i need preferably fanless pc for browsing/Office etc
1. i need mobo with ddr3 and sata 3, mATX
2. cpu with passive cooling, and maybe with lower TDP (i3-2100t with 35w?) or maybe my current i5 2500k with stock cooling and scythe ninja fan to fit in slim case
3. case itself and PSU (maybe pico psu?)

Get a G620 Pentium CPU for $64. You really don't need any more horsepower for web browsing and MS Office. Undervolt/underclock it. Or get the G620T for a specc'ed 35W TDP. I doubt you'd save more than a few watts on application load, though.
Anandtech chart:

There are cheaper Sandy Bridge CPUs now.
And that chart is close to useless. SPCR's are better for low-power stuff.

The problem is that you can't use a huge heatsink with room on top in a small case. So if you don't want any fans and you don't want a solid metal case to use as a heatsink with a heatpipes matching your board (see SPCR's reviews of such cases), you want less than 20W of power consumption. You can get that with Atom, Zacate and so on. You can get that with expensive and rare mobile gear.
Or you can get that by underclocking efficent mainstream gear such as a Sandy Bridge CPU. If you're going to underclock, the only reason not to pick the cheapest CPU is if you want a feature that comes only with the more expensive ones. But I haven't seen any such build report. The trouble is sourcing a suitable heatsink (fairly small but not too small yet designed for fanless cooling). Let me know if you find a successful build report...

DDR3 and SATA3 are utterly useless for office and web stuff by the way. Utterly.

Yeah, he can save $10-15 and get a G5xx. I was pointing out that he can get a $60-70 processor to meet his needs rather than a $200 one (i5-2500k reference by OP). The chart is also a reference to compare load power of the i5 vs a pentium. If there was an SPCR article with a SNB based Pentium I would have shown it...


Nonetheless, the OP gets his wish - they are on every mobo if you build a SNB based system.

Yeah, it's worth pointing out that the Pentium consume less but I thought that would be obvious since they're slower. What's needed if you want to pick anything but the cheapest dual-core is reliable numbers. Anandtech's number exagerate the difference between CPUs unfortunately. Even the G620T gets hot.
The SPCR article in which they looked at low-power Sandy Bridge CPUs has numbers which are more useful I think. You can cross reference these with the measurements made by random people on the web who measured a stock 2100 or 2100T as well as Penitums and/or undercloked i3s. Numbers less absurd than Anandtech's are not hard to find.

Sharing some specific link will be appreciated. Or more appreciated. If known, obviously.

FWIW and just for future reference (though it's more an HTPC-centric comparison and providing some lack of cross-platform consistency due to comparable but different setups), I would advice for some MissingRemote power-wise findings about Intel 2100T, 2390T, 2400S, which can be compared also with the 65W AMD A8-3800 (there is also a smaller set of figures for 2500K, IMHO not so useful).
The relevant power numbers were usually measured at the wall with a P3 P4460 Kill A Watt.

XBitLabs charts:









NOTES on Celeron G440: it has EIST disabled by Intel. The comparison with Atom and Zacate was performed with the ASUS P8H61-I, while the other is with ASUS P8P67 DELUXE.

Testbeds:

* Processors:

o AMD Athlon II X2 250 (Regor, 2 cores, 3.0 GHz, 2 MB L2);
o AMD Sempron 150 (Sargas, 1 core, 2.9 GHz, 1 MB L2);
o Intel Celeron E3500 (Wolfdale, 2 cores, 2.7 GHz, 1 MB L2);
o Intel Celeron G540 (Sandy Bridge, 2 cores, 2.5 GHz, 2 MB L3);
o Intel Celeron G440 (Sandy Bridge, 1 core, 1.6 GHz, 512 KB L3);
o Intel Core i3-2100 (Sandy Bridge, 2 cores, 3.1 GHz, 3 MB L3);
o Intel Pentium G850 (Sandy Bridge, 2 cores, 2.9 GHz, 3 MB L3);
o Intel Pentium G620 (Sandy Bridge, 2 cores, 2.6 GHz, 3 MB L3).

* Mainboards:

o ASUS Crosshair IV Formula (Socket AM3, AMD 890FX + SB850, DDR3 SDRAM);
o ASUS P5Q3 (LGA775, Intel P45 Express, DDR3 SDRAM);
o ASUS P8H61-I (LGA1155, Intel H61 Express);
o ASUS P8P67 Deluxe (LGA1155, Intel P67 Express);
o Gigabyte GA-D525TUD (Intel Atom D525, Intel NM10 Express);
o Gigabyte GA-E350N-USB3 (AMD E-350, AMD Hudson M1).

* Memory — 2 x 2 GB DDR3 SDRAM (Kingston KHX1600C8D3K2/4GX):

o DDR3-800 5-5-5-15 with Atom D525 and Celeron E3500 processors;
o DDR3-1067 7-7-7-21 with AMD E-350 processor and in the integrated system with Celeron G440;
o DDR3-1600 9-9-9-27 in all other cases.

* Graphics card: ATI Radeon HD 6970.
* Hard drive: Kingston SNVP325-S2/128GB.
* Power supply unit: Tagan TG880-U33II (880 W).
* Operating system: Microsoft Windows 7 SP1 Ultimate x64.
* Drivers:

o Intel Chipset Driver 9.2.0.1030;
o Intel Graphics Media Accelerator Driver 15.22.1.64.2361;
o Intel Graphics Media Accelerator 3150 Driver 15.12.75.50.64.2230;
o Intel Management Engine Driver 7.1.10.1065;
o Intel Rapid Storage Technology 10.5.0.1027;
o AMD Catalyst 11.8 Display Driver.

Not known but 20 seconds of Google yields one which has already been discussed here (disclaimer: I don't endorse it):
http://www.avsforum.com/avs-vb/showpost ... count=1473
I bet there are more.

