Power Consumption and Overclocking

If you remember back to our Core i9-7900X review, I was critical of the power consumption of the Skylake-X processor both because of its high power usage, but also the inaccuracy of Intel’s rated 140 watt TDP. The power consumption of the Ryzen Threadripper CPUs is 180 watts, 40 watts higher than the 7900X. How did it stand up?

In Cinebench, the 1950X uses 30 watts more power than the Core i9-7900X, pulling full system power at 268 watts. Compare that to the Ryzen 7 1800X at 155 watts (113 watts higher) or the Core i7-7700K at 122 watts (143 watts higher) and you cannot deny that the Threadripper platform is going to be a power-hungry beast. Idle power consumption is high as well – 95 watts on the 1950X is 40 watts higher than the 7900X and nearly 60 watts higher than the 1800X! We’ll see if BIOS tweaks can adjust this, or if our ASUS Zenith Extreme board just happens to be power hungry as well.

One of the messaging angles you will likely find in other reviews centers around power consumption per core. It’s an interesting metric even when amortized across the full system as we did below:

  • Threadripper 1950X: 16.75 watts
  • Core i9-7900X: 23.8 watts

The point of this comparison to show the advantage that AMD has on a per-core basis, but in truth, I don’t believe this metric is worth much today. When Intel releases its 12-core and higher Skylake-X parts in the near future, I do not expect power draw to continue to scale UP at 20-something watts per core; it’s just not possible. Similarly, the Threadripper 1920X has a rating of 21.75 watts/core in our testing, much higher than the 1950X.

Instead, performance per watt is the metric that consumers that even care about power consumption should be looking at. Can the Threadripper 1950X offer more performance in the applications that are important to them than the added power draw it requires over the Core i9-7900X? Our test setup shows that the 1950X uses ~12% more power than the 7900X in Cinebench, but it results in a 37% better score, netting an improved performance per watt metric. If performance on the 1950X is anything under 12% faster than the 7900X, then it is a net-negative on performance per watt.

Power per core matters in the data center market, but it isn’t a useful metric for consumers that are going with single socket solutions in workstations or enthusiast builds.

Overclocking Threadripper

Overclocking with Ryzen processors has been a fairly fixed art since release. The Ryzen 7 was able to hit somewhere around 4.0-4.2 GHz with all cores active and we have seen that trend continue through Ryzen 5 and Ryzen 3, for the most part. Because the Threadripper CPUs are using two Zen cores, of which AMD claims are in the top 5% of the binning from the fab, we expected to see similar levels of scalability. Our expectations were not far off.

By simply setting the Vcore to 1.395v (which is higher than we'd like), I was able to get our 1950X to run at 4.0 GHz completely stable through a mess of workloads.

Performance scales by 12-13% in Blender, POV-Ray, and Cinebench in our testing.

What does not scale by 13% is power draw. 

WOW. Under a full load, the overclocked Threadripper system uses an additional 140 watts of power, or 52%! Considering that is  all power being pulled from the CPU directly, it means we are pushing nearly 300 watts of total consumer power from the CPU. Need more proof?

If you are going to be overclocking the Threadripper platform, make sure you have a significant cooler on your CPU.

« PreviousNext »