Does Sandy Bridge idle power consumption correlate with TDP?

[Fjodor2001]

Hi,

I intend to build an Intel Sandy Bridge based computer that will be running 24/7. Therefore I want it to have as low idle power consumption as possible.

Now I just wonder if there is a significant difference in idle power consumption between the models that have lower TDP (e.g. 35/45W TDP) and those that have higher TDP (e.g. 65/95W)? Or is the TDP not a good indicator of the idle power consumption? I.e. is the Sandy Bridge CPU able to shut down the power consumption to very low levels when idle regardless of what TDP it has?

 

[Accord99]

Hi,
Or is the TDP not a good indicator of the idle power consumption? I.e. is the Sandy Bridge CPU able to shut down the power consumption to very low levels when idle regardless of what TDP it has?

This appears to be the case:

No difference at idle between the 65W i3-2100 and the 95W i7-2600K
http://techreport.com/articles.x/20401/5

No difference at idle between i3-2100 and 45W i3-2100T
http://www.hardware.fr/articles/819-5/amd-e-350-fusion-solutions-mini-itx.html

 

[Mopetar]

The power gating used is the same across all chips so the idle consumption is going to be very similar for all chips. A dual-core SB chip should be somewhat lower than a quad-core SB chip, but you're not going to see any difference between a 2500 and 2500S.

http://www.xbitlabs.com/articles/cpu/display/core-i5-2400s_6.html#sect0

 

[paperwastage]

for my i5 2500k, HWMonitor reports processor idling between 4-10W

At stress, depending on the program I use, it tops at 90W for Prime95, 105W for LinX(vcore = 1.32V)

again, these are numbers for the processor power only, and might not be totally accurate, but it seems to line up with what the norm for CPU power is

I assume that the rest of the power draw depends on your mobo, HDD, GPU, which should account to 50W?

 

[Mopetar]

No difference at idle between the 65W i3-2100 and the 95W i7-2600K
http://techreport.com/articles.x/20401/5

Interesting that there's no difference between the two considering that the 2100 has two fewer cores.

 

[Idontcare]

Interesting that there's no difference between the two considering that the 2100 has two fewer cores.

Not really, power-gating means the power to the cores can truly go to zero (including the idle leakage).

 

[BTRY B 529th FA BN]

Hard drives can tack on round 7-9w per drive. With my main rig, if I have just the SSD running it idles ~ 99w [taken from the wall (total system power)] With two Samsung F3 1Tb drives running the watts hover around ~ 116w

This is pretty good compared to my Q9650 rig

 

[Mopetar]

Not really, power-gating means the power to the cores can truly go to zero (including the idle leakage).

I suppose if regardless of the total number of cores it only leaves one running and wakes up others as needed the overall result would be the same. I just assumed that it wouldn't do that, but it makes sense if you don't need additional power.

 

[coolhand411]

I'm running 5.0 under offset mode (asus Pro) Idle 1.048 V (8.5W) bios1204
Real Temp 3.67 will show


Uploaded with ImageShack.us

 

[Fjodor2001]

Previously I thought I'd get a CPU for my 24/7 server that was "as low TDP as possible with a resonable performance". Now I'm leaning more towards "as high performance as possible for a reasonable price".

Previously I was actually considering some of the S or T models, since I thought they'd also have lower idle power consumption. Now that I know that's not the case, I don't see the point with these models, except when building computers that have a limited PSU (e.g. a PicoPSU & PowerBrick) or where there is limited cooling possibilities (e.g. when building a really quiet computer).

 

[Fjodor2001]

Hard drives can tack on round 7-9w per drive. With my main rig, if I have just the SSD running it idles ~ 99w [taken from the wall (total system power)] With two Samsung F3 1Tb drives running the watts hover around ~ 116w

This is pretty good compared to my Q9650 rig

Well, 2TB 5400RPM "green" HDs can consume as low as ~3W when idle.

I have one server based on the following hardware: Atom D510 / 2 GB RAM / 2x2TB 5400RPM "green" HDs, and a "standard" non-high-efficiency PSU. That system idles at around 28 W.

