Just to clarify then, the 35W T version only saves about 5W under load than the normal 65W version?

Seems like there is not much reason to chose other than the lowest cost 4 core Sandy Bridge.

Even if you are constrained by peak load PSU limitations, given that you can control peak load through Windows, it is pretty hard to argue for choosing those low wattage Sandy Bridge chips.

The table on page 1 is obsolete already.
If retailers are to be believed, some of the new Sandy Bridge CPUs are already in stock in some shops over here.


No. Page 2 of the review: the difference is 12W.
And that's not the whole story. The review understates the differences (I understand why and I'm not saying it's dishonest).

The real questions are:
-can the board you'd want to use undervolt? how expensive would a board with that feature be?
-how much difference is there between a regular and low-wattage part if you undervolt both? how much difference is there from one sample to the next?
The last question would be very expensive and time-consuming to answer. If you don't want to bother with the minutiae and don't need the performance of the regular part, paying 5-10$ extra for the low-wattage dual-core seems to be the obvious choice. If you could use the higher performance, then you've got a real choice to make and more information would help.

"The real questions are:" Does peak load wattage matter?

Answer: No

TDP != instantaneous peak... though no one (manufacturers) seems to want to define TDP at all.

And peak sustained obviously would matter (for thermals, power supply)! Depending on why you are building the machine, "winging it" with an aftermarket under clock/volt might be a risky venture.

Edit: I do agree, though, that load power is irrelevant when considering energy costs for many/most systems.

"This has been our biggest criticism of these specialty low power processors — you can create your own with simple undervolting on a decent motherboard."

I don't think you can undervolt with an H67 based motherboard.

I don't see the point to comparing the lower TDP CPU versions to regular CPU chips with the test setup used. I think the following test setup would be more relevant

H67 motherboard
no graphics card
low power hard drive - 2.5" or a 5400 RPM 3.5"
stock CPU heatsink/fan and a passive alternative
case with one 120mm fan

The reason I'd consider an S or T variant would be to allow me to use passive cooling in place of one or more fans in the system. Or a case that was more restrictive. If the CPU chip's peak power use is lower, I can consider cooling approaches that might be somewhat risky with a regular 95W TDP chip.

Bill

Uh yes you can undervolt a H67 mobo. See the Asus Review right below this one :ppp

And for us silencers, I'd say undervolting is by far the best solution over paying for an overpriced low voltage version from Intel.

Thanks so much for another great review! I've been watching these low power consumption chips for a while, and was very interested to see how much power they actually save.

Not sure if you still have the chips around, but I'd be very curious to see whether the use of integrated graphics effects power usage differently between the two sets of chips. For example, you could compare power usage under GPU load between the 2400 and the 2400S. A corollary would be to test the 2405S with more powerful graphics (if you can get your hands on it, of course ).

Thanks for the informative review! It's great that lower wattage desktop chips are becoming more common-place. I have great love for my 2.5GHZ AMD 45w chip [X2 4850] as well as my 1.9Ghz BE-2300.

I know that there might not be any microcenters wherever the review takes place, but even now you can pick up that Phenom II X4 3.2Ghz *with*, I admit, a mediocre motherboard for just over $100, which is almost twice as much as listed on the review. I ended up doing a whole PC for just over $250 including a 3.2Ghz Quad-core, Antec Earthwatts 380w, 4 GB memory, 500GB hdd....

Though I admit, the stock AMD heatsink is ... painful.

So what is the peak power consumption of the 35W CPU? I'd like to compare it to the 65W version and max stable undervolt of maybe a quadcore mainstream part.

I actually buy ASUS motherboards religiously for their BIOS features. What other manufacturers tend to include undervolt options? I know I've been disappointed by Gigabyte in the past

No idea for Socket 1155 as I'm very close to buying my first ones since I have two Core I3-2100's with no mobos sitting here.
Gigabyte did used to have undervolting features. I used them religiously with my old P35 boards, undervolting + overclocking = best of all worlds for me. I kinda wish we had some sort of comprehensive motherboard features wiki, but that would be hell to maintain for sure.
I do miss Abit and DFI for having some very good enthusiast features, even if I almost never bought their stuff.

The page 1 comparison chart is wrong. It lists the Core i5 2500 and 2500K as having 8MB cache, when it has only 6MB.

Thanks, Lawrence. Great comparison of 2400 undervolt vs 2400S.

there is a 5C degree difference between 2100 and 2400. that can make a difference for building a fanless system.
it would be interested in compare them in a desktop maybe with 1 fan at the exhaust, no other fans on the CPU or intake.

you reviewed a fanless system a few days ago. if you still have that one, maybe you can just swap the cpus
and run a simple load test.

if the overall system temperatures are not much different, that would be a better judgement against 2100.

As usual, power dissipation is the key to understanding the thermal issues. In CPUs, they are one and the same. Lawrence has documented clearly that...

1) There is no difference at idle (ok 1W, but that's not a difference imo)
2) There are more substantial power draw differences at higher loads, but the maximum 12W difference between the 2100 models and 14W difference between the 2400 models is not big enough to make the standard models "unsuitable" for fanless cooling. It is not a black or white issue, but a matter of degrees. With the right cooling hardwar, any of them can be fanlessly cooled. Perhaps some would run a bit hotter or it might take a bit more work to make them all run at the same temp at load.

