Performance per watt (W)

Ecological issues around computing. This is an experimental forum.

Moderators: Ralf Hutter, Lawrence Lee

Post Reply
maalitehdas
Posts: 166
Joined: Tue May 27, 2008 12:04 am
Location: Finland

Performance per watt (W)

Post by maalitehdas » Tue Sep 07, 2010 9:49 am

Calculated clock speed (MHz) per TDP (W) for max 10W processors, results:
W MHz/W model
2 915 Intel Atom Z540
2 830 Intel Atom N280
2 800 Intel Atom N270
2 800 Intel Atom Z530
2 665 Intel Atom Z520
2 550 Intel Atom Z510
5 332 Intel Atom N450
5 300 VIA Nano U2250
5 280 VIA Nano U3200
5 260 Intel Celeron ULV 743
5 260 VIA Nano U3100
5,5 254 Intel Core 2 Solo SU3500
5 240 Intel Celeron ULV 723
5 240 VIA Nano U3300
5 210 VIA Nano U1700
5 200 VIA Nano L3500
5 200 VIA Nano U2300
6,8 198 VIA Nano U2500
8 162 VIA Nano U2225
10 160 Intel Core 2 Duo SU9600
10 140 Intel Core 2 Duo SU9400
10 130 Intel Core 2 Duo SU7300
10 130 Intel Core 2 Duo SU4100
10 130 Intel Pentium SU2700
10 120 Intel Celeron Dual Core SU2300
10 120 Intel Core 2 Duo SU9300
10 120 AMD Athlon L110
*Please check that my TDP's are right, they are from different sources (en wiki, de wiki, nl wiki)
** All dual core processors MHz/TPD should have some multiplier for comparison. I would suggest to use a multiplier between 1.8-1.9
Last edited by maalitehdas on Tue Sep 07, 2010 10:37 am, edited 3 times in total.

maalitehdas
Posts: 166
Joined: Tue May 27, 2008 12:04 am
Location: Finland

Post by maalitehdas » Tue Sep 07, 2010 10:02 am

Calculated Passmark score per TDP (W) for a lot of processors, results:
W score/W model
2 155 Intel Atom N280
2 150 Intel Atom N270
2 150 Intel Atom Z530
2 125 Intel Atom Z520
10 92 Intel Core 2 Duo SU9400
10 88 Intel Core 2 Duo SU7300
5,5 87 Intel Core 2 Solo SU3500
10 83 Intel Celeron Dual Core SU2300
25 83 Intel Core 2 Duo P9600
45 82 Intel Core 2 Extreme QX9300
45 81 Intel Core i7 820QM
10 77 Intel Core 2 Duo SU9300
55 76 Intel Core i7 920XM
25 75 Intel Core 2 Duo P8800
45 74 Intel Core i7 720QM
25 72 Intel Core 2 Duo P8700
17 72 Intel Core 2 Duo SL9400
45 71 Intel Core 2 Quad Q9100
25 70 Intel Core 2 Duo P9500
28 68 Intel Core 2 Duo P9700
Unfortunately I didn't find passmark results for VIA processors. Anyway it was most interesting that few 45-55W processors are there too.
Please check that my TDP's are right, they are from different sources (en wiki, de wiki, nl wiki)

lm
Friend of SPCR
Posts: 1251
Joined: Wed Dec 17, 2003 6:14 am
Location: Finland

Post by lm » Tue Sep 07, 2010 10:12 am

But the TDPs are usually the same for the whole line of CPUs, with several differing clock speed models having the same TDP. Doesn't this make the numbers pretty much useless?

maalitehdas
Posts: 166
Joined: Tue May 27, 2008 12:04 am
Location: Finland

Post by maalitehdas » Tue Sep 07, 2010 10:17 am

Just for curiosity here's some models with worse results:
W MHz/W model
35 30,49 Intel Core i5 520UM
45 35,56 Intel Core i7 720QM
55 36,36 Intel Core i7 920XM
45 38,44 Intel Core i7 820QM
45 44,44 Intel Core 2 Quad Q9000
W score/W model
35 28,57 AMD Athlon X2 QL-64
35 27,43 AMD Turion X2 RM-70
15 27,33 AMD Athlon MV-40 Neo
35 25,71 AMD Athlon X2 QL-60
45 23,56 Intel Core 2 Duo T5870

maalitehdas
Posts: 166
Joined: Tue May 27, 2008 12:04 am
Location: Finland

Post by maalitehdas » Tue Sep 07, 2010 10:25 am

lm wrote:But the TDPs are usually the same for the whole line of CPUs, with several differing clock speed models having the same TDP. Doesn't this make the numbers pretty much useless?
Well, maybe only the best clock speed model of the same TDP-serie should be counted. But definitely not useless info for green computing.

