SPCR

It might be just me, but it seems a little dissapointing that the best solution in many tests was a standard i5-2400 that was undervolted. I am running this setup right now and I appreciate the performance. It's mind boggling how low the idle temps/power is now. I can't really see anything that these special versions bring to the table that you can't get from an undervolted/underclocked sandy bridge CPU. They are just done so from the factory.

djkest wrote:It might be just me, but it seems a little dissapointing that the best solution in many tests was a standard i5-2400 that was undervolted. I am running this setup right now and I appreciate the performance. It's mind boggling how low the idle temps/power is now. I can't really see anything that these special versions bring to the table that you can't get from an undervolted/underclocked sandy bridge CPU. They are just done so from the factory.

yup

they are more for OEMs than enthusiasts

DAve_M wrote: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 just don't see that as so. But I would be pleased if someone would correct my logic.

1. The 2100, 2100t and even the 2600K all idle at about the same 4 watts. Even with undervolting it how much better can you do? 10%? 20%? lower. You are already darn close to 0 watts.

2. If you are trying to reduce power consumption at load, can't you do that with Windows power management?

Why all the fancy footwork? Why bother.

HFat wrote:The trouble is that determining the max stable undervolt is not trivial and highly time-consuming.

3. In fact why is everyone automatically reaching for a 2100 if a 2400 is available for not much more. Apparently they even downvolt themselves if they get too hot. If you aren't running benchmarks and doing overclocking.... the 2400 isn't going to use much, if any, more power than the 2100T almost all of the time.

cordis wrote:...so I did some investigating. It turns out that there's a feature on these chips where if the cpu detects that it's getting to hot, it throttles the speed down until temperatures come back down. I think in my case, I'm probably exceeding the amount of heat the cooler can pull off, and that leads to the processor getting throttled, which means I'm not really running at 4.8GHz. I suspect that a lot of what you hear about in terms of amazing speeds people are seeing overclocking these is probably due to that. The bios will gladly take the new value, but if you load the cpu too much, the speed drops automatically and you are none the wiser.... But what I'm really saying is, have a healthy dose of skepticism about numbers you see.

Remember that these are better chip samples. It's all good and well saying buy a standard version and under-volt it but there's certainly no guarantee they'll all go down to this level. CPUs are not all created equal.

What's more, as others have said, it may be possible to push these cherry-picked samples even lower. If it really does translate as $10-$20 retail price difference then might be a worthwhile investment if you're prepared to put in the effort to see just how low you can push an S or T.

Maybe it's a little unfair to undervolt the 2400 and compare it to a stock 2400S. However, I certainly wouldn't want to undervolt + stress test for a few hours for every voltage downstep for the purposes of a review And the sample variance is so high here it could leave some less fortunate buyers disappointed.

protellect wrote:
Though I admit, the stock AMD heatsink is ... painful.

Only if improperly inserted

Having finally reached a stable CPU-undervolt on my i3-2100T sample the result was -0.095v. Leaving it idle at ~0.860v and ~1.015v fully loaded. Done on an Asus P8H67M EVO motherboard.

Now to do the same with the GPU

djkest wrote:a standard i5-2400 that was undervolted. I am running this setup right now

How many fewer watts did your undervolting attain? I would not have expected it to have made any difference.

Due in huge part to this review, I'm building a machine [alas, not for myself] this week with an i5-2400 that I'll be undervolting on an ASUS Z68 board. I won't have it in hand long enough to do any exhaustive testing, but I hope to learn enough while building it to get me the next step closer to the no-moving-parts Sandy Bridge audio workstation I hope to assemble for myself. Great work, Lawrence!

I begin in this of undervolt. Currently only touched the offset, unless any of you tell me you have to change some other parameter.

I have a Asus P8H67M EVO with a i5 2500.

OFFSET:-0,070v
EPU ON

IDLE
ECO: 0,792v
AUTO: 0,824v
TURBO: 0,904V

PRIME 95
AUTO: 1,072v
TURBO: 1,136v

I have not yet done stability tests.
In a few days to measure the differences between applying undervolt or not.

As I have read, really the difference in consumption will be full with the CPU, and especially temperature.
A wonder with a slight undervolt converted our 2500 in a 2500S with more power.

Come on, We could do the masters offer us the details for a correct undervolt. Novices as I would be eternally grateful.

Thanks for a great revies.

jmqnick wrote:I have a Asus P8H67M EVO with a i5 2500.

Be warned that the latest 1850 firmware fixes a "serious" voltage bug on this motherboard.

