Overclocking the Gigabyte GTX 670 OC Edition

So I've finally managed to put away a some time to write this post bit by bit so that I can have more time to work on the Oh Baby! rig post. I've managed to benchmark the overclock on six games that have actual benchmark programs built into the game. The games are Mafia II, Just Cause 2, Far Cry 2, Dirt 3, Dawn of War 2: Retribution, and Batman Arkham City. It is a bit odd how all the games happen to be sequels, but it seems more like a coincidence rather than anything else. One thing I like to look at is if the games are V-Sync stable. When I say this I mean averaging at 60+ FPS consistently. The reason I prefer V-Sync is because with this card it can cause tearing in the screen very easily if the FPS is much too low or very high and takes away from the visual experience. I recently downloaded the newest beta nVidia driver (306.02) as it has some minor improvements for the GTX 600 series cards. Here is the following settings of the system during these benchmarks as well as the nVidia Control Panel settings:
--------------------------------------------------------------------------------------------------------------------------
Processor: Intel Core i7 2600K @ 4.5Ghz, 1.38v
Motherboard: MSI Z68A-GD80 (G3)
Memory: Patriot Memory G2 Series PC3-12800 9-9-9-24 @ 1600MHz
Heatsink: Phanteks PH-TC14PE w/ included Phanteks PH-F140TS fans and PH-NDC Paste
Power Supply: Thermaltake Toughpower Grand 80+ Gold 750W Power Supply
Graphics Card: Gigabyte GTX 670 OC Edition in PCI-E 2.0 Mode
Hard Drives: Patriot Pyro 60GB SSD + WD Caviar Black 1TB HDD + WD Caviar Black 2TB HDD
Optical Drive: LG DVD-RW
OS: Windows 7 Ultimate 64-bit
Case: Coolermaster CM690 II (with optional side window installed)

Factory Clock Settings: Core Clock 1100MHz & Memory Clock 1502MHz @ 100% Power Limit
Boosted Clock Settings: Core Clock 1200MHz & Memory Clock 1852MHz @ 100% Power Limit

Part One of nVidia 3D Settings

Part Two of nVidia 3D Settings

--------------------------------------------------------------------------------------------------------------------------
BATMAN: ARKHAM CITY

So the first benchmark I ran was Batman: Arkham City. I ran this one first as I already covered it partially when I did my overview on the GPU. This time, I decided to do benchmarks of both clock settings with and without Physx enabled to see if, without Physx, it would allow a bit more of a spread between the factory clock and my boosted clock. In case you didn't read the overview of the Gigabyte GTX 670 OC card, here were the settings used on Batman: Arkham City:

The Batman: Arkham City display settings

I did benchmarks using those settings with both the "normal" Physx setting and the "off" Physx setting. Once again, just simply trying to see if the amount of clock speed would net a large spread or not on BAC. Here were the results:

The factory benchmark result for Batman: Arkham City with Physx on

This is the exact same result copied over from the overview page. This keeping in mind that I'm using the GPU in PCI-E 2.0 mode rather than 3.0 mode so the performance may improve if I get an Ivy Bridge processor, but that probably won't be for a bit anyway. The game had an average of 85 frames and the highest frame rate was 125. The Physx parts seemed to be approximately the same result between both the factory overclock Physx frame rate and the boosted overclock Physx frame rate. That being said, even with factory clocks the card seems to be pretty qualified to be V-Sync stable even with Physx turned on.

The boosted benchmark result for Batman: Arkham City with Physx on

It's a bit disappointing on the end result of the boosted overclock, at least in terms of the average clock, coming in at 1 frame less than the factory clock. This is most likely due to the intense Physx part where it has all the ice breaking apart in the museum when Penguin is shooting Mr. Freeze's ice gun. Looking at the highest frames though, the highest frame rate was 131. This is a bit of an improvement over the 125 by 6 frames. That's not too bad I guess. The minimum frame rate also seems to have increased from 20 frames per second on the factory clock to 24. So something is definitely being boosted but the average seems to end up about the same, simply because of the intense Physx used in the game. I may consider reinstalling my GTX 460 back into the system as a dedicated Physx card and run this benchmark again to see if there's a better discernible difference between the two when running it with Physx enabled.

The factory benchmark result for Batman: Arkham City with Physx off

So as you can clearly see, having Physx off really allows the GPU to perform a lot more optimally. The average frame rate went from 85 to 124 (a 39 FPS increase). A huge jump but expected as there isn't as much work that the 670 has to do with processing all the Physx objects. The minimum frame rate went from 20 to 36, which is also something nice to see. The biggest difference though was the max frames (even though the average is the most important value of the three) went from 125 to 168 (a 43 FPS increase). Really goes to show how much work Physx puts on nVidia cards, even ones as powerful as the GTX 670.

The boosted benchmark result for Batman: Arkham City with Physx off

Running this test without Physx on really allowed a much better comparison between the two clock settings. The minimum frame rate of the boosted clock settings was 50. That's a 14 FPS improvement over the stock clocks and that's only the minimum frame rate. The average frame rate was pretty close (with the boosted clock having 128 frames which is a 4 FPS increase) so it goes to show how optimized the game has come since it was originally released. The maximum frame rate has quite a bit of spread between the two clock speeds, with the boosted clock having 191 FPS (a 23 FPS increase). So running the game without Physx on these clock settings would almost always guarantee a pretty excellent frame rate regardless of what was happening on screen.