ilovejedd
Posts: 676
Joined: Tue Oct 21, 2008 4:14 pm
Location: in the depths of hell

Post by ilovejedd » Tue Sep 07, 2010 11:25 am

There's just way too many factors to consider than purely processor TDP.

My CULV laptop (Asus UL30A-X5: SU7300, 10W TDP) consumes a maximum of around 20W when stressed and 8~10W during regular use (web browsing). Compare that to my brother's netbook (Acer Aspire One 8.9": Atom N270, 2W) which normally uses around 10~12W. Haven't really measured the netbook's power consumption when stressed.

scdr
Posts: 336
Joined: Thu Oct 21, 2004 4:49 pm
Location: Upper left hand corner, USA

Post by scdr » Tue Sep 07, 2010 12:19 pm

[quote name=maalitehdas]Calculated clock speed (MHz) per TDP (W) for max 10W processors[/quote]

The ratio of two almost meaningless numbers - the perfect benchmark. ;-)

(The clock speed is primarily useful for comparing processors within one architecture. Have to be sure that things like number of cores, amount of cache, etc. are held constant.)

To get something like a meaningful comparison for systems that are even remotely close in performance per watt, you really need to measure how much energy it takes to perform a particular task or in typical use.

For instance, from Folding@home - "Points per day" per watt
viewtopic.php?t=42087&highlight=ppdpw

(Or the inverse killo-watt hours/killo-point )
viewtopic.php?t=46406&highlight=

May serve as a useful measure.

Also, considering that different CPU families integrate differing amounts of the support chips, it may not be so meaningful to compare just the wattage of the CPU. (A CPU that includes a memory controller, graphics processor, etc. might appear to fare worse than one that requires several other chips to do the same job.)

So, for instance, atom systems generally fare poorly when compared to efficient Core2 or i3 systems in performance per watt when one includes all the components necessary to make a computer. (See for instance the Mac mini.) Atom are low power consuming, but they are also low performance.

Also, since typical computers spend most of their time idle, or not fully loaded, a useful performance per watt for general applications would include the wattage when the system is idle.

maalitehdas
Posts: 166
Joined: Tue May 27, 2008 12:04 am
Location: Finland

Post by maalitehdas » Tue Sep 07, 2010 2:23 pm

Thanks for all the info. So let's start the seek of green CPU. With everything above in mind I try this one next, waiting for opinions / challengers:
Intel® Core™ i7 Mobile Processor i7-660UM
Arrandale 1.33 GHz (Turbo 2.4 GHz)
2 cores, 4 threads. 64-bit
cache 4 MB, 2.5 GT/s DMI bus
TBT,VT-x,VT-d,EIST,AES-NI,TXT,Integrated graphics
TDP 18 W

scdr
Posts: 336
Joined: Thu Oct 21, 2004 4:49 pm
Location: Upper left hand corner, USA

Post by scdr » Tue Sep 07, 2010 4:59 pm

Hope my previous post wasn't unduely negative. I don't mean to discourage seeking efficiency. Performance per watt is interesting, and I would like to have some better information readily available to compare consumer grade computers. Just haven't found much in the lines of a simple metric that is readily available.

Performance per watt isn't everything (have to consider absolute performance too). Especially with low power CPUs like Atom. Marketers tout low power CPUs as green, but that may or may not be reasonable. Depends on the application. In general the computer that gets the job done faster (within a reasonable budget for power, hardware, etc.) often uses less resources than a slower machine which takes longer.

Especially if one focuses on the low power end - the savings to be achieved from improved efficiency may be minimal. Not saying one shouldn't look at efficiency for low power systems, but not clear why that would be a focus. (As compared to just flat out efficiency.)


