Overclocking the EVGA 2080 Ti Video Cards on air
I wanted to overclock the video cards while they are still air cooled just to see what they would do with the stock air cooling, and then be able to compare how much more I can squeeze out of them under water.
Plus since there are so many reports of these 2080 Ti cards dying an early death, I also wanted to give these cards a thorough testing before installing the waterblcocks. While I was overclocking the CPU using Real Bench that program also stresses the video cards, so they got a good three weeks of heavy use there before I started with the video card overclocking.
Over the years I have used both the MSI Afterburner software which is probably the most popular program to overclock video cards with, and the EVGA Precision XOC software. The Afterburner software is easier to use, but I liked the hardware monitoring better on the Precision XOC software.
EVGA has an all new overclocking program called Precision X1, and that is what I’ll be using for overclocking these cards. It has a completely new, much easier to use interface, and it also has a “VF Curve Tuner” that automatically scans your card to determine how much of an overclock you are likely to be able to achieve. It also has control of the illumination of the card’s air cooler and the EVGA NVLink if you have one.
I am going to test each card separately first, and then together with the NVLink. Like CPUs every video card/GPU will be a little different, and I want to put the best card in slot #1 on the motherboard.
First I’ll see what the VF Curve Tuner does and then I’ll over clock both the core clock and memory clock manually. The VF Curve Tuner only tests the clock speed, and not the memory overclock, so the memory has to be tuned manually anyway.
For testing purposes I’ll be using the Heaven benchmark. This is my favorite benchmark program for overclocking video cards for several reasons. First the program will continuously loop until you hit the “Benchmark” button, and also because this benchmark has been around for years now, and I have stored benchmark scores from many different video cards and multi-card set ups that I’ve had over the years. It’s always fun to see how new hardware compares with what you have previously used!
After opening the program you click on the left arrow along the left side, the lower portion of the program slides over to the VF Curve Tuner. Here all the sliders on the top are at the stock settings.
I moved the VOLTAGE slider all the way to the max as well as the POWER slider. I’m not sure if it makes a difference while the VF Curve Tuner is running because I had the fans on max too, and as soon as I started the VF Curve Tuner the fan speeds went way down.
Anyway just push the SCAN button on the right side and the VF Curve Tuner will run. It takes about 20 minutes to complete. When it completes it will put the score in the box where the zero is now. I didn’t grab a screen shot with a completed score, but basically whatever the score is you should be able to put in the CLOCK box to add to your core overclock.
Using the arrow buttons here is the fan curve screen.
There is also a temp tuner.
Since both cards were already installed I ran the VF Curve Tuner first, and it came up with a score of 158, which I thought was really good. Remember that number because I’ll be testing it again later.
I pulled out video card #2 so I could test the first card by itself. Now the the Precision X1 software would not load properly, and 3dmark software would not run any benchmarks because it was not able to detect the system information. You would think just removing one card that everything should have still worked fine!
To make a long story short, after much frustration uninstalling and reinstalling the Precision X1 software, the lighting component that goes along with it, a registry edit I found on the EVGA forum, and a complete uninstall of the video card drivers with DDU and reinstall of the video card drivers, I finally got it all to work again
I’ll start by overclocking the memory clock and leaving the core clock stock. Then I’ll put the memory clock back to stock and overclock the core clock by itself. Then I will combine them, usually you will have to back down one or both of the max settings you got individually to maintain stability when they are both overclocked together.
With the VOLTAGE and POWER sliders already on max, I also turn the fans to maximum speed for all of this testing. Then open up Heaven and just let it loop through.
Overclocking Video Card #1
After running the VF Curve Tuner on video card #1 it scores 104. I started with the memory overclock by entering 500 in the Memory Clock box and hitting the apply button at the bottom. Then just watch the Heaven benchmark, usually you can see pixilation once you get too high on the memory overclock.
Then just go up 100 at a time until you start to see the pixilation. +900 would run fine, +1000 would not, so I ended up getting +930 on the memory overclock with no pixilation.
Now to find my core overclock, after resetting the memory back to zero, and entering 120 in the GPU Clock box and then the apply button, it crashed right away. With the clock speed too high usually it will lock up the Heaven software, it can even lock up the whole system. Most of the time you can just restart Heaven, lower the clock and keep going.
I found that +110 on the core clock would work fine. I’m only testing for about 15 minutes on Heaven to check for stability here because I’ll be doing this all again once the waterblocks are installed, and then I’ll let it run much longer to dial in the final settings.
Now to do the memory and core clock together I start with +900 on the memory and +110 on the clock… crash. With +900 and +105 it still crashed, but +900 and +100 ran fine.
Next it's time to run the actual benchmark on Heaven, Superposition 4k, Fire Strike, Fire Strike Extreme, Fire Strike Ultra, Time Spy and finally Time Spy Extreme benchmarks. Sometimes you will crash one of these benchmarks, which obviously means it’s not stable, and will probably crash when gaming. The +900 and +100 make it through all these benchmarks. I put together a spread sheet of all the benchmark scores below.
Overclocking Video Card #2
I took the first video card out and put video card #2 in the first motherboard slot. Thankfully everything still works. I ran the VF Curve Tuner again, this card scores 136. This is quite a bit better than the first card!
For the second card since I already have a good feel from the first card, I start with the memory at +900 and work on the core clock. Starting with +140 it crashed, but +135 was stable. It seems that the VF Curve Tuner was pretty close here!
Now to see if I can get the memory to go even higher. Going right to +1000 caused an immediate crash. +960 ran good and I worked it all the way up to +990 on the memory and it was still good, so I tried +1000 again and this time it worked ok
Here’s two screen shots I took right after starting Heaven with the memory at +990 and +1000. The core clock only stays above 2000 for a short time, then bounces around between 1950 and 1980.
Then I ran all the benchmarks again, Heaven passed with the +1000 and +135 settings, but crashed on Superposition. I have heard that the Superposition benchmark will trip up a video card overclock easier than most other benchmarks.
I decided to move the memory overclock back to +900 like the first card and leave the core clock at +135. This way I can compare the benchmarks of both cards with the same memory overclock, and see how much difference just the core clock makes. Those setting passed all the previously mentioned benchmarks.
Notice how close both cards were manually overclocking the core clock to what the actual score was from the VF Curve Tuner. It appears that the VF Curve Tuner is good to use to give you a rough idea of what you can expect, and gives you a great place to start your overclock testing.
Benchmarking Both Video Cards with NVLink bridge
Leaving the video card #2 in the first slot I reinstalled the Card #1 and put the NVLink bridge on, and ran the VF Curve Tuner one more time just for the heck of it. Interestingly it scored only 109 this time compared with the prior score of 136. The only difference is that the two cards are reversed in the two slots.
I ran all the benchmarks through again with the +900 on the memory and +135 on the core clock and it passed all the benchmarks. I wasn’t sure if it would because of card #1 only really stable with +100 on the core.
Here’s a chart I put together comparing Video Card #1 at +100 +900 with Video Card #2 at +135 +900, and with both cards with the NVLink bridge at +135 +900. Once I retest everything with water cooling I can simply add to this chart.
The only abnormality is that Video Card #2 scores lower on the Heaven benchmark than Card #1. I reran that Heaven benchmark several times since Card #2 was still in place too just to be sure, but did not bother to retest Card #1. All the other benchmarks are higher on Video Card #2 as I would expect due to the higher core clock.
Finally here’s the screen shots for both cards together on the Firestrike and Time Spy benchmarks.
Now it's time to install some waterblocks