+- PINE64 (https://forum.pine64.org)
+-- Forum: PINE A64(+) (https://forum.pine64.org/forumdisplay.php?fid=4)
+--- Forum: General Discussion on PINE A64(+) (https://forum.pine64.org/forumdisplay.php?fid=3)
+--- Thread: Some hard numbers on Pine64 Power Usage (/showthread.php?tid=1532)
Some hard numbers on Pine64 Power Usage - ataylorm - 06-26-2016
Good Day Everyone,
I've been playing with my Pine64's for a couple weeks now, and thought I would share some details on observed power usage.
I'm using a multi-port 2.1 Amp per port power supply and running 3 Pine64's on it.
Output Voltage: 5.14v
Boot Max Amps: .66
Idle Amps: .30
I'm using mine as "servers" so they are running multiple shards of Redis on them right now for testing. I have search shard on a separate processor affinity, keeping 0 for the Ethernet, and 1-3 for Redis shards.
Running Redis-Benchmark against all 3 shards at the same time achieving near 100% CPU utilization on each core as well as putting over 100,000 requests per second on the network.
Full CPU and Network Amps: .81
So this gives me some basic calculations.
5.14 * .30 = 1.542 watts idle usage
5.14 * .81 = 4.163 watts peak usage
Now I do not have anything connected via USB, which would draw additional power to power those devices, so keep in mind that this is board only. I'm also not running HDMI as I am just using an SSH terminal.
Adding a Microsoft Comfort Curve Keyboard, Microsoft Optical Mouse, and HDMI monitor added about .15 amps to my usage.
Given that my electric cost fully delivered is $0.092 / Kwh, and that there are 24 hours in a day, and 30 days in a month (I know it's not exactly 30, but let's keep the math easy).
Idle Cost Per Month: $0.102
Peak Cost Per Month: $0.275
While additional devices may cause the Pine64 to draw more ancillary power, overall the power efficiency is really amazing for what all it can accomplish.
As a comparison, let's look at my Dev Redis server. It's a Intel Xeon E5-2620v2 x 2. Each core maxes out at about 105,000 requests per second. Giving it a max capacity taking into account a core for ethernet and io of 105,000 * 23 = 2,415,000 requests per second.
At full draw, with dual network, SSD drives, and such, it's about 130 Watts usage.
That gives me an output of 18,576.92 requests per watt.
In comparison the Pine64 running full out is producing about 84,000 requests per second or 20,177.75 requests per watt!
Now think about that for a moment, the Xeon is running 2.10 Ghz, and the Pine64 is 1.2 Ghz...
RE: Some hard numbers on Pine64 Power Usage - ssvb - 06-26-2016
How do you measure these 130 Watts for Xeon? How do you measure the current draw for the Pine64? Also isn't maxing at around roughly 84,000 - 105,000 requests per second in every test where you did real measurements a little bit suspicious? Wouldn't it point to the network possibly being the bottleneck rather than the CPU?
Other than this, thanks for the interesting results.
RE: Some hard numbers on Pine64 Power Usage - ataylorm - 06-27-2016
(06-26-2016, 06:52 PM)ssvb Wrote: How do you measure these 130 Watts for Xeon? How do you measure the current draw for the Pine64? Also isn't maxing at around roughly 84,000 - 105,000 requests per second in every test where you did real measurements a little bit suspicious? Wouldn't it point to the network possibly being the bottleneck rather than the CPU?
Other than this, thanks for the interesting results.
Good Questions,
So let me give you a few details. For measuring current draw, I am using a few devices.
USB Power: DORK USB Digital Multimeter - https://www.amazon.com/gp/product/B00J3JSEG6/ref=oh_aui_detailpage_o06_s00?ie=UTF8&psc=1
AC Power: P3 International P4460 Kill A Watt Usage Monitor - https://www.amazon.com/P3-International-P4460-Electricity-Monitor/dp/B000RGF29Q/ref=sr_1_2?s=hi&ie=UTF8&qid=1467028463&sr=1-2&keywords=kill+a+watt
AC Power: Cyberlink UPS
As for testing of the actual throughput capacity for Redis, I am using Redis-Benchmark to provide as close to equal comparison against both devices. I have also run it both locally each device, and across the network to compare and ensure that I do not have network bottlenecks in the way. My back end network is all CAT 6 wiring through a single data center grade DLink 48 port gigabit switch.
Since using Redis-Benchmark locally also technically uses the local ethernet chipset, I have set processor affinity for eth0 to CPU 0, and each Redis shard is then set for it's own affinity to a single CPU core as well.
A single core on the Pine64 can deliver approximately 28,000 request per second, compared to a single core on the Xeon processor delivering approximately 105,000. The Xeon is hyperthreaded giving me a possible 23 usable threads (reserving 1 for network), and the Pine64 is 3 usable threads (reserving 1 for network).
Hope that all make sense?