Some of xbitlabs's numbers are pretty absurd as well (I think I've seen worse on their site by the way). If you do believe their numbers (I wouldn't) and want significantly better single-threaded performance than you can get with Atom or Zacate, the G440 seems like a good candidate for fanless operation in spite of the lack of EIST. I expect the new Atom D2700 would perform better on most multi-threaded loads while consuming less power however.

I would advise to limit sometimes the risk of sounding haughty, maybe even unfairly: here we don't share - enough well, at least - the same language, how can we undoubtedly share the same threads, or the same search terms (not to mention that the very same words would gives different SERPS on google.it, google.ch or google.com)??


Are you referring to the power consumption figures of all those CPUs?
If in case, picking the 2100 relevant figures, you may see: SPCR, idle 31W, load 79W (DC estimated) - AVS idle 39W, load 74W (AC measured) - XBL idle 41W, load 79W (AC "measured").
At first sight, even giving all credits for setup differences, the latter ones doesn't look like absurd "de plano".
Maybe you could try to set forth with details your thoughts.


I wish that... unfortunately, it will not come for cheap, as it will cost as much as the dual core 2.5GHz Celeron G540.

Compare the G540 over 2 charts and the G440 over the 4 charts. I can't make sense of those values without assuming artifacts caused by the interaction between their PSUs and meter. Maybe you can.

Atoms are sold with a board so they're cheap when you compare them to the total cost of a mainstream board and a mainstream CPU (even a Celeron). But they're not as good. The point of Atoms is the power consumption. They only compete with mainstream CPUs which are underclocked and limited to a core. If that's what you need to do to get passive cooling, Atoms could be a better option.
But ideally you'd want a G540 (or better) instead of the G440. Atoms can't compete with that.

I'm not sure to have well understood you, but anyway, the G440 on the P67 high end board sports an HD6970 (more than 20W at idle), while on the basic H61 run off the IGP: the difference between the two scenario is around 30W, not unreasonable at first sight.
Were you talking about this?


You're right, and there are evidences that Cedar Trail will be the cheapest Atom ever.
At anyway, on NewEgg something like an ASUS D525+NM10 is around 95 USD (maybe an Intel might cost some bucks less), while a G530 plus a FoxConn/BioStar/ASRock H61 goes for 105-110 USD. G540 accounts for 5-10 USD more, I suppose. Not so close, but not so far: Intel probably doesn't love too much the Atom.

Sorry to have gone too deep into OT.

If you want to buy an Atom board, don't buy an Asus. They're expensive and come with a large heatsink but it doesn't work well. Asus also made overpriced Zacate boards with flashy heatsinks which don't work as well as those of their competitors. It's not Intel's problem. Their fanless Atom boards sell for $50-70 (depending on the model) and their unassuming heatsinks are adequate.


Look at the values I pointed at. What idle power consumption would you expect for Sandy Bridge dual-cores at idle with the H61 board based on the differences between CPUs with the P67 board? It is a plausible value? I think not.
They don't tell what PSU they used which makes their numbers dubious (these numbers can't be AC for the 880W you mentionned). Maybe they used a bogus correction for PSU inefficiency. My recollection is that they did so in other articles, with absurd results.

Or, for better energy efficiency, the OP could buy a Pentium based laptop and use it 'docked': external keyboard/mouse/monitor. I just saw an HP listed for $400 via Dealzon. Don't know how quiet it is..then again, the load of the OP's apps are low. He could always replace the 5400rpm HDD with an SSD for a bump in performance- it's also an opportunity to shed all the bloatware.

I start to see your point, but looking at charts 1 and 3 (Windows idling), at least for the PSU (in)efficiency: maybe it's worth to note that in their E-350 review, Xbit Labs declares the same PSU, an 80+ Bronze unit.

A tad suspicious, indeed: if it were as efficient as my Antec Signature (80+ Bronze), it should be under 55% at 15W AC level (230V), if it were as efficient as my Seasonic M12D (80+ Silver), it should be under 65% at same level/voltage... say definitely 55%? Let's try.
DC should be ~8W for the entire system: as an i5 draw just from the 12V line about 5W (according to Lost Circuits, a credible source for me, up to now) it doesn't sound fully right (even if I have read a 15W AC at idle for the 2100T from Missing Remote using - IIRC - an Antec 380W PSU), but comparing these figures with the 18-20W AC level recorded for Atom/Brazos by SPCR (with a PicoPSU 120!), it becomes even more suspicious.

Even about the difference between the P67 and H61, I see some further questions: in the only review with power data I was able to find about the H61 (TweakTown), the relevant board (Gigabyte mITX) draw 15W AC less than an H67 one (ASUS mITX) using the same 2600K CPU. If we add the ~20W DC usually draw at idle by the graphics card vs the presumable IGP power consumption, the difference should account for about 40-50W AC less than... (the Xbit declared) 45W AC for the G440+P67.

I'm not sure about my nightly math, but I have more doubts than before, now.

I don't trust generic power meters for PSUs operating at a small fraction of their capacity. I got suspicously low readings in some cases as well. But these low-wattage AC readings aren't merely suspicious: they're absurd.

The only credible low AC power readings are done with low power(<75W) level IV-V bricks and PicoPSU or equivalent; the EPA level categorization is similar to the 80+ ratings, requiring efficiency measurements @20, 50 and 90%, IIRC. Both level IV and V are rated with the efficiency over 80%@20% of load, so if you want to estimate somehow accurately the AC power consumption at low levels, this configuration is a must. If a (low level)power measurement is not conducted in a similar way, it's not worth the carrying electrons, not to speak about the time spent reading it.
Sorry for the OT, but I hope I mange to shed some light on the problem.