AMD Radeon RX 6800 XT with Overclocked overlayed

Introduction

AMD launched the Radeon RX 6000 series of video cards on November 18th, 2020, also known as Big Navi.  At the very high-end is the $999 Radeon RX 6900 XT.  Right below that is the $649 Radeon RX 6800 XT and then below that is the $579 Radeon RX 6800.  The video card that compares to NVIDIA’s most recent video card launch, is the Radeon RX 6800 XT at $649.  It competes directly with NVIDIA’s GeForce RTX 3080.  At only $50 apart, with the advantage in price going to the Radeon RX 6800 XT, squeezing every bit of performance out of the video card is important.

We aren’t going to go into detail on either video card today, we’ve done that in each launch review for both video cards.  You can read our launch review on the Radeon RX 6800 XT as well as the GeForce RTX 3080.  Today we are interested in finding out how high we can overclock the Radeon RX 6800 XT, and how much performance benefit we get from that.  We will then compare that performance to the GeForce RTX 3080 Founders Edition to see how they compare.  If you wish to see how well the GeForce RTX 3080 Founders Edition overclocked, we have a separate review on that as well.

Before we dive in though, let’s do a quick brief on the specifications and clock speeds the Radeon RX 6800 XT runs at.  The $649 Radeon RX 6800 XT is based on AMD’s new RDNA2 architecture and manufactured on TSMC 7nm.  It has 72 Compute Units, 72 Ray Accelerators, 4,608 Stream Processors, 128 ROPs, and 128MB of AMD Infinity Cache.  The Game GPU Clock is set at 2015MHz and the Boost GPU Clock (up to) is 2250MHz.  It has 16GB of GDDR6 at 16GHz on a 256-bit memory bus providing 512GB/s of memory bandwidth.  The Board Power is rated at 300W.

How to Overclock the Radeon RX 6800 XT

AMD Radeon RX 6800 XT Overclocking GPUz Default

How do you overclock the AMD Radeon RX 6800 XT video card?  Before we find out how high it can overclock, we first need to find out what it runs at by default.  AMD has two different clock domains.  There is the Game Clock and the “Max” Boost Clock that the GPU is capable of boosting to in certain workloads.  For the most part, the Game Clock of 2015MHz is what you should expect while gaming.

AMD Radeon RX 6800 XT Overclocking GPUz Sensor Data Default

However, because of the dynamic nature of the GPU frequency, it can clock up to 2250MHz while gaming.  Basically, the clock speed will be anywhere between 2015-2250MHz while gaming.  It will jump all over the place based on many sensor inputs about thermals, temp, power, voltage, and even the level of workload and types of workload that is stressing different parts of the GPU.

In our review of the Radeon RX 6800 XT, we actually looked at what the real-world GPU frequency is while gaming.  You can see in the graph on this page how that turned out.  The GPU frequency did jump all over the place, but it ended up being much higher than we thought.  It actually remained well above the 2015MHz clock domain at all times.  In fact, it was even in excess of the 2250MHz clock domain!  It started very high at 2300MHz and then dropped down after a minute to mostly around that or upwards of that 2250MHz mark.  The average clock speed turned out to be 2265MHz. 

Therefore, that is our starting baseline for overclocking.  You can also see in GPUz that at default Voltage was at 1.150V, GPU Temp was 75c at 46% fan speed on automatic and GPU Chip Power Draw was 279W.  That’s the default operation we are comparing the overclock with.

Highest Overclock on Radeon RX 6800 XT

Overclocking the AMD Radeon RX 6800 XT is limited at the moment in what software programs support overclocking.  At this time, MSI Afterburner will not work.  We have to rely on the features available in the AMD Radeon Software itself for overclocking.  Thankfully, these days that support and features are very comprehensive.  It allows customizing power target, fan speed, GPU frequency, and memory frequency and timings.  It does not, however, allow us to raise the Voltage at the moment on the Radeon RX 6800 XT GPU.  Our maximum overclock is shown in the screenshot below.

AMD Radeon RX 6800 XT Overclocking AMD Radeon Software Performance Tuning Overclock

The first thing to note from the AMD Performance Tuning is that we were able to manually overclock.  We were able to increase the Power Tuning up 15%.  This is essential, and the most we could raise it.  Second, we were able to turn on advanced control for the fan speed and raise it to 100% at all times and also turn off Zero RPM fan feature.  We set the fan at 100% to ensure the highest overclock.