In previous BIOS versions the CPU would idle at dangerously low voltages after a hard-reboot, and if you had set the undervolting options directly in the bios (not using the TurboV software) this would lead to the machine hardlocking because the vcore could drop to low. This happened to me and was frustrating as h*** when I was trying to find the undervolting sweetspot.

Jordan wrote:Remember that these are better chip samples. It's all good and well saying buy a standard version and under-volt it but there's certainly no guarantee they'll all go down to this level. CPUs are not all created equal.

It's not all cherry picking, market forces are also at work. Most i3-2100T can go lower than most i3-2100 but I'd say you would be extremely unlucky if you found that your i3-2100 wouldn't run stable at i3-2100T settings.

Locklear wrote:

jmqnick wrote:I have a Asus P8H67M EVO with a i5 2500.

Be warned that the latest 1850 firmware fixes a "serious" voltage bug on this motherboard.

In previous BIOS versions the CPU would idle at dangerously low voltages after a hard-reboot, and if you had set the undervolting options directly in the bios (not using the TurboV software) this would lead to the machine hardlocking because the vcore could drop to low. This happened to me and was frustrating as h*** when I was trying to find the undervolting sweetspot.

If I have noticed that sometimes the computer hangs at startup, the LED lit up without loading the bios. Will update the firmware, thanks.

Please can you tell me what the parameters are modified in the bios for undervolting?.

UPDATED THE BIOS

I updated the new firmware, and have left the same parameters undervolting, but not with the same voltage than the previous version.

Before a value of -0070v obtained at idle: 0.792v , now it does not lower to 0.880, seems to have increased the voltage considerably.

As someone who's less technically literate than most of you in terms of cooling, can I say that I'm pretty disappointed with the review. For a site that has the word "silent" in the domain name, I didn't see the mention of the word "fan" or "noise" even once.

It would have been nice to summarize if one of the two CPUs is capable of running the quietest or near the quietest so far (roughly in its power range), and also if one can even go fanless with it (as my apx 5 year-old Q6600 can be noisy sometimes). I have a PSU with a very quiet fan, so my PC does have one fan, but I'd like to avoid another for the CPU... maybe the 2100/s is the best? If so, it would have been nice to summarize (again for less knowledgeable users like myself) that the 2100 is the one to go for if silence is paramount.

On another note, I'm surprised Amazon are still stocking the Q6600 after all these years considering it's also 50-100% more expensive than when I paid for it.

No Sandy Bridge can go all passive (in a normal build) but "semi-passive" is doable and you don't need one of the factory underclocked T or S-versions for that. Those underclocked versions are for special uses, if for example you want to stay within a certain power envelope or want to build very small and quiet.

twinbee wrote:if one of the two CPUs is capable of running the quietest or near the quietest so far (roughly in its power range)

There are other CPUs which consume less. Of course with lower power consumption comes lesser performance.
But for multi-threaded loads there's a trick. The ratings are only an indication so if you get the 2400 and clock it down to 1.6Ghz as low as possible for instance, it should consume less at load than a 2100 that's not underclocked and should get roughly the same multithreaded performance.
Note that it's not clear you keep much of the low-power CPU's efficiency advantage if you underclock.

If you want a small amount of heat, you could get a 2100T (or a cheaper CPU) and underclock it but you'd probably get better performance with mobile CPUs which don't need to be underclocked as much.

Thanks, yes what would be ideal is if I could have the power when I need it, but then for the rest of the time switch down to a lower gigahertz rating so as to avoid having to use the fan.

I wonder if there's a CPU which lets you choose a completely arbitrary GHz level from anything from say 100 Mhz to say 5Ghz. If the fan adjusted automatically too, that would be amazing.

twinbee wrote:Thanks, yes what would be ideal is if I could have the power when I need it, but then for the rest of the time switch down to a lower gigahertz rating so as to avoid having to use the fan.

That is basically what a stock Sandy Bridge does... and that is why they all idle at about 4 watts.

They have have similar bottom ends (is 3.5 watts really that different from 4.1 watts?). It is only at the high end that they differ much from each other.

These Sandy Bridge processors have HD 2000 graphics, but there are CPU variants that have the better HD 3000 graphics: the i3 2105 and the i5 2405S.

Does HD 3000 consume more power than their HD 2000 equivalents, at idle and under load?

That is basically what a stock Sandy Bridge does...

Yes but can you force them to that speed (or any speed for that matter), or do they only drop to that speed when you aren't using the CPU? I might want a slow speed even if I'm pushing the computer to its limits.