GPU Clock – The Real Story

When I previewed the Radeon R9 Nano (the second time) I theorized that even though AMD was claiming the GPU was rated at “up to” 1.0 GHz clock  rate – just 50 MHz slower than the Fury X – it would likely not hit anywhere near that in real-world gameplay. It turns out that I was correct and the real clock speed of the AMD R9 Nano varies quite a bit; from game to game as well from resolution to resolution.

To test this, I captured clock speeds of the R9 Nano using GPU-Z during all of my benchmark tests, something that I don’t usually do as it adds a couple of extra steps to an already elongated process.

Let’s take a look at the clock speeds of the R9 Nano when running through 6 different games at 2560×1440.

nano-clock-25x14.png

I have zoomed in on the graph some (non-zero starting point on the scale) to demonstrate the variability a bit more directly. Also, I have taken a 90 second interval from the middle of our testing and game play for the same purpose. There are a lot of lines on this graph, so bear with me.

Immediately you can see that the clock speed of the R9 Nano can waver dramatically from game to game, hitting as high as 1000 MHz in GRID 2 and as low as ~825 MHz in Bioshock Infinite. Even when looking at data in the same game, for example Grand Theft Auto V, clock speed moves between nearly 1000 MHz and ~920 Mhz regularly.

As mentioned on the previous page and in the previous story, AMD is doing this balancing act of clock speed (and voltage) in order to maintain the card at the 175 watt TDP level that it is marketed and sold at. If an OEM or an end user designs a system to properly dissipate 175 watts of heat, the card must maintain that output level even if that means it’s at the expense of clock speed, and thus, performance. Clearly some games put a different strain on the system in order to get these different clock speed results from the R9 Nano: GRID 2 is less GPU bottlenecked and thus it can run at a higher clock speed while Bioshock Infinite is using the shader resources heavily, keeping clock speeds lower.

What changes if we bump this up to 4K?

nano-clock-4k.png

First, none of the 6 games here is able to get close to 1000 MHz this time and in general we see a drop in the clock speed result across the board. Bioshock Infinite is now hovering in the 825 MHz range while GRID 2 sticks to the 900 MHz mark.

nano-clock-bar.png

Looking at the above results in terms of average clock speeds over that same 90 second span, you get a better idea of what’s happening. At 2560×1440, both GTA V and GRID 2 have the highest average clock rate while Bioshock is only averaging 863 MHz. When we jack up the resolution to 4K though the averages drop through all of our games’ testing, as I would expect with the additional load on the GPU. At 4K, only GTA V averages higher than 900 MHz with Bioshock as the lowest clocked game, only getting 830 MHz or so.

Clearly the “up to” rating on the R9 Nano is a bit misleading. In the past AMD has marketed clock speeds as “up to” some number but in every other case that I have seen with Radeon cards the GPUs were hitting that maximum GPU clock nearly the entire time. (The lone exception was the initial release of the R9 290X/290 and their reference coolers.) Because of that, the “up to” clock speeds have been mostly ignored. That can’t continue going forward, clearly.

But how does the NVIDIA counterpart to the R9 Nano behave? The ASUS GTX 970 DC Mini is the highest performing GeForce card with a form factor similar to that of the Nano, though as you’ll see later, it can’t hold up to this new Radeon card’s performance levels.

nano-clock-gtx970.png

Keeping in mind that higher clock speeds do not mean higher performance when comparing different architectures, we are looking at this graph only to gauge the clock speed consistency between the two offerings. The GTX 970 is able to maintain a much more static clock rate than the R9 Nano in GTA V and GF4 (two random samples). Why does that matter? It is possible that with highly varying clock rates a GPU could run into issues of frame pacing even in a single GPU environment. It’s something to watch for as we get into our Frame Rating performance testing later in the review.

« PreviousNext »