Hi. Im interested in knowing how well desktop sandybridge (i5-2500k,i7-2600k) undervolt at stock speeds, and lower frequences. Has anybody tried it yet?......cheers.

I just ordered my 2500K with a P series motherboard and the very first thing I will do with it is see how low it can go at stock speed. I will post detailed results on the forum in the style of my MSI 5770 review.

Sandy Bridge, Part 2: Intel DH67BL & Asus P8H67-M EVO H67 Motherboards

SPCRs review included the Asus board where the EPU utility undervolts the CPU by about 0.05V at idle and 0.8V under load. I would guess that you can probably undervolt more. But like withe the Clarkdale CPUs, there's little to no change in idle power. The modest undervolt did help bring the load power down by about 4W on the lower powered CPU and 8W on the higher clocked one.

Thats great...thanks

yeh...im hoping they can be undervolted to use no more than 1.0v under load at stock speed.....looks promising.

OK, I just got my i5-2500k setup with Asus p8p67 PRO mobo. I've got prime95 running on all four cores. CPU-Z reports 3400 MHz, Speedfan reports 1.03V (set 1.09 in bios) and cores currently at 44 C. I have this on a Prolimatech megahalems with no fan so I expect it to max out a few degrees hotter with enough time.

We'll see how hot it ultimately gets and whether it can sustain a whole night of prime95, but my initial impression is that this already far exceeded my expectations. If it gets through a few more hours of prime95 with no error I'll undervolt some more.

Edit @ 7:47 PM, a few minutes later and we seem to have maxed at 48 C. Cores 0/1 are about a degree cooler than cores 2/3. All I can say is holy shit, do I have a golden sample? This thing is producing at least 220% of the raw compute performance of my E8400 with even lower thermals. And this is supposed to be 95W TDP versus the E8400's 65W?? I get the feeling I haven't even approached rock bottom voltages yet.

maybe you guys can help me to clear things up a bit. I got a 2500K and an Asus P8p67. In the Bios I enabled EPU and there is also the Offset Option. If I understand that correct then offset is the new way to Undervolt right? It determines the voltage difference regardless if the CPU is in idle or full load.
So I just enable EPU and for further undervolting I experiment with the Offset to see how much I can push there?

edit: this is what looks stable so far. EPU on Auto, Offset to -0.09V. Prime 95 large FFTs test (don't know if this is the right one) 52° (from CPUID HWMonitor)@1.024V (from CPUZ) cooled by a Megahalem with a Scythe Slipstream@5V

edit2: @Silentplummet: which tool did you use to meassure temps? If it was speedfan only the beta version 443 show the correct temps. 442 will display 10 degrees less than it is (on my P8P67 at least).

Here these guys did some undervolting
http://www.computerbase.de/artikel/proz ... dervolting
(in german)

Thanks so much for this tip, I knew something had to be up but I had no reason to distrust what speedfan was telling me! It turns out I had been hitting around 61 C and not 48 C as shown. Not dangerous temperatures but still too hot for my liking.

I broke down and relocated a superfluous Noctua 800 RPM fan from the drive bay to the Megahalems (paradoxically, this reduced the noise it produces to nearly undetectable levels even at max RPM). This change and the minimum stable voltage I reached of 1.065 V (bios)/ 1.000 V (after vdroop in 4 core prime95 load conditions) brought the load temps down to 41 C max, confirmed by the speedfan beta and the recently released 2.6.36 Linux kernel. Now 41 C and blessed quiet I can live with.

Now begins the tedious scientific process of finding minimum stable voltages at various multiplier and turbo settings. I'm not sure exactly how to go about this as the motherboard offers so much flexibility. The turbo function adds yet another dimension to the testing. Apparently the turbo function at factory settings scales with the number of loaded cores. Max turbo on the i5-2500 is 3700 MHz but it will only attempt this with 1 core loaded. When you load all four cores it scales them down to 3400 MHz. Unfortunately the motherboard allows you to set only a base voltage and then an extra voltage offset for when turbo is enabled. This means you have to find a single voltage (base + turbo extra) stable over the entire range of turbo states.

It seems you can artificially limit the range so that under 1-core load conditions the turbo frequency target is the same as for fully loaded conditions, but this is untested and theoretical for me at this point. There are many settings in the BIOS related to turbo function and honestly I have no idea how any of them work in practice. The details given in the online help are characteristically unhelpful.

Update. Right now I'm folding at 4 GHz on the SMP core in amd64 linux. I set the voltage to 1.275 but have no means of checking Vdroop in linux. Average core temp is 57 C with the 800 RPM noctua fan I slapped on it. I think there is still some headroom with the voltage but probably not too much. The same setting was on the threshold of stability at 4.2 GHz. Will tinker some but I think I'm going to stick with ~4 GHz for my target settings. Can't beat the silence and the thermals aren't TOO bad. I'd like to get them under 55.

If they can be undervolted so well, is there any point in buying the expensive 2390 or T/S versions of the CPUs? Might as well get a 2500K on Z68 with overclock potential and better GPU, right?

I agree with you...plus you have the choice of increasing clocks and volts in the future, to keep up with software.

+1

I'm interested about this Sb cpu's. First i look S or T versions. But after reading this
http://www.xbitlabs.com/articles/cpu/di ... 00s_6.html
and this http://www.silentpcreview.com/article1143-page5.html
review i give priority to non expensive S or T cpu's.

Undervolting i can make alone too, question is only:
How consumption is your SB system? I mean whole system with cpu+ddr3+hdd+psu.

System i'm thinking about:
GIGABYTE Intel H67 + I5 2300.

I have undervolted my i5 2400.

Runs on 1.025V (~1.2 V stock) in full load. Temps are *very* low (especially in idle, around 5 C over ambient using Scythe Shuriken mounted with Thermalright Bolt-through kit).

Fan runs in 600-1100 rpm (100 mm fan) which is virtually silent.

Do you tried measure power consumption system with and without undervolted cpu?

I'm so interesting about I5-2500 K with min. undervolting-underclocking points.

It idles at under 4 watts. So what is the benefit of undervolting it? It undervolts and underclocks itself, automatically.

In either case (undervolted or not undervolted) it will ramp up if you put a load on it and in any realistic situation that load won't tax it even 50%. In either circumstance it will use the same wattage... just enough to process the load... then return to idle. Right?

So many flaws in your reasoning. Maybe some of us have usage patterns that don't fit the parameters of what you just described?

I run mine flat out, 400% load 24/7. Protein folding simulation for the SPCR F@H team. Moreover, this being SPCR and not HotHardware or HardOCP, I have an obsession with silence. I want to ensure that load temps don't exceed 45C* under ANY load for ANY period of time, and I want to do so while avoiding any cooling solution that would compromise the peace and quiet of my den.

(*achievable at stock frequency by undervolting. In reality my target max temp is 55C but I run at 4 GHz)

Also please realize that wattage (electrical power consumed) and processing power are not correlated. I could run my CPU at 3 GHz@1.0 volts or at 3 GHz@1.5 volts. 'Processing the load' will take exactly the same amount of time in both scenarios. One consumes 50% more power to do so.

The idea is to run the hardware as efficiently as possible by finding the minimum stable voltage at any given clock frequency. That's what we do when overclocking; when we do it at stock frequency we just call it a different name, "undervolting".

Moreover, 'minimum stable voltage' vs. 'clock frequency' is not linear. It is typically an exponential curve. Clock frequency vs. processing time is (ideally) linear. Therefore, given a CPU always running at minimum stable voltage, you will use more power for a given load the faster you try to complete it.

Basically you are wrong on about 3 levels.

Brief update on my SB performance / silence quest. I now achieved 4.5 GHz/ 1.28 V (load; 1.32 bios LLC high). Core temps are at 61 C (ambient 25 C) as I fold on the Linux SMP client. My system is now what I would call 'whisper quiet' whereas it was silent before... here's what I did:

1. Lap CPU and heatsink (megahalems) down to the copper, voiding over 9000 warranties
2. Polish with 1-micron honing compound to a mirror finish
3. IC Diamond TIM
4. 2x Scythe slipstream fans @ 720 RPM push-pull on HSF, case fans are 2x Noctua NF-P12 @ 810 RPM, 1x Nexus Real Silent @ 740 RPM

It's a lot of work (took me probably 5 hours total and ~$80 of equipment) but I managed to shave off 8-10C core and bring the temps down into the region I consider 24/7 safe (<65C).

This is a lot of performance for such a quiet system. SB exceeded my expectation in every way. I thought it couldn't get any better than my Core 2 E8400

11th hour update: 4.5 GHz isn't really 24/7 stable on my chip for folding. My chip hits the point of diminishing returns around 4.3 and I wasn't comfortable pushing the volts or thermals high enough for real stability at 4.5. I backed it down to 4.3 for 24/7 OC and real silence

Old thread but let's bump it anyway

Was doing some experimenting with the voltage on my 2500K last night.
Dropped clocks to 40x100 and started lowering the voltage.
1.152v at 4Ghz

Dunno if that's good or good but I like it. It was Prime95 stable with 1792k FFT with 4096mb mem load, good enough for me
Didn't go any lower yet, gonna try soon though, gotta just do some things first.


Have you guys undervolted your SB's more?
Imma see if I could go 1.1v with 4Ghz

topic may be old but really interesting!

i'm especially interested in findings about a 2500k at (most probably) stock speeds, I'd like to undervolt it to "silence" it as much as possibile; what are your "best" parameters?

I'm going to buy the 2500k for the ability to fine tune its performance to the use, but most of the time I'd like for it to be "dead quiet at full load". It will be provided with a shuriken.

since I'm probably going to use a z68 board (asrock pro3-m) I'd be interested in integrated gpu downvolting also, to bring the heat down as much as possible.


thanks a lot!!

I5 2400 OCed to 3.6 GHz on all cores (4 bins for 400 MHz + 100 MHz Turbo Boost) idles at < 0.8 V and works under load at 1.1 V. Naturally, the board increased Vcore when I increased the multiplier, but I used the available offset (- 0.200 V) to bring it down. This is on a GA-P67A-UD3-B3 board. Perhaps other boards have higher negative/positive offsets available. Still, 0.792 V @ 1600 MHz is not bad. I was initially hoping to get 2500K but either I wouldn't OC it heavily, but rather undervolt it and have it at around 4-4.2 GHz perhaps. Now, with the 3,6 GHz on 2400 I guess I'd rather save up for IB.

2500K stock consumes around 81 Watts, at 3.4Ghz 1.200V

According to this formula: 80W/(1.200*1.200)(1.050*1.050)~60W

At 1.050V it should be 20 watts lower, down to 60Watts As the load in Games and rendering would result even less impact to be only 10 Watts lower with undervolting. is it worth it? I'm stuck with non undervolting motherboard.