I intend to build a new very similar one where I change to a 1155 socket motherboard, an i3-2100/i5-2400, 4 GB RAM. Shouldn't that system be able to idle at around 35W?

 

[gevorg]

The low power Sandy Bridge S and T variants are good only for mass system builders like Dell, HP, etc. For the rest of us, just pick any reasonable CPU (K variant for maximum flexibility) and tweak it in the BIOS for your preferred TDP and clock frequencies. There is nothing special about S and T CPUs other than what "build-in" settings they have.

 

[IntelUser2000]

Now I just wonder if there is a significant difference in idle power consumption between the models that have lower TDP (e.g. 35/45W TDP) and those that have higher TDP (e.g. 65/95W)? Or is the TDP not a good indicator of the idle power consumption? I.e. is the Sandy Bridge CPU able to shut down the power consumption to very low levels when idle regardless of what TDP it has?

On the mobile chips, the lowest power states even that enable power gating show substantial differences between various voltage and number of cores, but on the desktops there's only 10% difference between standard voltage and low/ultra low voltage models.

Or is the TDP not a good indicator of the idle power consumption?

Two answers to that question.

1. Binning: Desktop processors probably all come from same bin regardless of power levels, which is why with the C6 state(power gating) that has clocks and everything off, has identical power levels.

Laptop processors have different values for even C6/C7 likely because higher wattage processors come from a different bin from lower wattage ones.

2. While C6 is identical for 35/45/65/95W desktop processors, C3 is not. CPUs go to the next level power state one by one rather than all at once, so C3 is just an important state as C6, especially if whatever you are doing doesn't allow C6 to engage.

 

[jiffylube1024]

Well, 2TB 5400RPM "green" HDs can consume as low as ~3W when idle.

I have one server based on the following hardware: Atom D510 / 2 GB RAM / 2x2TB 5400RPM "green" HDs, and a "standard" non-high-efficiency PSU. That system idles at around 28 W.
I intend to build a new very similar one where I change to a 1155 socket motherboard, an i3-2100/i5-2400, 4 GB RAM. Shouldn't that system be able to idle at around 35W?

35W idle? 7W more going from an Atom platform to Sandy Bridge? That's pretty unrealistic. 5-10W more just from the SB mobo is quite likely. ~50W+ is a more sensible target.

Any Sandy Bridge platform is going to have a higher power consumption motherboard, CPU and possibly even graphics. Are you going to be using the integrated graphics? If you have a modern GPU in there, mid-range or above, those idle at between 20-30W alone. If you're just using the intel IGP then it will probably be more like 2-5W.


I currently idle at 86W with the following system:

Sandy Bridge i5 2400 @ 3.7 GHz (stock voltage)
8GB DDR3
2x 1TB Seagate HDDs
1x Intel X25-M G2 160GB SSD
ATI 6870 HD
3x 120mm ~1000 rpm case fans, 1x 200mm slow case fan

So, if you go to integrated graphics from the 6870, probably subtract 15-25W for the idle power there, and maybe ~3W from downgrading fans. I'm overclocking from 3.1 GHz to 3.7 GHz by adjusting the multiplier only; maybe take another 1-3W off for that.

~50W idle may be doable; 35W on SB is probably impossible.

 

[Concillian]

Previously I was actually considering some of the S or T models, since I thought they'd also have lower idle power consumption. Now that I know that's not the case, I don't see the point with these models, except when building computers that have a limited PSU (e.g. a PicoPSU & PowerBrick) or where there is limited cooling possibilities (e.g. when building a really quiet computer).

The main usage for these low power versions of CPUs are for co-located server farms where you have a contract with the facility stating a specific current draw your systems must stay under. If you exceed that specific maximum current, even briefly, you incur a significant surcharge.

For anyone not subject to this kind of limitation, it's been proven many times that faster CPUs with good idle power management are actually more power efficient than "low power" versions which limit clock speed.

I had my home PC with an i3-530 down to about 37 or 38W idle measured by a Kill-a-Watt. max clock speed made very, very little difference, but I did have to underclock the bus speed to get the last 1-2 watts (since this also lowered the idle clock speed), which did hurt max speed. This is something almost nobody would actually do though, I was just playing around.

