Power Consumption Tests

Our watt draw tests look at GPUs, CPUs, briefly looks at additional watt draw from GPU overclocking, and then provides some templated system builds for power tests.

GPU testing looks at full load. CPU testing looks at heavy encode workloads (100% load) and idle power consumption. The templated systems go through “full load” (unrealistic load level for a gamer – 100% CPU, 100% GPU), “gaming” (DiRT benchmark on loop), idle, and “synthetic load” test passes, defined more below.

Saturation of our test PSU does affect watt draw in the form of efficiency. SLI/CrossFire configurations, despite consuming the most power, will trend toward peak efficiency on the high-wattage unit. Small adjustments would have to be made to extrapolate for precise watt draw / builds. See above / previous page for more information on this.

Video Cards Tested

Here's a list of video cards tested:

• AMD R9 Fury X 4GB HBM (& CF) ($630)
• MSI R9 390X Gaming 8GB ($430)
• Sapphire R9 390 8GB ($345)
• Sapphire R9 380X 4GB ($230)
• PowerColor R9 380X 4GB for CF ($230)
• Sapphire R9 380 4GB ($210)
• XFX R9 280 2GB (Deprecated)
• PowerColor R9 290X 4GB (Deprecated)
• XFX R9 285 2GB ($206)
• AMD R9 270X 2GB Reference (Deprecated)
• NVidia GTX 980 Ti 6GB Reference (& SLI) ($660)
• MSI GTX 980 4GB Gaming ($515)
• EVGA GTX 970 SSC 4GB ($315)
• EVGA GTX 960 SSC 4GB ($225)
• ASUS GTX 950 Strix 2GB ($170)
• EVGA GTX 750 Ti 2GB ($125)

You'll notice we did not test GPUs for idle. This is an artifact of old methodology that we are revising as we continue to shift our bench hardware to an updated rig. Because we no longer have access to the Fury X or R9 380 (non-X) and R9 390 (non-X), we maintain older power consumption methodology for GPUs. This is in exchange for keeping the once-loaner hardware present on the bench, as moving to idle/load testing on the new equipment would force-out the now-unavailable hardware. Going forward, we will be adding idle power consumption measurements for GPUs.

CPUs Tested

For CPU testing, we looked at the following products:

• Intel Pentium G3258 3.2GHz (HW DC) ($60)
• Intel Core i3-4160 3.6GHz (Haswell) ($122)
• Intel Core i5-4690K 3.5GHz (HW DC) ($210)
• Intel Core i5-6600K 3.5GHz (Skylake) ($270)
• Intel Core i7-4790K 4.0GHz (HW DC) ($300)
• Intel Core i7-6700 3.4GHz (Skylake) ($350)
• Intel Core i7-5930K 3.5GHz (HW-E) ($460)
• AMD Athlon X4 3.8GHz 760K (Richland) (Deprecated)
• AMD A10-7870K 3.9GHz (Godaveri) ($129)
• AMD FX-8320E 3.5GHz (Vishera) ($120)
• AMD FX-8370E 4.0GHz (Vishera) ($200)
• AMD FX-9590 4.7GHz (Vishera) ($200)

CPUs were left at stock clocks for this test. As mentioned elsewhere in this article, overclocking will increase watt draw and should be factored-in to PSU selection (add overhead if planning to OC considerably).

System Templates

These system templates are mostly just thrown together from parts we have access to, but they're reasonably built and represent what we think would occur in the real world. We carefully created a mix of AMD/Intel/nVidia components to best fill the power draw gaps between each configuration. The PSU used in every bench was the same for testing – we would, obviously, recommend lower wattage PSUs for these configurations.

Windows 10 was used for all systems. The same peripherals and monitor were used for all systems (ASUS MG278Q, Tt eSports Poseidon Z keyboard, Logitech G502 mouse). Some components were provided previously by manufacturers for tests, all of whom have been named in the “Courtesy of” column. iBUYPOWER and CyberPower additionally provided some CPUs for these tests.

System 1 (note: Memory pieced-out from 32GB kit, 1x8GB single-stick, single-channel configuration; price is not defined for this reason).

System 2 (note: Same Z97 motherboard was used for all Haswell/DC systems, despite the fact that an i3-4160 CPU should not likely be paired with this motherboard; this was for testing purposes and, simply, availability purposes):

System 3:

System 4 (Note: Dual 380X chosen for AMD representation and to look at other power draw scenarios):

System 5:

This testing was made possible by sponsorship from Enermax.

Power Consumption Testing Methodology – GPUs

GPU power consumption was conducted using our Z97 test bed, as defined below. The power supply used was Enermax's high-end Platimax PSU, with 80 Plus Platinum-rated efficiency. To create a workload similar to a real-world gaming environment, we deployed 3DMark FireStrike Extreme, looping the Graphics 2 benchmark. This benchmark places the GPU under 100% load that, although synthetic, closely resembles a gaming workload.

This benchmark was left to run for 20 minutes, allowing for load and heat generation over the test period. Power draw was recorded manually from a wall meter. The benchmark does not heavily load the CPU and is primarily a demonstration of GPU power consumption under 100% load.

Overclocks were applied where defined. For each overclock, we have in-depth tables of overclock stepping and the process used to achieve the frequency. These tables are found on individual video card reviews.

For additional metrics and test diagnostics, we used GPU-Z and AIDA64 background monitoring utilities.

Power Consumption Testing Methodology – CPUs

CPU power consumption was conducted using test benches with platform modifications as dictated by CPUs. The power supply used was Enermax's high-end Platimax PSU, with 80 Plus Platinum-rated efficiency. To create a replicable, constant load on the CPU, we used HandBrake encoding of a 4K video file captured at 50Mbps during a Battlefront gameplay session. This benchmark places the CPU under 100% load that resembles a processor-intensive workload.

The benchmark was left to run for 10 minutes, allowing for load and heat generation over the test period. Power draw was recorded manually from a wall meter. The benchmark is primarily a demonstration of CPU power consumption under 100% load.

For additional metrics and test diagnostics, we used CPU-Z and AIDA64 background monitoring utilities.

All power saving features, including C-states and APM, were disabled for these tests. This was to create a reliable, replicable load in a worst-case out-of-box scenario. Overclocks were not applied for our CPU power draw benchmarks.

AMD AM3+ Bench

AMD FM2+ Bench