Power, Temp and 7nm

We wanted to editorialize our thoughts about power, temperature, process node technology and architecture.  These are our opinions, please take them as such.  Though the AMD Radeon RX 5700 series of video cards are based on the wonderful 7nm technology, the process node isn’t always the most important thing.  The architecture still plays a powerful part of whether a GPU is power efficient or not. 

Trying to get the most work done, with the least amount of power is a tremendous challenge.  7nm helps Navi, big time.  The RDNA architecture helps Navi, big time.  This is by far AMD’s biggest leap in efficiency.  However, it still has room for improvement on the architecture side of things.  AMD has it licked in the process node department, 7nm is awesome for them.  Consider that it takes that process node for AMD to achieve the clock speeds and performance AMD has with the Radeon RX 5700 series.

To understand what we mean by this you have to look back at NVIDIA and what NVIDIA has done in terms of efficiency.  For the longest time the industry was stuck at 28nm, waiting for a process shrink that just wasn’t coming in time.  AMD struggled with this, its GPUs were power hungry, hot and thermal throttled or had loud fans to keep them cool.  Remember the whole “Uber” performance mode era? 

NVIDIA managed to make its GPUs at the time way more efficient on that process node.  This was the Maxwell line of GPUs.  These GPUs at 28nm shocked everyone at how much more efficient NVIDIA could make the architecture at 28nm.  It really was mind blowing, and showed that even though the process node had stayed the same NVIDIA was able to get so much more out of it and had.  At the time NVIDIA had the best performance per Watt.  NVIDIA continued this trend with Pascal at 16nm, and now with Turing at 12nm. 

At 12nm NVIDIA is still generally more power efficient than the Radeon RX 5700 XT at 7nm.  In our testing the Radeon RX 5700 XT demanded the most power.  Even at equal to RTX 2070 FE performance, it was using more power to do so, so the RTX 2070 FE is more power efficient, even though it’s 12nm versus 7nm.  This is the Turing architecture at work.  Imagine what NVIDIA will be able to do at 7nm. 

The other component is clock speed.  One thing that held back AMD GPUs of the past at 28nm and so forth was clock speed.  AMD just couldn’t get the clock speed high enough with its GCN architecture.  However, NVIDIA was able to do so.  With Maxwell NVIDIA had frequencies operating near 2GHz.  Overclockers could certainly get the frequency to those levels, and that was on 28nm!  With Pascal and Turing those clock speeds have remained high.  In order for AMD to get clock speeds that high it has had to shrink and shrink the process node just to get there. 

Finally, at 7nm and with RDNA architecture AMD is now able to give us very high clock speeds on the Radeon RX 5700 series GPUs.  In our testing the Radeon RX 5700 XT was gaming at an average of 1854MHz, and peaked at 1945MHz.  That’s really great!  However, clock speeds this high weren’t possible until 7nm and the RDNA architecture. 

Even with RDNA there is room for improvement.  If NVIDIA can pull of 2GHz on its architecture at 12/16nm, imagine what it will be at with even more efficient architectures at 7nm.  AMD is just now reaching upwards closer to 2GHz, but NVIDIA’s been there, and by the time its on 7nm it may surpass that. 

These are just interesting facts to consider.  Don’t get us wrong, the RDNA architecture is the biggest leap AMD has made yet!  It’s great, we love it, it’s huge for them, it’s a great architecture it looks like.  A lot of the issues of GCN are gone.  We just ultimately don’t know how scalable it is yet.  Maybe AMD has more up its sleeve, maybe AMD will surprise us all and we will all be blown away. 

Right now, though, the Radeon RX 5700 XT and RX 5700 run a little hot for our tastes, and the RX 5700 XT a little more power hungry than we thought it would be.  The clock speed results are impressive though, but we don’t quite know the headroom just yet.  Perhaps custom video cards with custom cooling will be a hardware enthusiasts delight, that is yet to be seen. 

The real gem is the Radeon RX 5700.  Its power efficiency is great, it is more power efficient, with more performance per Watt than the GeForce RTX 2060.  Custom video cards based around it must be good, and this may be the video card that has the best potential for overclocking out of them all.  If there is some good headroom there, we can see the Radeon RX 5700 surpassing the RTX 2060 SUPER in performance.  We will just have to see.  We will also just have to see if AMD will scale Navi upwards or not with fastest video cards to compete with the GeForce RTX 2080 and 2080 Ti.  If RDNA can be scaled up, we look forward to it.