--------------------------------------------------------------------------------------------------------------------------
DIRT 3

This was the second game I decided to run a benchmark on as it has DX11 features. I set all the settings to the maximums that I could (unfortunately I don't have a picture of the settings as the list is too long) in order to stress test and see how the GPU performed. One thing about Dirt 3 is that it's DX11 optimization is pretty good. I unfortunately didn't do multiple runs through on this game, simply because I didn't have enough time to sit and wait several minutes to drive around the track. That being said, the results were fairly close on this benchmark as well. Here were the results:

The factory benchmark result for Dirt 3

The factory overclock performed quite well. It resulted in an average FPS of 98.47 and minimum of 76.68. So no matter what, running this game with V-Sync on it should never drop below 60 (at least according to this benchmark anyway). Pretty excellent I must say.

The boosted benchmark result for Dirt 3

The boosted overclock actually resulted in a lower average FPS than the factory clock (only 94.17 FPS). That being said, this benchmark isn't the most consistent as the AI drivers don't always perform the exact same so the results can be a bit different dependent on if an AI driver decides to crash or not etc. In the case of the boosted overclock benchmark run, there was actually a massive crash between two of the drivers (one hit a wall and the other T-boned him) causing the frame rate to drop a bit simply because of the amount of visual car damage that occurred. The minimum frame rate does indicate some increase in performance however, with the boosted clock coming in at a minimum of 77.53 FPS. So once again, the GPU should have no problems with V-Sync instability.

--------------------------------------------------------------------------------------------------------------------------
FAR CRY 2

I decided to put a couple DX10 games into the benchmark list, simply to show it's performance on games that don't have DX11 enabled. As such, Far Cry 2 has a great benchmark and is a great game to play visually. Here were the settings I used:

The General Settings tab of the Far Cry 2 benchmark

The Game Settings tab of the Far Cry 2 benchmark

I disabled the AI though so that the test would be a bit more consistent instead of having the AI placed randomly and dying randomly. Of course, all the in game visual settings were maxed as that's the only real way to benchmark this GPU and see how far it can be pushed.

The factory benchmark result for Far Cry 3

Far Cry 2 records the results a bit differently than a number of other benchmarks built into some of the newer games. It uses a utility and then records the results onto graphs and into a html page instead of just giving results on screen. The factory clock averaged 83.81 FPS over the course of 3 loops of the benchmark. The max FPS was 136.70 and the minimum was 56.86. Keeping in mind, I ran 3 loops of the test to allow for an average output just in case something visually different happened with one barrel exploding or something like that. This being said, the GPU performed quite well and with the results has almost V-Sync stability in terms of performance. Now lets see how the boosted clock did.

The boosted benchmark result for Far Cry 3

Far Cry 2 seems really favor the boosted clock quite a fair bit, with the results in all three categories reaching over 10 FPS more than the factory clocks. The average FPS of the 3 loops with the boosted clock turned up as 95.67 (a 11.86 FPS increase). The max FPS was 152.87 (a 16.17 FPS increase) and the minimum was 66.54 FPS (a 9.68 FPS increase). So with the boosted clock, V-Sync stability would be very capable (provided a massive number of explosions didn't all occur at once). Very good results indeed.

--------------------------------------------------------------------------------------------------------------------------
JUST CAUSE 2

Just Cause 2 is a game I purchased more recently (even though the game released back in March of 2010). What recently peaked my interest was just that I felt like playing a lot more sandbox adventure games and this one isn't QUITE that but it's pretty fun to goof around in. This game is another DX10 game as well, so this should shed a bit more light on how the GPU performs in a DX10 environment. Here were the settings I used:

The Standard Display Settings options

The Advanced Display Settings options

Anything that could be turned on or set to high were set accordingly (except of course V-Sync, cause I want to see how much higher it goes after 60 FPS of course). The benchmark I ran was "The Dark Tower" map. I was considering running through all three of the benchmarks, but just running "The Dark Tower" map would be sufficient enough to have a general sense of how the FPS will be in a game. Here were the results:

The factory benchmark result for Just Cause 2

Unfortunately Just Cause 2 does not give a min and max FPS designation (I guess it wasn't built into the benchmark?) so the results only yield the average. With stock clocks, it averaged at 58.42, which is pretty good and almost V-Sync stable. Now lets see how much far ahead the boosted clock is performing.

The boosted benchmark result for Just Cause 2

I was expecting the GPU to have a bit more of a spread on the FPS with this boosted clock, especially after doing the Far Cry 2 benchmark. Unfortunately, the FPS only turned out to be 64.52. I was expecting more like a 10 FPS increase rather than only 6.1 FPS. That being said it still appears to be V-Sync stable according to this benchmark, so that's good by me.