Most of the energy efficiency benchmarks that I have found so far focus on servers, or those with heavy computing needs.

http://en.wikipedia.org/wiki/SPECpower
http://www.spec.org/power_ssj2008/resul ... j2008.html
(has a list of systems, but they are all Xeons and Opterons)

http://en.wikipedia.org/wiki/Average_CPU_Power
Again - rating for server CPUs.
Note how "Average CPU Power" is generally much less than TDP.

And, of course, the Green500 ranking of supercomputers.
maalitehdas wrote: Well, maybe only the best clock speed model of the same TDP-series should be counted. But definitely not useless info for green computing.
TDP is defined for a whole range of CPUs. Even if you take the top CPU currently available in the range, there is no reason to expect that it will take anything like the TDP. (The envelope may have been defined to accomodate CPUs as yet unbuilt, or that never will be built.)

Also, TDP estimates vary systematically between manufacturers, so an Intel TDP watt is not the same as an AMD TDP watt. So that doesn't make a reliable basis for comparison between manufacturers.

scdr
Posts: 336
Joined: Thu Oct 21, 2004 4:49 pm
Location: Upper left hand corner, USA

Post by scdr » Tue Sep 07, 2010 6:12 pm

maalitehdas wrote:Thanks for all the info. So let's start the seek of green CPU. With everything above in mind I try this one next, waiting for opinions / challengers:
Intel® Core™ i7 Mobile Processor i7-660UM
Arrandale 1.33 GHz (Turbo 2.4 GHz)
2 cores, 4 threads. 64-bit
cache 4 MB, 2.5 GT/s DMI bus
TBT,VT-x,VT-d,EIST,AES-NI,TXT,Integrated graphics
TDP 18 W
First question - CPU for what purpose?
(Are looking for FLOPS/watt? IOPS? Scalar performance? Best for "general purpose"?)
How much weight give to idle power?

Thought this table might be helpful in this quest (or maybe folks have already seen it). It doesn't cover Passmark benchmark but has several others.

http://www.notebookcheck.net/Mobile-Pro ... prime_32=1

Wish there was something that collected measured watts while performing the benchmarks.

maalitehdas
Posts: 166
Joined: Tue May 27, 2008 12:04 am
Location: Finland

Post by maalitehdas » Wed Sep 08, 2010 1:20 pm

I found this list for mobile processors, passmark scores are there for most of them. In the specification TDP's are too rough (0 digit), at least some of them are different from what I found. Nice list thou.
http://notebookspot.nl/charts/cpu/0/

I asked around my friends about this performance per watt stuff, one of them posted me a link to a blog. Writer seems to share scdr's opinion about meaningless numbers:
http://silenssimo.blogspot.com/
About i7-660UM, scdr wrote:First question - CPU for what purpose?
(Are looking for FLOPS/watt? IOPS? Scalar performance? Best for "general purpose"?)
How much weight give to idle power?
I'm looking for none of those mentioned, more like a green computer that doesn't consume too much power for doing nothing. well, "general purpose" can, of course be anything including that. Idle power should be at high level of importance, that's what most computers do.

Performance of the computer should be close to what is needed, that varies bu user of course. Atm I'm looking for a way to measure it - a formula to define what is the most energy efficient processor for "you". i7-660UM might be a good reference CPU including so much with low TDP. If it's performance doesn't satisfy, you can compare how many watts you need more in CPU which has your required performance, and if it has some extras you don't need, you can see if you can find one without them with less TDP.

atmartens
Posts: 80
Joined: Thu Jul 23, 2009 3:52 pm
Location: Baltimore, USA

Post by atmartens » Thu Sep 09, 2010 9:22 am

To find the "greenest" CPU, first determine what you need to do. Then, find the processors capable of doing that task. Take the one that is capable of performing that task and that consumes the least power. It's wasteful to buy more CPU than you need, and it's wasteful to buy a CPU that can't do what you need.

maalitehdas
Posts: 166
Joined: Tue May 27, 2008 12:04 am
Location: Finland

Post by maalitehdas » Fri Sep 10, 2010 2:31 am

That's exactly what I mean. But what kind of compare/review/formula tells me the one which is most energy efficient? Let's start with something pretty typical: How do you search for a CPU with lowest wattage, aiming for a computer able to:
- show 1080p movies smoothly (Dark Night is a hard one)
- show modern games with 1280x720 resolution, 50Hz, 20fps average, 12fps min