Realistically, a desktop motherboard with a desktop CPU, you are going to be looking at 40-50W from the wall at idle. There's just more features on the boards and those all take power. People buy features and backwards compatibility on mobos, not lack of features so they can save 3 or 5 watts. The motherboard MFRs make what sells in volume, and lack of features doesn't sell in volume unless it's CHEAP. Too many people spend an extra $10 for this or that feature "just in case" or for "future-proofing". The low end mobos really only sell in volume if they're priced very significantly lower, which means no profit... so you don't see too many desktop boards that are pointed towards low power consumption, especially when the difference between a stripped board and a reasonably featured one is probably only ~5W.

 

[lemmo]

Is idle power in Sandy Bridge significantly lower than Clarkdale?

I haven not yet seen a review which, say, compares a mini-ITX Sandy Bridge board with a similar mini-ITX Clarkdale board. There is also often a significant difference between boards using the same CPU, e.g the Intel and ECS Socket 1156 mii-ITX boards have a lower power draw than the Gigabyte and Zotac.

 

[Seero]

Hi,

I intend to build an Intel Sandy Bridge based computer that will be running 24/7. Therefore I want it to have as low idle power consumption as possible.

Now I just wonder if there is a significant difference in idle power consumption between the models that have lower TDP (e.g. 35/45W TDP) and those that have higher TDP (e.g. 65/95W)? Or is the TDP not a good indicator of the idle power consumption? I.e. is the Sandy Bridge CPU able to shut down the power consumption to very low levels when idle regardless of what TDP it has?

TDP is not a good indicator to idle power consumption. IMO CPU is not to worst power sucker in the system when it is idle. TDP is the maximum current draw by the CPU that is said to be safe, and Turbo mode is governed by that. If you are a nerd, have no problem messing with bios, and knows how to, you are specify the amount of voltage to CPU instead of letting it sit at auto.

Mathematically speaking, say the idle power draw is 50W. Say you mess if voltage and saved 5Watt (that is 45W drawn by CPU at idle), and say electricity cost you 8cents a Kilowatt. 5watt * 24 hr * 3 days = 3.6kilowatt. 3.6 * .08 = 28.8 cents a month or 3.456 dollar a year. That is, if you can reduce Vcore by 10%, you can save 28.8cents a month! Suppose the total system draw is 100 watt only, running 24/7 means 28.8 * 20 = 5.76 dollars a month, or above 70 bucks a year.

So if saving electricity in the name of green or cash, you should shutdown your PC. If you want your PC to be ready when you need it, you can have it in standby mode, where most of the PC is shutdown exception RAM(in abstract), so it takes around 5-10 second to wake up. Have OS to put PC to sleep after 10-30 minutes of idle and if it is a home fileserver than set it to "wake on LAN" or "wake on WiFi".

(note that it should actually by watt-hour and KWH, but i drop the hour to make the case simpler.)

 

[Concillian]

Is idle power in Sandy Bridge significantly lower than Clarkdale?

I haven not yet seen a review which, say, compares a mini-ITX Sandy Bridge board with a similar mini-ITX Clarkdale board. There is also often a significant difference between boards using the same CPU, e.g the Intel and ECS Socket 1156 mii-ITX boards have a lower power draw than the Gigabyte and Zotac.

Clarkdale CPUs themselves draw so little at idle, there can't be a big difference between them. Clarkdale is well under 10W idle on the CPU itself, probably closer to 5W.

1 Watt may be 20-25% and maybe SB idles that much lower. 20-25% sounds significant, but in reality is 1 Watt really significant? I think it's gotten to the point where there are not significant gains to be had from the processor.

The only place left for significant gains are South bridge and motherboard components. My guess is that these don't idle well because things like PCI or USB or whatever probably don't have the same idle state functionality that a CPU does. It's like a stack up of <5W devices that combined make a significant amount of power, but to add idle capability to each individual piece is asking a lot.

