MP 1,1-5,1 cMP 4,1, Power consumption, and the real world

[Darmok N Jalad]

I'm sure much has been discussed about power usage on the cMP, but I wanted to run some everyday numbers on my cMP 4,1 just to see the costs of operating the machine 10 years later. With the passage of time, efficiency gains in technology might lead us to think that these old machines are going to hit us in the pocket book when it comes to energy consumption, but just how much does it cost? Are machines from 10 years ago really so inefficient to consider replacement? Is tossing what works for something newer more environmentally friendly? Well, those questions were the purpose of my testing! Not long ago, I got myself a Kill-A-Watt meter, so now my questions can be answered!

First off, I'm running an updated version of the 4,1. I've flashed it to 5,1, installed a W3690, pulled the spinning drive for an SSD, added an RX 580, and updated to Mojave. Since I can't effectively use my Mac without a display, my measurements will include my monitor--a very modern LG 27" 4K IPS display. I may eventually measure the display just to be able pull it from the results.

Assumptions:
I let my machine sleep after 15 minutes of idle, and I don't turn it off. Sleep is a big contributor, as it will be the state of my 4,1 probably 80-90% of the time throughout the year. I allow the monitor to also go to sleep. For these results, I'm assuming approximately 20 hours of the day my machine will sleep. The other 4 hours of the day will be mixed use. I rarely load my machine (no games, rare encoding), so I don't expect heavy load usage rates very often. I will be generous and say I load my machine for 1 hour a day and it idles for the other 3 hours. I'll use the average US national rate of $.12/kWh to estimate a yearly cost and this site to do my math.

Results:
Sleep: 5W, or 37 kWh per year ($5)
Idle at desktop: 142W, or 155 kWh per year ($19)
Heavy workload: 315W, or 115 kWh per year ($14)

So for my use case, my 4,1 Mac Pro with display costs me about $38 a year to operate. I'm actually quite surprised that the idle power draw was so low. The energy efficiency of such an old machine is commendable. I think my usage is being quite generous to the high side--I probably am overestimating my usage by at least 10%.

I'll run a few other household items and see how they compare. I have a late model PS4 Pro attached to a 48" UHD that I'm curious to measure as well.

 

[kohlson]

Years ago I had an APC UPS with an LED on the front that showed power consumption. IIRC, it was 30 watts in sleep mode, and 190 watts at idle. I don't recall many other readings, though it certainly bumped up a bit when I was doing "work." I've since replaced the stock 4-core 2.93 Nehalem with an x5680. The drives now seem to be spec'd for lower power consumption. But an RX-580 has replaced the dual GT-120s. It starts up in less than a minute (compared to 3 or so minutes with an HDD), so I keep it off when not using it.

 

[MIKX]

I have a " Watt meter" arriving from Amazon next week !

One thing I'm certain of is that my electricity bills went down by 30% ( in Japan ) after i replaced my 2,1 2007 Dual Quad 3Ghz with a 2.93 Ghz Dual Quad 2009 4,1. . much, much better energy management and substantially cooler too.

Both cMPs basically ran 24 hrs/day, didn't sleep them much but now use my 8,1 MBP after 11pm.

 

[donluca]

The X series Xeon should be more efficient.

I think I'll get a watt meter as well, silly me, prime day just ended yesterday.

 

[tsialex]

The X series Xeon should be more efficient.

I think I'll get a watt meter as well, silly me, prime day just ended yesterday.

Definitively no, X5xxx versions are the 95/115/130W TDP. L5xxx (80W) and E5xxx (60W) are the power consumption optimised versions of Westmere Xeons.

W3690 is the X5690 with just one QPI link, power consumption could be even a little less.

 

[bsbeamer]

In case it's helpful, dual X5690 MP5,1 with RX580 and all SSDs (no spinning HDD in system) is about 200 watts initially during boot/startup, spikes to about 290 and 325, kicks to above 400 when monitor starts, then settles around 310 watts and seems to be a steady 260-275 watts while "resting" without doing much activity and the monitor on. Spikes to 325-375 watts are common. Rarely does this pull more than 425 watts during heavy activity lately. All numbers from CyberPower 1500PFCLCD UPS with MP5,1 and 27" ACD monitor connected to it. (And no external HDD's turned on.)