GPU Frequency

Next up is the GPU frequency, this is where it gets a little complex.  There are two options, a Min Frequency and a Max Frequency.  Just raising the Max Frequency will not result in an overclock, or a very high one.  You need to actually raise the Min Frequency, but you don’t want to set them to the same value, that will be unstable.  We found keeping them about 200MHz apart worked best for us.  There isn’t really a science here, it is all trial and error.  In our trial and error, which took a lot of time to finesse, we found the highest overclock for our card was 2400MHz on the Min Frequency and 2600MHz on the Max Frequency.  This resulted in positive performance increases in every game, at 4K and with Ray Tracing. 

We tried to set it a bit higher to 2500MHz Min Frequency and 2700MHz Max Frequency, and while this setting actually worked, it actually lowered game performance at 4K and with Ray Tracing.  So, when we backed it down to 2400MHz-2600MHz then we saw the most positive performance increases.  We’ll see what this turns out in real-world frequency below.  As for the GPU Voltage, we were not able to increase it at all, it was locked at a maximum of 1150mV. 

Memory

On the memory side of things, Performance Tuning only lets us set the memory up to a maximum of 2150MHz.  The default speed is 2000MHz.  The slider simply would not go any higher, so the memory is locked at a maximum of 2150MHz.  This results in a maximum memory frequency of 17.2GHz versus the default of 16GHz.  That boosts the memory bandwidth up to 550GB/s up from the default of 512GB/s.  We think the memory might be capable of more, but it’s simply software locked.

Overclock Comparison Graph

AMD Radeon RX 6800 XT Overclocking GPU Frequency Comparison Graph

Here is the graph that shows how the overclock compares to the default frequency.  The orange line is the overclock, blue is the default.  At default, it’s all over the place, and our overclock definitely stabilizes the frequency and makes it more consistent.  The overclock stays between 2400MHz-2530MHz.  It maxes out at 2550MHz briefly.  The average default frequency in this new run is 2257MHz.  The average of the overclock frequency is now 2504MHz.  That’s an 11% overclock on the GPU frequency.  Then with the memory, the memory is a smaller 8% memory overclock.

GPUz

AMD Radeon RX 6800 XT Overclocking Overclocked GPUz
AMD Radeon RX 6800 XT Overclocking Overclocked GPUz Sensor Data

According to GPUz the Voltage remains at 1.150V, GPU temperature is 61c and GPU Temperature Hot Spot is 104c at 90% fan speed (for some reason it didn’t quite get up to 100% even though we set it.)  The GPU Chip Power Draw increases to 300W.

Brent Justice

Brent Justice has been reviewing computer components for 20+ years, educated in the art and method of the computer hardware review he brings experience, knowledge, and hands-on testing with a gamer oriented...

Join the Conversation

11 Comments

  1. Good tip on adjusting the minimum frequency. I’ll have to try that on my 6800. I set the max freq to 2450 with an undervolt to 975mV, but didn’t touch the minimum.
  2. What is the performance gain of adjusting memory independent of the GPU clock? Is the card bandwidth starved, so we should be looking for AIB cards with higher mem frequency, or does it not matter much?
  3. What is the performance gain of adjusting memory independent of the GPU clock? Is the card bandwidth starved, so we should be looking for AIB cards with higher mem frequency, or does it not matter much?

    I don’t think big navi is bandwith starved thanks to the Infinity Cache.

  4. I don’t think big navi is bandwith starved thanks to the Infinity Cache.

    In which case, does keeping the memory clock lower allow any extra room on the GPU side? Could you even underclock the memory and see gains as you free up a little power draw for the GPU?

  5. The video card is definitely more engine starved than memory. It benefits more from GPU frequency increase in my experience. Especially in the case of Ray Tracing.
  6. This looks like a first take, on the Nvidia side, depending upon card, upping the fan speed can decrease GPU clocks due to having less available power to the GPU. Wonder if 100% fan speed here was taking some power away from the GPU? Also same with memory.

    Power requirements for fans follow roughly, double the speed, power goes up 8x, of course fan efficiency curves can affect that ratio a lot. Something to look at maybe, on Nvidia more constrain power wise cards, fan speed does affect GPU clock rate negatively at times when upped (the benefit of cooling sometimes does not outweigh the loss of available power to the GPU in other words). Some of the Nvidia users are decoupling the card fans and using board power to supply them and getting a significant bump in the GPU clock speed.