These distinctions will diminish further with the arrival of the Ivy Bridge chips which are supposed to use half the wattage of the Sandy Bridge chips... while processing even faster.

Maybe they only put it on their fancy boards? Or certain boards meant to have low power features? I have a Gigabyte LGA 775 board, and had a few Socket A and Socket 939 boards, and they all only allowed voltage increases (not decreases). Was a bit frustrating, since clearly the hardware supported lower vcores for other chips.
But then, none of those were particularly recent purchases, either.

Yes, yes. But it is useful to know exactly how much power you are dealing with from the CPU for cooling and PSU requirements. Measured on the 12v line to the CPU. 12W difference under load. Ok but that's system power. Can we just assume that:

- A conservative estimation for PSU & cooling is 35W for the T series.
- The regular version only requires 47W for PSU / cooling.

Why not undervolt AND underclock the 2100? See if it's basically the same chip as the 2100T. Rumour is the 2100T is binned for better efficiency, undercloking and undervolting a 2100 could reveal the truth.

Great review! I'm going to be building a HTPC in the near future, and I've actually decided on a slightly different CPU. Have you looked at the i5 2500T? Here are the spec sheets from Intel on the i5 2500T and the i5 2400S (the one you reviewed).
As far as I can see, these are the only differences...

2500T: 45W TDP
2400S: 65W TDP

2500T: 2.3ghz clock, 3.3ghz turbo boost
2400S: 2.5ghz clock, 3.3ghz turbo boost

2500T: 650mhz base graphics, 1.25ghz max graphics
2400S: 850mhz base graphics, 1.1ghz max graphics

These things should perform pretty close to the exact same, but I love that the TDP is only 45W for the 2500T. I find it interesting that the 2500T has a "higher high" as well as a "lower low" for the graphics chip.

It seems like the 2500T is tough to find right now, but I'd absolutely love to see it thrown into the mix. I think it is going to be a bit more expensive, but I'm fairly certain that is the chip I'm going to go w/ for my HTPC.

heh.

That might be true if you are a thermal engineer designing for complete precision, making maximum cooling only just good enough to cover the worst case. But it is far more practical to take a more general approach, and it is all most of us can do. For years, we have espoused the use of the biggest, baddest heatsinks with the slowest, quietest fans for good reason: It works. With this approach whether the CPU is 35W or 47W at max load hardly matters for cooling, especially with current gen coolers. And I suppose that in some cases when using a just-big-enough PSU, +14W could cause an overload -- but it really should not, there should be safety headroom to cover that.

Probably a safe assumption. (We have documented for years how Core Intel chips are impossible to get even near the rated TDP with stress tests, never mind normal apps. Sandybridge cores continue this trend.)

I would have to disagree. Considering this is SPCR, what I had in mind would be along the lines on pico PSU and external power brick. It would be helpful to know how many amps you are dealing with as there can be a limited amount to go around depending on the choice of GFX card.

Or if you want to go fanless then the CPU power consumption becomes rather important. Maybe it is only 25W? Probably not. But is helpful to know.


What was concluded was the T series is poor value, just undervolt a regular CPU. But if the T series are from the golden bins, you should do a comparison of max stable undervolt of the T series vs a normal part. Maybe there is a difference in which case the conclusion would need to be changed IMO.

I actually went with an i3-2100T for that reason. I have a few 60W bricks and I would like to use one of them for the system. Price for i3-2100 and i3-2100T were about the same and since I knew I wouldn't be able to underclock/undervolt much I went with the factory underclocked version since the performance is way overkill for my uses anyway. I wanted to minimize peak load (at startup, when all the disks are spun up). But as you say, these cases are rare.
Today I'm thinking I should perhaps have waited for the cheaper Sandy Bridge's, but I was impatient.

No. As I and others have said, the actual diffrence between maximum power consumption is going to be higher.

I wouldn't be surprised if the 2100T's TDP was close to 35W. It's the fastest 35W CPU of that type. But I expect the actual power consumption of the 35W G620T to be lower because it's slower.
The 2100 is not the fastest 65W. So I'd be very surprised if its maximum consumption was anywhere near 65W. Intel does not usually rate CPUs individually the slowest models of the bunch always consume less than the more expensive ones. So the G620 should consume even less than the 2100 even though they're rated the same.

Fair enough. Had not thought of that. May try a quick undervolting of the low TDP parts just to see...

The trouble is that determining the max stable undervolt is not trivial and highly time-consuming. The result also varies from one sample to the next.
If you are prevented by arbitrary limits from going near the max stable undervolt, it makes the determination considerably easier of course.

As others have pointed out, at stock there is no real reason to get the low power derivatives of SB chips, they are all very power efficient anyway.
But if the 2100T can undervolt even further on a mid ranged H67 board, could be worth the $10 extra.

Either way, i was very tempted to get one until i saw SB Pentiums announced recently, hoping they will be even more power efficient thanks to low clocks and turning off features. Its for an HTPC, any SB chip will be good enough performance-wise.