Wonder how notebooks do it, they seem to be able to idle at like 15-20W for the whole system, including LCD screen, so it must be possible, but then they don't have to support nearly as much backwards compatibility. I mean, when was the last time you saw a notebook with a Floppy connector, an IDE port and a PS/2 port?

 

[lemmo]

Clarkdale CPUs themselves draw so little at idle, there can't be a big difference between them

That's what I thought, I just haven't seen a review that compares them.

I think it's gotten to the point where there are not significant gains to be had from the processor

I agree, and it's clear that some motherboards are more efficient than others, even with very similar specs/components.

I suppose laptops achieve efficiency through careful choice of components, and locking things down in the BIOS. For a desktop board you'd be looking at BIOS options that allow you to switch off components you don't use, e.g. USB3 controller, and underclock graphics etc. But all a bit of a chore...

 

[Fjodor2001]

Well, 2TB 5400RPM "green" HDs can consume as low as ~3W when idle.

I have one server based on the following hardware: Atom D510 / 2 GB RAM / 2x2TB 5400RPM "green" HDs, and a "standard" non-high-efficiency PSU. That system idles at around 28 W.

I intend to build a new very similar one where I change to a 1155 socket motherboard, an i3-2100/i5-2400, 4 GB RAM. Shouldn't that system be able to idle at around 35W?

35W idle? 7W more going from an Atom platform to Sandy Bridge? That's pretty unrealistic. 5-10W more just from the SB mobo is quite likely. ~50W+ is a more sensible target.

~50W idle may be doable; 35W on SB is probably impossible.

Check out the Power consumption graph for the Gigabyte H67MA-UD2H motherboard:

http://www.anandtech.com/show/4241/h67-a-triumvirate-of-tantalizing-technology/11

The Sandy Bridge i5-2500K system idles @ 33 W. Hate to say I told you so... :biggrin:

 

[Zap]

35W idle? 7W more going from an Atom platform to Sandy Bridge? That's pretty unrealistic. 5-10W more just from the SB mobo is quite likely. ~50W+ is a more sensible target.
...
~50W idle may be doable; 35W on SB is probably impossible.

You are so wrong.

My mom's system is a Pentium G9650 (basically a cut down Core i3 Clarkdale) with an SSD and using IGP. It idles in the 30W range. Heck, Intel themselves were showing off such a mini ITX rig when Clarkdale launched that was idling around 32W.

Silent PC Review has estimated power draw from the power supply using calibrated equipment and they got as low as 17W for the whole system. This was using a 5400RPM notebook HDD, integrated graphics, ultra low voltage RAM and a Core i3 2100.

The Sandy Bridge i5-2500K system idles @ 33 W. Hate to say I told you so... :biggrin:

You know you love to! :twisted:

 

[Matt355]

Ive been considering the same thing, check this review.

50W idle may be doable; 35W on SB is probably impossible.

Check out the impossible.

http://forums.overclockers.com.au/showthread.php?p=13030093

Main Components
1. Intel Core i3-2100T
2. RipJaws 2 x 4GB DDR3 (That's the only spare ram I have lying around, I know it's kind over kill )
3. 256GB SSD (Salvaged from 2009 MBP)
4. Zotac H67 ITX-C-E (B3)
5. Windows 7 Pro 64bit


CPU Power Consumption

IDLE - 21.9 WATT

LOAD 100% IBT Test - 47.9watt

 

[Suspicious-Teach8788]

Sorry for bumping. Iw as doing some googling about low power systems.

You can get idle down to sub 40 watts with a SB processor:

http://www.xbitlabs.com/articles/cpu/display/core-i5-2500-2400-2300_10.html

The question is what OP wants....If he uses integrated graphics, ~30 watts is possible.

http://www.tomshardware.com/reviews/d510mo-intel-atom,2616-11.html - Toms got an i3 down to 30W.

This is say if you run a bare bones system designed for a file server or whatever and you're benching idle conditions.

From personal experience...

my i3-2100 and i5-2400 idle at about the same wattage. This is on a NAS with 7 HDs spinning, DVD drive, an 5 fans (excessive). Zero power optimization except for CPU idling. I idle at 70W with both CPUs.