11 is using almost 20% more processing power than 12.3
We again see that the last three threads on 12.3 are Yo-Yoing all over the place but thread usage remains low in comparison to RC11
Not much to see here just like at 1920x1200 the starting temps are the main difference, mostly based on whether I died and had to redo the benchmark or not. You can however see the CPU temp actually rise a lot more in the RC11 run compared to the 12.3 run
Well that's it. It definitely looks like AMD is walking backward at least in this game with 12.3 Pre-Certified. It's a shame too because I like some of the additions made to 12.3 and some of the reorganization they've done in the Catalyst Control panel. We can only hope this gets fixed by the time 12.3 is final. Or if not for god's sake by 12.4. I'd advise you to stay well clear of 12.3 if your a competitive BF3 gamer for now based on these results. I have reported all issues encountered to both AMD tech support and the AMD driver feedback team.
If you want to look at the raw files used to put this together then those can be had here
Test Setup
OS
Windows 7 x64
Hardware
Case
Thermaltake Armor Series VA8003BWS Black Full Tower Case w/ 25CM Fan
Cooling
Noctua NH-U12P SE1366 120mm SSO CPU Cooler
3 x Thermaltake DuraMax 12 AF0060 120mm Case Fan
4 x Thermaltake TT-9025A-2B 90mm Case Fan
Motherboard
GIGABYTE GA-EX58-UD5
Processor
Intel Core i7-920 BX80601920
Memory
2 x OCZ Platinum 6GB (3 x 2GB) 240-Pin DDR3 SDRAM DDR3 1600 OCZ3P1600LV6GK
Storage
Pioneer Black Blu-Ray Burner BDR-203BKS
PLEXTOR 24X DVD Writer Black PX-880SA-26
Intel X25-M SSDSA2MH080G1 SSD
6 x WD Caviar Black WD1001FALS 1TB
WD Caviar Black WD2001FASS 2TB
Power
BFG Tech EX-1200 BFGR1200WEXPSU
GPU
XFX FX-797A-TNFC Radeon HD 7970
SAPPHIRE 21197-00-40G Radeon HD 7970
Displays
BenQ G2400WD
Lenovo ThinkVision L2440x
SAMSUNG 2443BW
Connectivity/Peripherals/Sound
SIIG SuperSpeed USB 3.0 Bay Hub+Host Kit Model JU-H40411-S1
Creative Labs GH0100 Sound Blaster World of Warcraft Wireless Headset
Logitech X-530 70 Watts 5.1 Speaker
GRIFFIN - radio SHARK 2
Philips Webcam SPC1300NC
Logitech G13 USB Advanced Gameboard
HP Scanjet G3110
Fragpedal Quad
APC XS 1500 UPC
CPU-Z Report Belarc Advisor ReportSoftware Used
HWiNFO64 3.93-1530
Fraps 3.4.7
AfterBurner 2.2.0 Beta 14
BF3 Latest Version
Software Configuration
HWiNFO64- Set to scan interval 5000 ms(5 seconds)
MSI AfterBurner - Set to fan speed update period 5000 ms(5 seconds), hardware polling period 5000 ms(5 seconds)
Fraps - Standard benchmarking mode with FPS, Frame Time, MinMaxAvg checked
Drivers Used
8.921.2 RC11 with 12.1 Cap 3
12.2 Pre-Certified Driver(Actually reported as 12.3) with 12.1 Cap 3
The 12.2 drivers had the atipdlxx.dll from RC11 copied to to "C:\Windows\SysWOW64", and atipdl64.dll from RC11 copied to "C:\Windows\System32" while in safe mode after the initial installation. These files are present in the RC11 driver extraction location which is normally "C:\AMD\AMD_Radeon_HD7900_Win7_64\Packages\Drivers\Display\W76A_INF\B132075\" where they will have the extension .dl_ so that they have to be renamed either before copying or after they are copied to the correct folder. If this was not done then there were clock irregularities, difficulties with some programs, and other odd behavior.