--------------------------------------------------------------------------------------------------------------------------
DAWN OF WAR 2: RETRIBUTION

Dawn of War 2 is actually quite a demanding game when a lot of the explosions are getting rendered on screen all at once. As such, even a powerful GPU such as the GTX 670 can take quite a beating. Now, with the benchmark the AI isn't always doing the exact same thing, so sometimes the results can cause for a higher or lower minimum/maximum FPS. As such we will be focusing more on average FPS. Dawn of War 2 runs on DX10, just like Just Cause 2 and Far Cry 2. Here were the settings I used for the benchmark:

The display settings for Dawn of War 2: Retribution

I turned off automatic settings and cranked everything to max. Effects density is one of the main FPS killers in DOW2 because a lot more explosions and debris can be rendered on screen with higher settings.

The factory clock benchmark result for Dawn of War 2: Retribution

The GTX 670 factory clocks performed quite well for the game, resulting in 55.87 FPS as an average. Now as I said previously, effects density + AI can cause a lot of explosions to render on screen and as such the minimum FPS was 12.96 because of the massive number of explosions that occurred on screen. The maximum FPS achieved was 110.22, but of course that was pretty much just looking at some Assault Squad Space Marines running through some rough terrain. Lets see how the boosted clock does.

The boosted clock benchmark result for Dawn of War 2: Retribution

So the boosted clock did improve in terms of average FPS. It achieved 64.65 FPS (an 8.78 FPS increase) over the factory clocks. The maximum FPS also increased coming in at 121.03 FPS (a 10.81 FPS increase). However, the minimum FPS was actually lower during this test. It resulted in 9.9 FPS (a 3.06 FPS decrease) which was due to the inconsistency of the volley of rockets that occur during the benchmark. That being said, the average and maximums did increase, so unless a massive number of rockets occur on screen, the GTX 670 with the boosted clock does perform at V-Sync stable standards.

--------------------------------------------------------------------------------------------------------------------------
MAFIA II

The last game I decided to do a benchmark on was Mafia II, which happened to come free with my GTX 670. This is another game that has Physx built into it, and thus I decided to once again benchmark with and without Physx enabled to see the difference. Here were the settings I used:

The display settings for Mafia II

As usual I turned everything to maximum possible settings. Then I did tests with both Physx on (high setting) and off to see if the situation was just like BAC in terms of improvement. With this benchmark, Mafia II only shows the average FPS (same as Just Cause 2) and as such this will be our basis of how much performance I can get out of the boosted clock. First, lets see how the factory clocks did.

The factory benchmark result for Mafia II with Physx on

So with Physx on, the average FPS was 50.4. That's not too bad considering that Physx can pretty much full load any GPU, even one as powerful as the GTX 670. I believe the frames did drop a bit below the average during the playback of the benchmark, but I can't quite remember the exact FPS. In any case, almost V-Sync stable. 

The boosted benchmark result for Mafia II with Physx on

I expected the boosted clock to perform a bit better than it did. The average FPS was only 55.1 so not quite V-Sync stable even at these clocks. However, that being said, it is very close and definitely playable at that frame rate. Now lets see how the factory clock did with Physx turned off.

The factory benchmark result for Mafia II with Physx off

As to be expected, without Physx on, the GTX 670 performed quite well at factory clocks. The average FPS came out to 85.3 which is definitely V-Sync stable. Never once did I see it drop below 60 FPS at any point during the benchmark, even during the fire sequences, so I'm pretty sure that the game will have no problems at all if I play it without Physx. But who wants to do that when you have the option to use it?  Now lets see how the boosted clock compares with Physx off.

The boosted benchmark result for Mafia II with Physx off

The GTX 670 performed extremely well with the boosted clocks. It averaged 93.5 FPS (an 8.2 FPS increase) with Physx off. That's certainly V-Sync stable. Although it gets such a high FPS, I really would much rather enjoy the game by having Physx on simply because it adds more depth and realism to the game with added particles that would otherwise not be there. If I had my GTX 460 as a dedicated Physx card, the game would probably be V-Sync stable with Physx on. Something I may revisit sometime in the near future.

--------------------------------------------------------------------------------------------------------------------------
TEMPERATURES & CONCLUSION

In my overview I mentioned that the GTX 670 did not reach over 68 degrees celcius in a 26 degree room. That was with the factory overclock and the BIOS fan profile. After overclocking the GTX 670 to the boosted clocks, the temperature never reached over 65 degrees in a 22 degree room. And that's on just the BIOS fan profile settings, after setting it to my own personal settings, it doesn't even reach that half the time. All in all, an excellent cooler design by Gigabyte.

In conclusion, the GTX 670 performs excellent in all current generation games. Physx does put it's wear on the card, but that being said, Physx will put a huge workout on pretty much any nVidia card that has Physx built into it. Sometime in the future I will revisit benchmarking Mafia II and Batman: Arkham City to see how much the FPS improves with Physx turned on and the GTX 460 set as a dedicated Physx card. I expect the FPS to improve simply because the GTX 670 will have tons of breathing room to work on everything but Physx processing, but I'll cover those two games as well as some other Physx enabled games to see just how much the dedicated GTX 460 Physx card really affects the FPS.