    A lot of variations from benchmarks, particularly from site to site, is from ambient temperature, where fans don’t have to turn as fast or not, due to these higher power cards, they require beefier more powerful fans, affecting power to the GPU more than previous cards. So better coolers, more efficient fans could give 15w+ more to the GPU in other words. Getting an amp reading from the fan from 50% to 100% fan speed would be interesting data to consider and for comparison sakes from ref to AIB to AIB.

  7. Wonder if 100% fan speed here was taking some power away from the GPU?

    Industrial 120mm (loud and powerful) fans are upwards of 1W, and even if you have three of them on a cooler, most fan headers are only rated for 2W, as most computer fans are much less than 0.5W. Not sure what’s on a GPU header but can nearly guarantee it’s single digit wattage.

    I know on overclocks you want all the power you can get, but I have a hard time believing it’s the difference in 75% fan and 100% fan speed.

    But hey, if you have some good evidence otherwise I’m willing to believe, it is 2020 after all and anything can happen.

  8. Industrial 120mm (loud and powerful) fans are upwards of 1W, and even if you have three of them on a cooler, most fan headers are only rated for 2W, as most computer fans are much less than 0.5W. Not sure what’s on a GPU header but can nearly guarantee it’s single digit wattage.

    I know on overclocks you want all the power you can get, but I have a hard time believing it’s the difference in 75% fan and 100% fan speed.

    But hey, if you have some good evidence otherwise I’m willing to believe, it is 2020 after all and anything can happen.

    I think you got amps and watts mixed up. Fans at full speed, the higher spinning ones will take way more than 1W. If we have the fan data for the 6800 XT, power rating then that would be the easiest way. Measuring load either directly with amp meter or setting up the computer with a constant load as in all case fans constant speed, constant CPU power etc. then measuring the change in power from wall, by ramping fans from off to max and correcting for power supply efficiency maybe would give a rough measurement.

    Anyways here is a replacement fan for a 1070Ti 1080Ti Zotac Mini: It requires .46amps, times that by 12v at max speed, 5.5w. I would suspect the bigger fans on the newer cards may exceed that power of 5.5w for each fan. If 6w each at 100%, 18w total when at 100%, at 50% fan speed it would be roughly 1/8 the power or around 2w.

    Once I get my 3090, I can setup a loop like with 3dMark or something with a consistent GPU load, manually adjust fans from 50% to 100% and watch clock frequency, since there would be a time delay before the cooler heats up or cools down you should be able to catch any additional power or loss to the GPU with frequency change. Also of note, those that put water blocks on their cards can see big jumps in clock speeds for the 6800 XT, it maybe more than just cooling, you also have more available power to the GPU as in 10w-20w more without having to power the fans.

  9. OK, I was able to find a relatively easy way to see effect of fan speed and power on a video card. Article used 100% fan speed while OCing, question is will the increase power from the fans running at 100% have an effect on the outcome? I cannot answer that dealing with the 6800 XT, but was able to test with an EVGA 1080 Ti SC Black.

    EVGA ICX has extra sensors which also helps quantify any power increase by ramping up the fan speed. The EVGA 1080 TI SC Black has two 90mm fans which at 50% turn at 1840 RPM and at 100% 3699 RPM.

    With card basically idle and relatively steady clock with fans at 0 RPM, Board Power Draw was 59.2w, GPU-Z is using the video card power sensors:

    View attachment 712

    At 50% fan speed, 59.9w board power:

    View attachment 713

    At 100%, 64.4w board power:

    View attachment 714

    With two fans, from 0% to 100%, power changed 5.2w, from 50% to 100% power changed 4.5w where most of the power increase came from going to 100% from 50%. This is expected from how fans behave with power, Pwr is roughly proportional to (RPM2/RPM1)^3

    Anyways I am having a hard time concluding this would have a significant impact to power to the GPU, with a 3 90mm fan configuration using the same EVGA fans as is on this 1080 Ti, we would be looking at 8w max power consumption, which out of let say 300w OC is only 3% of the power budget. Are the 6800 XT fans much more power hungry or on other GPU’s?

    I did a test using Aida 64 GPU stress test, which put a constant load on the GPU, raised the clock frequency until it did not change anymore which I am thinking indicated power limited. Changed fan speed from 100% to 0% and saw 0 percent change in clock speed. I tried 3dMark but as soon as I lost focus with 3dMark it would stop working. At least with my 1080 Ti fan speed appears to have very little effect from power usage of them.

    On EVGA forums it was discussed that powering the fans on MB will give extra power to GPU and increase max clock speed. While true, not sure it will be that beneficial or significant, especially if fans are like 60% or less drawing less than 2w. Once I get the 3090, I will explore this more.

Leave a comment