 

[donluca]

Definitively no, X5xxx versions are the 95/115/130W TDP. L5xxx (80W) and E5xxx (60W) are the power consumption optimised versions of Westmere Xeons.

W3690 is the X5690 with just one QPI link, power consumption could be even a little less.

We're saying two different things.

One thing is TDP, which tells us how much power the CPU can use, another thing is efficiency, which means how low a CPU can consume while idling.

X series have higher TDPs and better efficiency, if you take the same processor in its E, W and X variants, you'll see that while idling the X is more efficient.

https://forums.tomshardware.com/thr...-and-x-series-processors.249394/#post-2536596

L Xeons are a completely different category and, yeah, they are geared towards power efficiency.

 

[tsialex]

We're saying two different things.

One thing is TDP, which tells us how much power the CPU can use, another thing is efficiency, which means how low a CPU can consume while idling.

X series have higher TDPs and better efficiency, if you take the same processor in its E, W and X variants, you'll see that while idling the X is more efficient.

https://forums.tomshardware.com/thr...-and-x-series-processors.249394/#post-2536596

L Xeons are a completely different category and, yeah, they are geared towards power efficiency.

We are talking W3690 and X5690, both are Westmere-EP cores with step B1, your linked post is not valid here.

A W3690 is a X5690 with one QPI link disabled. It's not a different processor like with other Xeons series and use the exact same 206C2 microcode as X5690.

There's no difference in efficiency whatsoever, besides a marginal VID difference, it's the same processor with a disabled QPI link and artificial memory limitations.

 

[h9826790]

AFAIK, X5690 has tiny bit lower voltage range than W3690, but I doubt if there is any measurable power consumption difference in real world.

 

[tsialex]

AFAIK, X5690 has tiny bit lower voltage range than W3690, but I doubt if there is any measurable power consumption difference in real world.

Marginal VID difference is probably because X5690 is better binned than W3690.

 

[orph]

Iv always wanted to now how accurate the power readings are on the mac pro, has anyone compared them to a wattmeter or can anyone do a quick check to see how close they line up?

Iv been using Hardware Monitor for years and watching power use for fun, tend to see 100w-300w ish from the history graph.
https://www.bresink.com/osx/HardwareMonitor.html

I guess it will also depend where the power reading is, before or after the PSU lose (so maybe 20% higher than the reading?) have noticed when the computer is under low use the PSU tends to get a lot hotter so assume it’s not at its most efficient till it’s past the 200w measured point?

I still find it amazing that it’s not normal to have all the power and temp sensors the mac pro has, relay cool to see each PCI slot’s power use and the two 6 pins power use.

from Darmok N Jalad post if his display is using 20-40w then id gess the macpro readings are vary close

 

[Darmok N Jalad]

Ok, so I plugged in my display to the KAW, and I must say, I am impressed. When it goes into full sleep mode, it goes to 0W somehow, even though it wakes with my Mac. So the monitor effectively costs nothing to operate when not in use. When the display is awake, is uses about 29W. That adds up to 42kWh yearly (4 hours on, daily in the example above), or $5. So the "control" readings would be the following:

Results (no display):
Sleep: 5W, or 37 kWh per year ($5)
Idle at desktop: 113W, or 124 kWh per year ($15)
Heavy workload: 286W, or 105 kWh per year ($13)

Not bad at all for a 10 year old machine.

For fun, here's my PS4 Pro with TV. In my life, I only game about 1 hour a day, but I do use the PS4 idle mode for game updates.
Sleep: 3W, or 25 kWh per year ($3)
Idle at dashboard: 70W
Gaming: 154W max, 56 kWh ($7)
48" LG UHD: 80W, 29 kWh ($4)

So my gaming habit costs me $14 a year! Combined, these 2 systems cost just under $60 annually to operate.

 

[orph]

fun to see an update, thanks

 

[howiest]

Funny story...

I used to leave my cMP running 24/7 and only letting my dual monitors sleep (because I had EyeTV recordings scheduled).
But we found ourselves leaving home for commitments for 3 days a week, and so I started powering down each week. Not long after that we received a hydro bill that said... "congratulations!! you've lowered your hydro bill by 19%!!"