Both drivers were installed with the following process. Any previous Catalysts were fully removed via the Catalyst uninstall manager. The system was booted into safe mode where it had a driver cleaning with Driver Sweeper 3.2 from Phyxion.net(this is the latest version vs the one from Guru3d). The system was rebooted and then the driver was installed on a "clean" installation.
Testing Methodology
HWiNFO64, FRAPS, AfterBurner started, Origin launched, BF3 Campaign launched, single player game in Operation Swordbreaker started. The difficulty was set to hard. If at any point during the test I died then the test was repeated until I didn't die. The test starts when you step out of the troop carrier and gain control of your character. At that point I pause the game, alt+tab out, start logging for HWinfo64 and MSI AfterBurner. After those two are started I alt+tab back into BF3 and un-pause the game. After un-pausing I immediately start Fraps's benchmarking mode. The test ends when you take out the sniper with the AT rocket and the "saving game" notice appears on the screen. At that point I shut down FRAPS's benchmarking mode, alt+tab out and kill logging in HWiNFO64 and AfterBurner as well. All except the Eyefinity resolution runs were launched from a mirrored display desktop to avoid the BF3 Crossfire scaling bug which occurs when the game is launched from an Eyefinity desktop but played at a single monitor resolution.
Note On Time Graphs
There are 8 saves in the portion of the game that I play through for the benchmark, but there are 9 long frames that occur in the same spots every time. I initially thought there were 9 auto-saves occurring within the portion of the game I was testing but the more frame time graphs I piled through the more I began to wonder. So I recorded a testing session from beginning to end and counted them up. There were only 8. Taking note of where they occurred and how far they were apart. My best guess is that the first major time frame delay is not from an auto-save. I can only guess that new textures or assets are being loaded for the rest of the map at this point. I have fired off an e-mail to one of the engine developers at DICE to see if he can give me any further insight. You can see and feel the stutter on that first frame time spike. If this occurs in multiplayer at a bad point your dead it's as simple as that. I think this might only occur on large maps in multiplayer where the card can't hold all of the assets at once even with 3Gb of ram, and consequently new assets have to be loaded in on the fly.
If you look at the graph below I've numbered where the auto-saves happen and long frame time outliers are. That first spike tends to be the only one that really affects game play noticeably mostly because you are on the move when it happens, but at 130-190 ms it's a heck of a pause. If you correlate this to the FPS over time it doesn't stand out because it happens in less than a second and then your back to your normal frame rate in the area.
1. This is not an auto-save. A big jerk occurs as you turn the corner into the building with the door that leads to the alleyway before the school.
2. This is the auto-save that occurs as the guy kicks the door in so you can run down the alleyway toward the school
3. This is the auto-save that occurs as your getting ready to go into the building before the firefight
4. This is the auto-save that occurs after you've dragged your buddy to safety during the first part of the firefight
5. This is the auto-save that occurs right before the guys with RPGs let loose in the second part of the firefight
6. This is the auto-save that occurs after the firefight as you enter the building to go up the stairwell
7. This is the auto-save that occurs right before you go through the door onto the roof
8. This is the auto-save that occurs right after you pick up the TOW anti-tank missile
9. This is the auto-save that occurs right after you take out the sniper with the TOW anti-tank missile
This is it until you get to the 5292x1050 test for 12.3 you'll see what I mean. It becomes very hard to tell exactly which ones are the auto-saves and which ones are just long frame times. As such I included all long frame times in calculations. The only time frames which are excluded from a calculation is if the frame time is 0 and I'm calculating minimum frame time or average frame time.
Now down to it
What can you say about that. Both drivers never drop below 88 frames per second according to the individual frames per second that are recorded. Both drivers trade blows as to which is more chaotic in frame rate. Interestingly FRAPS own MinMaxAvg frame rate data disagrees with this. I can only guess the MinMaxAvg data is being polled more than once per second whereas the FPS data is being polled only once per second.
12.3 has the higher minimum frame rate as well as the higher average but they are less than a frame apart for average frame rate
Click these to get huge original graphs! It appears as though RC11 has 12.3 beat here in terms of just having fewer random long frame times and having frame times that are just less chaotic overall
RC11 has 12.3 beat in terms of max frame time. While 12.3 has the lower minimum frame time and the lower average frame time but not by much.