Last requirement makes it difficult to choose between integrated graphics and energy efficient GPU-cards, doesn't it? What data is needed to make the right decision?

scdr
Posts: 336
Joined: Thu Oct 21, 2004 4:49 pm
Location: Upper left hand corner, USA

Post by scdr » Sat Sep 11, 2010 1:16 am

atmartens wrote:To find the "greenest" CPU, first determine what you need to do. Then, find the processors capable of doing that task. Take the one that is capable of performing that task and that consumes the least power. It's wasteful to buy more CPU than you need, and it's wasteful to buy a CPU that can't do what you need.
Agree definitely don't want to underbuy - which is what makes a lot of the low power, so-called green machines of dubious greenness. From a green perspective, better to overbuy (to a modest extent) than to get just what does the job now. (Since the software almost always gets more demanding as time goes on - Wirth's Law.)

So the greenest CPU would be (in approximate priority order):

0) Except for special circumstances, there are probably other areas where a home user can save a lot more than getting really hung up on CPU power efficiency. (Special circumstances would be things like: going to get several machines, running off grid.) (Other areas - things like heating, cooling, lighting, transportation)

1) CPU that don't spend an inordinate amount of time and energy shopping for, compared to the energy used by the CPU ;-)
[For instance, if one spends extra time shopping for the low-energy using CPU, and if average energy use of a person in US 10kW, then for each hour one spent shopping, that CPU has to use 10kW-hours less during it's lifetime in order to make up for the energy expended to support the person while shopping.]
http://en.wikipedia.org/wiki/List_of_co ... per_capita
(of course if one can put together a resource so many people use more efficient device, that would help offset the cost of doing the research.)

2) the one that will do the job at hand, plus handle the job for a reasonable time in the future, (longer is better)
3) That can get second-hand
4) That has low idle power (unless have special type of job)
5) Support chips/motherboard low power
6) Balances cost of chip versus cost of energy (return on investment for higher efficiency)
7) High performance per watt when loaded

electrodacus
-- Vendor --
Posts: 372
Joined: Sun Jul 12, 2009 10:30 am
Location: Canada
Contact:

Post by electrodacus » Tue Sep 14, 2010 10:28 am

CPU performance vs TDP is not important if you want low power system.
The CPU will not work alone it also need (memory controller + IDE or SATA controller and GPU as minimum) you will probably also need sound and network.

If you need HD video playback then CPU will not be efficient at playing HD video you need GPU acceleration.
The best platform for energy efficiency are at the moment Z5xx + GMA500 I have a Sony P based on this platform and if I exclude the LCD and use good drivers I have a 5W idle, 6W 1080p, and 9W max power consumption for Z530+US15W+2GB RAM+16GB compact flash+sound and wired LAN.
Then is this VIA C7-M ULV 1GHz (3.5W TDP) + VIA VX855 integrated VIA Chrome9 HCM DX9 3D video (2.3W TDP) also support 1080p HW decode of H.264 that is 4W idle ,8W max about 5W 1080p playback.
But if you are interested in lower power the only option I know is ARM platform for example the beagleboard is using 2W max probably only able to play up to 720p but there will be soon 1080p capable ARM devices as Tegra2 with similar low power consumption.

scdr
Posts: 336
Joined: Thu Oct 21, 2004 4:49 pm
Location: Upper left hand corner, USA

Post by scdr » Tue Sep 14, 2010 9:52 pm

I found these references interesting on the relative efficiency of Atom vs ARM vs Lynfield (core2)

http://arstechnica.com/business/news/20 ... tion.ars/3

http://arstechnica.com/business/news/20 ... enters.ars

Which suggest that Lynfield more efficient than Atom (for the sorts of loads they are measuring), and that ARM might have a hard time doing much better (big issue being support chips).


Another article comparing processor families (again using theoretical performance and TDP).
http://realworldtech.com/page.cfm?Artic ... 050230&p=1

Suspect Atom is more of an outlier because of the specifications rather than real performance.

Post Reply