That 19% was solely due to me powering down my cMP for 3 days a week. And, it may only amount to $20.00/month or so... but that adds up over time. Just sayin'.

 

[clystron]

Ok, so I plugged in my display to the KAW, and I must say, I am impressed. When it goes into full sleep mode, it goes to 0W somehow, even though it wakes with my Mac.

So it is EU compliant, in theory many devices (including PCs and displays) sold in Europe must not consume more than 1W when in standby. Many devices have a specific setting to enable this "mode" and offer less features if its enabled so unfortunately its not quite as "automatic" as it should be....

One thing to consider about power-consumption is that it also means heat, if you live in a mostly warm climate a higher power-consumption might also increase your cooling needs. In the winter it may actually reduce your heating bill by a tiny amount.

 

[orph]

A consumer power meter will not be 100% accurate but even with a 10% leeway the numbers look close.

From what I can tell it’s only idle/light power use that is higher than new computers, the new top end consumer CPU’s can use much more power than the cmp. The 9900K is closer to 170w+ under full load https://www.anandtech.com/show/13400/intel-9th-gen-core-i9-9900k-i7-9700k-i5-9600k-review/21

I assume the main difference will be work done per watt hour or something like that with newer chips.

On the waist heat side i half feel lucky to live in Wales most of the year we have a toasty 16c
https://www.holiday-weather.com/cardiff/averages/

Given the chance ill move somewhere hotter, id give up a cool computer for something closer to 20c every day of the week. (In the past I have used my computer as a hand warmer had ice inside my window)

 

[Darmok N Jalad]

So it is EU compliant, in theory many devices (including PCs and displays) sold in Europe must not consume more than 1W when in standby. Many devices have a specific setting to enable this "mode" and offer less features if its enabled so unfortunately its not quite as "automatic" as it should be....

One thing to consider about power-consumption is that it also means heat, if you live in a mostly warm climate a higher power-consumption might also increase your cooling needs. In the winter it may actually reduce your heating bill by a tiny amount.

It’s in the basement, so it’s automatically a cooler location, but yeah, I’m sure it’s more heat than an iMac would produce, and you’d probably get the same to better overall performance. It would just depend on need for that much power. I need something to process 16MP RAW photos, and the GPU does most of that lifting.

 

[amedias]

I've done fair bit of testing in this area....below machines were tested powered up to a desktop, left to sit for 10 mins and confirmed properly idle. They were then stress tested with a combination of 'yes' threads to occupy the CPUs and 'rember' to busy the Ram. You could stress them more with other CPU intensive workloads and also bring the GPU into it but for my testing I wanted a repeatable baseline with known components and without the added variable of GPU utilisation.

They were tested with a single SSD and for the 1,1/2,1s a single 7300GT, for the 3,1 a single GT510, and for the 4/5,1s a single GT120. Screen was turned off while readings were taken so this can be considered just the results for the Mac Pro alone. Fans were at minimum in all idle cases and ambient was ~25deg C, fans did spin up a bit on the 1,1-3,1s when stressed but only marginally and the impact seemed to be 3~5W extra for the fans. The 4/5,1s registered less than 2W increase even when the fans did increase a bit. Readings are rounded UP to nearest 5W as generally the reading fluctuated by 2-3W from second to second.


Mac Pro 1,1 4 Core 2x 5130 @2.0Ghz (65W TDP)
----------------------------------------------------
4x 4GB FB dimms
idle: 165W
load: 260W

8x 4GB FB dimms
idle: 195W
load: 310W

Mac Pro 1,1 4 Core 2x 5150 @2.66Ghz (65W TDP)
----------------------------------------------------
4x 4GB FB dimms
idle: 170W
load: 260W

8x 4GB FB dimms
idle: 200W
load: 310W


Mac Pro 2,1 8 Core 2x L5335 @2.0Ghz (50W TDP)
----------------------------------------------------
4x 4GB FB dimms
idle: 155W
load: 230W

8x 4GB FB dimms
idle: 185W
load: 280W


Mac Pro 3,1 8 Core 2x E5462 @2.8Ghz (80W TDP)
----------------------------------------------------
4x 4GB FB dimms
idle: 150W
load: 310W

8x 4GB FB dimms
idle: 185W
load: 360W


In all cases below machines were tested with 3 dimms on each CPU, so for single CPU systems they were 3x16GB and for dual CPU systems they were 6x16GB.

