1 |
On Mon, Jul 30, 2012 at 10:14 AM, Michael Mol <mikemol@×××××.com> wrote: |
2 |
> On Mon, Jul 30, 2012 at 1:04 PM, Mark Knecht <markknecht@×××××.com> wrote: |
3 |
<SNIP> |
4 |
>> |
5 |
>> MB power? |
6 |
>> Hard drive power? |
7 |
>> Hard disk power |
8 |
>> GPU power? |
9 |
>> DRAM power? |
10 |
>> |
11 |
>> The 5 above can easily become the dominant power hogs. |
12 |
>> |
13 |
>> I use an Intel i7 980X 6-core hyper-threaded CPU, so that's 12 CPUs in |
14 |
>> top, which burns _lots_ of power, but I suspect it's not the biggest |
15 |
>> power consumer when compared to the total of the 6 500GB 7200 RPM hard |
16 |
>> drives I have in the box. |
17 |
> |
18 |
> Spinning disks consume surprisingly little power once they're up to |
19 |
> speed. My GPU, by comparison, doesn't seem to reduce heat generation |
20 |
> very much when relatively idle. |
21 |
> |
22 |
|
23 |
OK, point taken about the disks at least when talking about a single |
24 |
disk. I see we measure the same way with a Kill-a-Watt so that's at |
25 |
least consistent. |
26 |
|
27 |
And I think we're in more or less violent agreement, but reducing the |
28 |
CPU power in the end won't save all that much on the electric bill, or |
29 |
so I think. |
30 |
|
31 |
I ran around really quickly to find some spec values for the machine |
32 |
I'm on right now. No idea if the numbers are right. They are just what |
33 |
I found quickly: |
34 |
|
35 |
CPU - i7-980x - 130W |
36 |
|
37 |
vs |
38 |
|
39 |
Asus Rampage II Extreme - Can't find so far, so let's guess 25W for |
40 |
the chipset, NICS, audio devices, etc. |
41 |
WD RAID Edition 500GB - 8W * 6 = 48W |
42 |
NVidia GTX 465 class card = 80W |
43 |
24GB DRAM - Total guess, but about 1W/DIMM looks safe, so 6W total |
44 |
|
45 |
So a __really__ rough guess is my machine should use about (130 + 25 + |
46 |
48 + 80 + 6) Watts, or 289W assuming I added it up correctly. That's |
47 |
under full load though. My UPS has a power meter in it. The UPS is |
48 |
driving this machine, 3 monitors, a small switch, a wireless access |
49 |
point and maybe one or two other small things I've forgotten or have |
50 |
plugged in somewhere. When the machine is essentially idle that mater |
51 |
reads 330W. When I start a VM that uses 6 processor cores and runs for |
52 |
30 minutes at full tilt the power consumption is 385W. I no longer |
53 |
remember what I have set up in terms of the CPU clocking stuff. It's |
54 |
on the machine and let's the box go full tilt, but I'm not sure what |
55 |
it does when the machine is idle. |
56 |
|
57 |
Anyway, if you assume that the 55W jump was the difference between the |
58 |
980x idling, and then using 3 cores full tilt, then 6 cores (12 |
59 |
threads) might be more like a 100W jump which seems about right |
60 |
according to Intel's spec. |
61 |
|
62 |
In terms of the electric bill, don't forget the PC power supply is |
63 |
only 80-90% efficient, so 10-20% is thrown away there also. |
64 |
|
65 |
Now, assume you get a CPU that draws half the power. This setup would |
66 |
still likely draw something close to 330W when it's idling, and might |
67 |
only jump up by 60W when running full tilt. That would save maybe |
68 |
(330+60)/(330+100) or only about 10% on the whole system power |
69 |
consumption. For that reason I don't think skimping on the CPU makes |
70 |
much sense to me. I'll happily turn the box off 2 hours a day vs go |
71 |
slower all the time, but that's just me. |
72 |
|
73 |
>> |
74 |
>> WRT to money spent to run a machine I hope someone stated earlier than |
75 |
>> this that it's the whole system that matters and not just the CPU. |
76 |
> |
77 |
> I didn't state so explicitly, no, but I believe I mentioned the two |
78 |
> machines had been otherwise comparable in their equipment loadout. If |
79 |
> I missed that, my bad. |
80 |
|
81 |
And I don't know that you did as I haven't read the thread, but part |
82 |
of my argument is that you have to know the WHOLE system and not just |
83 |
the CPU to decide if changing the CPU costs or saves much power. |
84 |
|
85 |
Cheers, |
86 |
Mark |