GTX 780 Graphics Card Dimensions:
Thermal and Power Specs:
95C Maximum GPU Temperature (in C)
250 WGraphics Card Power (W)
600 WMinimum Recommended System Power (W)
One 8-pin and one 6-pin
On the previous page we saw the result of the real-world gaming GPU frequency with no manipulation. We found that the default frequency was running from 862-875MHz after 30 minutes of gaming in each game. This put the frequency right at the GPU Boost frequency for the video card. However, we feel this can go higher if we just raise the fan speed. Therefore, in the tests below, we are going to maximize the fan speed on the GeForce GTX TITAN. That is all that is done, nothing else is touched
Above you can see that we have manually set the fan speed to the highest value of 85%. This is all that we have done, increased fan speed. This of course creates a lot of noise, and is loud compared to TITAN’s default near-silent configuration. However, will this sacrifice of decibels allow the real-world frequency to rise? Let's look below at the photographs again after 30 minutes of gaming.
Yes, increasing the fan speed, and keeping the GPU cooler (even though it still has the 80c limit) makes for a much better real-world frequency. Crysis 3 settled to 993MHz at 64c. Tomb Raider settled to 993MHz at 62c. Far Cry 3 settled to 993MHz at 64c. Sleeping Dogs settled at 993MHz at 62c.
By simply setting the fan speed higher the frequency has increased by 15%. It is consistent across all four games, none of these varied. At 993MHz the real-world frequency while gaming is now above the GPU Boost frequency, without overclocking. This is a 131MHz higher clock speed! That is a nice "free" boost, just by increasing the fan speed.
Above, we have increased all three values. The maximum Power Target on this card is 106%, which seems a bit low to us. However, the fan is maxed, the Temp Target is maxed and the Power Target is maxed.
Our results didn't turn out like we thought these would. We thought for sure the clocks go higher, but not near what we expected. All games were operating at 993MHz, exactly the same as just turning the fan speed up to maximum. This just shows how important it is for the TITAN to keep the GPU as cool as possible for the best clock speeds. We can conclude from this that it is important to have a good robust cooling system for TITAN, and a decent fan speed, to keep the GPU cool enough. This will result in a 130MHz+ frequency over the GPU Boost clock speed, while gaming, before even overclocking.
On the next page, we now get to overclocking by raising the GPU Offset with everything else maxed out, fan, power, and temp. We will also raise the voltage as well in the last section.
GeForce GTX TITAN Overclocking GPU Offset
On the previous pages we found out that our NVIDIA GeForce GTX TITAN runs at 862-875MHz in games at default settings. This puts it right at the default stock GPU Boost clock frequency, no further. We also found out that by simply raising the fan speed we were able to get a 131MHz boost to 993MHz in games. A simple fan speed improvement, creating a cooler GPU, allowed the GPU to boost to higher levels, providing a nice boost to the frequency without overclocking.
Of course, what we all want to know is how high the GPU can really be pushed once we factor in GPU Offset. So for these tests below we have the fan speed maxed, the temp target maxed, the power target maxed, and we are simply raising the GPU Offset as far as we can to increase the clock speed further. We have to watch out for diminishing performance. The GPU can throttle itself in performance if we raise the frequency too high. We have to make sure every frequency increase also gives us a boost to performance. Once we reach the plateau where we get no more performance, and it's not crashing of course, that is our limit. To do this we tested in all four games for at least an hour to make sure it was stable in all four games.
We managed to maximize the GPU Offset to +150. Any further, even to just +155, or +160, cause games to crash. At +150 this boosted the in-game frequencies higher than 1.1GHz. We have taken pictures of each game below to show you the frequency in each game at +150 GPU Offset.
In Crysis 3 the real-world frequency settled at 1110MHz. In Tomb Raider it settled at 1137MHz. In Far Cry 3 and Sleeping Dogs it also settled at 1137MHz. The most demanding game was Crysis 3. The frequency would switch between 1110MHz and 1137MHz while gaming, depending on the load.
GeForce GTX TITAN Overclocking with Voltage Tweaking
Unique to GPU Boost 2.0 is the ability to set the voltage higher. However, we found out that it isn't by much, and in fact isn't enough to really make a real-world difference unfortunately. Let's look at the option below and what it means.
After clicking the "Voltage and Tweaking" button you get a separate window with the voltage. You can raise the voltage only by +38 or 38mV. In all of our gameplay, we experienced a maximum voltage set automatically by GPU Boost of 1.162V. If you look through all the screenshots on the previous pages you will 1.162V as the settled voltage. By increasing +38mV we are telling the GPU it can now go up to 1.200V. This is a small change in the grand scheme of things.
What we experienced, however, is that even by setting +38mV to 1.2V, the GPU would not stay there while gaming. GPU Boost was still changing the voltage under that, sometimes 1.185V, or sometimes back down to 1.162V or even lower. It wasn't remaining at 1.2V giving us the best chance for highest clock speeds. In order to lock in 1.2V we had to choose the "K-Boost" button and enable it. Unfortunately, this also locks in the voltage in 2D mode, increasing power and temperature.
Now, here is the interesting part, even while in K-Boost mode, we still saw the voltage automatically downgrading at times. We experienced 1.2V more often with K-Boost enabled, it tried to keep it there, but there were times when it would drop for a few seconds, depending on the scene load. In the more demanding games like Crysis 3 we never saw sustained 1.2V, it was bouncing down to 1.162V, or lower, and then back up to 1.2V. It wasn't stable, and therefore neither was the frequency of the GPU.
Did we achieve a higher overclock with the voltage increase? The answer is no. We were still not able to set +155 or +160 on the GPU Offset option. This caused games to crash. What's more, with the voltage increased, we were now also experiencing crashing at the +150 GPU Offset that was stable without the voltage increase. We did see that the frequency increased slightly in the less demanding games, up to 1150MHz. However, we were still getting game crashing, so this setting was not usable. We literally had no success with increasing the voltage, and felt that it is best to leave GPU Boost up to deciding that in the end. Therefore, our maximum GPU overclock of 1137MHz remained the highest stable frequency we got.
Kahless wrote:I have 4X HBAO. And 120% resolution scaling. Which they said is like a "super sampling" I tried 200% resolution scaling but I didn't care for the performance hit and it didn't really make it look that much better. They said its like running the game in 4K and shrinking the image down to fit at 2560x1440.
Hammer_Time wrote:I see, well not sure why you are getting such high frame rates in BF4 ( compared to others who own same GTX 780 card ), but why worry?? Enjoy!! ( must be your 2133 Mhz ram methinks! )
Hammer_Time wrote: I will always be a "single" videocard gamer, and avoid any SLi or CrossFire issues...
We used the BF4 Settings Editor for our configuration -- the Open Beta client refused to save in-game graphics settings changes normally.
We tested using FRAPS on a 64-player server in Buffalo, NY (geographically the closest server to my own home). Game frame rates were measured over the course of seven minutes.
We've also graphed the frame time latencies using FRAFS, a handy tool for displaying FRAPS frametime results. If you aren't familiar with frame timing, this is a metric of how long it takes to draw each frame. It shows variation within a single second -- something that FPS, by its nature, fails to catch. And the differences it captures are exceedingly useful. For example, here's the traditional FPS results (min, max, average) for the GTX 770 and the Radeon HD 7970 over an entire run.
Users browsing this forum: No registered users and 3 guests