Mac Pro 4/5,1 4 Core 1x W5520 @2.66Ghz (130W TDP)
----------------------------------------------------
idle: 115W
load: 260W

Mac Pro 4/5,1 4 Core 1x W5530 @2.8Ghz (130W TDP)
----------------------------------------------------
idle: 115W
load: 255W

Mac Pro 4/5,1 4 Core 1x W5590 @3.33Ghz (130W TDP)
----------------------------------------------------
idle: 120W
load: 265W

Mac Pro 4/5,1 4 Core 1x X5677 @3.46Ghz (130W TDP)
----------------------------------------------------
idle: 115W
load: 240W

Mac Pro 4/5,1 6 Core 1x L5640 @2.26Ghz (60W TDP)
----------------------------------------------------
idle: 105W
load: 155W

Mac Pro 4/5,1 6 Core 1x X5670 @2.93Ghz (95W TDP)
----------------------------------------------------
idle: 115W
load: 220W

Mac Pro 4/5,1 8 Core 2x E5520 @2.26Ghz (80W TDP)
----------------------------------------------------
idle: 145W
load: 310W

Mac Pro 4/5,1 8 Core 2x X5570 @2.93Ghz (95W TDP)
----------------------------------------------------
idle: 155W
load: 340W

Mac Pro 4/5,1 8 Core 2x L5630 @2.16Ghz (40W TDP)
----------------------------------------------------
idle: 115W
load: 220W

Mac Pro 4/5,1 12 Core 2x X5690 @3.46Ghz (130W TDP)
----------------------------------------------------
idle: 165W
load: 355W


Obviously things can vary by a few Watts from one machine to the next anyway but hopefully people might find the above useful, it's interesting to see the impact the change to Westmere has, and also L series Xeons are quite a bit lower draw. If you're interested in other things too then a 5400RPM HDD spinning but doing nothing adds between 3-5W depending on the make and model, 7200RPM adds about 5-6W.

FB dimms are hungry little things, 4 extra Dimms added ~35W of idle load on the <3,1 machines, and ~50W under load! I did do some brief testing with fewer DDR3 Dimms in the 4/5,1s and it appears that 4 Dimms there is ~15-20W at idle ~25W at load, so they are significantly less power hungry.

Variation in GPUs is obviously massive, I've been using GT710s as my HS and Mojave card of choice, I don't need a lot of GPU power, the most stressful thing my GPUs have to render is my desktop so they are great, cheap, passively cooled and drop another 10W or so off a GT120 at idle.

I have one machine running as a headless ESXI box with no GPU, a single L5640 and all SSDs for storage and that idles at about 85-90W, which is about as low as I think you'll ever get a 4/5,1 to go. It's not a very busy host and 6 slower cores is plenty, and with the L series it only bursts up to ~140W when really busy.

 

[Darmok N Jalad]

Just for fun, I plugged in my Apollo Lake-based BRIX to the KAM. Poor little guy can't even break 20W when loaded down! It's by no means fast, but it can handle streaming and legacy console emulators just fine. With an SSD, it's pretty responsive.

 

[jscipione]

I have 2 Mac Pros hooked up to a power meter that measures the power coming out of the wall. I have a 5,1 with a single 2.26Ghz L5640 which uses 175W idle, 250W under load and a dual 3.33ghz X5680 4,1 which uses 250W idle and 500W under load. Both are fully loaded with optical drives and hard drives and PCIe slots and include monitor and speakers. Both use around 7W asleep. My numbers are a bit higher than what others are reporting.

 

[amedias]

If your figures include monitor that will add 20-60W depending on the model.

The HDDs and PCI cards will be adding extra Watts too.

I tried to do all my tests in as basic spec as possible to minimise contributions from other componentss and get figures for just the base cMP