RC11 is looking better and better as far as responsiveness goes. 99% of the frames from the RC11 are almost a full millisecond faster than 12.3.
It becomes pretty clear that 12.3 just has quite a few longer frame times than RC11 has. On that basis alone I would recommend RC11 over 12.3 if you're a competitive BF3 gamer. Even though 12.3 has a lower frame time for it's smallest auto-save frame time when you look at all of the numbers it goes back and forth on different auto-saves as to whether RC11 or 12.3 has the lowest frame time.
12.3 looks to have RC11 beat in terms of GPU utilization. At one point RC11 chokes and falls below 80 % where 12.3 never drops below 90%
I really don't know what to make of these numbers. Either BF3 is using so much memory that it is having to be streamed over the PCIE bus or AfterBurner has a weird way to calculating video memory usage. Either way it looks like 12.3 uses more memory than RC11 in BF3
Well I don't know if it's a good or a bad thing but RC11 total CPU usage is just higher on every account.
Well there's the reason for the low total CPU usage right there. On 12.3 the thread usage is not only low for each thread but also it's all over the place. This is especially true for the last three threads. The usage is still low overall though. I'm not sure what to make of this. You could argue it either way. You could be making fuller use of your hardware on RC11 or maybe 12.3 just have more optimizations for better efficiency.
Nothing to see here the difference are mostly due to whether I nailed the benchmark on the first try or whether I died and had to repeat it a few times.
OK now time for 5292x1050
12.3 just falls apart at this resolution. It starts low and stays there the only good thing about it is that's it has slightly less extreme swings in fame rate
Both drivers have minimum frame rates that I would classify as too low to play competitively. Never mind that multiplayer worst case scenarios are more demanding than the campaign worst case scenarios. RC11 does manage to get an average frame rate that is closer to playable. 12.3 is just dropping the ball....hard
Both drivers clearly pick up quite a bit more long frame times at this resolution. RC11 is about where you'd expect but 12.3 just gets purely chaotic. You can count the number of frames above 50 ms in RC11 just looking at the graph, I challenge you to do the same with 12.3
RC11 schools 12.3 on every metric of frame time
Now you really begin to see the true performance picture 99% of the frame times produced by RC11 are 26.581 ms faster than the frame times produced by 12.3.
Wow what can you you say about that. 12.3 is only 120 frames away from having ALL of its frame times above 20 ms with just over four times RC11's number of frames above 20 ms. 22% of 12.3's frame times are above 50 ms with thirty six times RC11's number of frames above 50 ms. There has to be a serious bug in the 12.3 drivers somewhere that is causing this king of performance devolution.
RC11 does a better job keeping the GPU's pegged than 12.3 at this resolution
Both drivers are being reported by AfterBurner to be going above the actual amount of VRAM on the cards again and RC11 is still lower in it's usage than 12.3. Maybe this is what's causing the horrible performance of 12.3?
at this resolution you are of course much more GPU limited. However RC11 is using almost 20% more processing power than 12.3
We again see that the last three threads on 12.3 are Yo-Yoing all over the place but thread usage remains low in comparison to RC11
Not much to see here just like at 1920x1200 the starting temps are the main difference, mostly based on whether I died and had to redo the benchmark or not. You can however see the CPU temp actually rise a lot more in the RC11 run compared to the 12.3 run
Well that's it. It definitely looks like AMD is walking backward at least in this game with 12.3 Pre-Certified. It's a shame too because I like some of the additions made to 12.3 and some of the reorganization they've done in the Catalyst Control panel. We can only hope this gets fixed by the time 12.3 is final. Or if not for god's sake by 12.4. I'd advise you to stay well clear of 12.3 if your a competitive BF3 gamer for now based on these results. I have reported all issues encountered to both AMD tech support and the AMD driver feedback team.
If you want to look at the raw files used to put this together then those can be had here
Last edited:
