Possible to underclock a i3 by BLCK@66MHz

[sanbeiyan]

Hi,

I am planning to build a silent (with low RPM fan) and cool HTPC. My question is that,
is it possible to run the i3-530 with BCLK adjusted to 66MHz ? i.e. 22x66MHz = 1.45GHz
From various review, the i3-530 idle at 16Watts, and Full load at ~60Watts. With BLCK@66MHz,
I hope that it can be idle at 16Watts, however, full loat at ~30Watts, of course, with much less
heat and noise from a high rpm fan.

If yes, which board (GIGABYTE ?) allow me to do so ?

And what is the minimum value I can use ?

I remembered that I came across a forum that someone tried to run a i5 with 100MHz instead
of the default 133MHz and hence, saving about 33% power consumption, i.e. heatness
and noise !

Thanks !!

[alecmg]

Hi,

I am planning to build a silent (with low RPM fan) and cool HTPC. My question is that,
is it possible to run the i3-530 with BCLK adjusted to 66MHz ? i.e. 22x66MHz = 1.45GHz
From various review, the i3-530 idle at 16Watts, and Full load at ~60Watts. With BLCK@66MHz,
I hope that it can be idle at 16Watts, however, full loat at ~30Watts, of course, with much less
heat and noise from a high rpm fan.

If yes, which board (GIGABYTE ?) allow me to do so ?

And what is the minimum value I can use ?

I remembered that I came across a forum that someone tried to run a i5 with 100MHz instead
of the default 133MHz and hence, saving about 33% power consumption, i.e. heatness
and noise !

Thanks !!

From what I read, BCLK doesn't let adjust it too low. 93-95 being the limit.
But on the other hand I'm pretty sure you can achieve your goal by simple undervolt. SandyBridge don't need much voltage at all (my i5-2500 needs 1.085V on load, 0.140 less than default).

[sanbeiyan]

Maybe it is a little bit off topic.

But I just discover that the Celeron E33/34/3500 seems more easy to archive my purpose. From the datasheet, they are running at 200MHz x 12.5X/13X/13.5X, resp. Thus, if I set the FSB to 100MHz, the theoretically power consumption would be 50% off during full load.

That will bring up a box with idle ~25W and full load ~40W, assuming using SSD and on-board/low power vga.

Anyone tried this setup ? and comments ?

[fumino]

power consumption is more effected by voltage, which its tough to beat SB at. though i wont lie, i have heard tell of people with celerons that sip single digit watts at idle. so for the cost savings too, it might be worth it to try that route.

[alecmg]

slap me sideways, I'm becoming senile

Why adjust BCKL at all, when ALL Intel cpus have UNLOCKED multipliers below official spec.
You can simply change multiplier to get less MHz.

But I will reiterate again, lowering voltage gives more tangible results.

[sanbeiyan]

Actually, my build target is a 24x7cool PC which will be ran under a harsh env (no air con in summer, ~35 deg C ambient max).
Low power consumption + low heat is the target !
My initial research found the Zotac SU2300 hd-nd22 box, various review claim that idle ~25W and full ~30W. However.
it seems that this box is a bit suck that the ventilation/case is poor designed. Look like it cant survive under a harsh env without custom mod to the case ...

After some more study, I found that a modern i3 (or even last year's E3300 celeron) did a good job in power management during idle/low load. So, I am heading the road of underclock+volt of a modern i3/E3300. No idea if it could work for me.
But according to various post and my past experience (Pentium age), undervolt is not easy to success, or the V reduced
is marginal.

P=C * f^2 * V

i3-530 and E3300 at 50% CLK still out performance a SU2300, and the ideal power should be 40% off from original.

Tonight, just managed to borrow a MSI+E3300 from a friend. Did some tests,
Config: MSI G31+E3300+2G*1+2.5" samsung
Stock V+MHz:
idle: 25W from device (pf 0.4)
prime95: 45W

Stock V+MHz, disable EIST in BIOS (CLK Multiplier auto fall to 7X, software indcate it is 6.5X):
idle: 25W from device (pf 0.4)
prime95: 39W

It seems that the clock was downed nearly 50%, however, the power consumption just -12% !!!

Also, as the BIOS doesnt allow me to adjust Volt/freq, nothing more can be tested. RMCLOCK
failed to adjust the CPU volt from default 1.28V to any other value !

Any suggestion/advices ?

[Big Pimp Daddy]

Changing the BCLK on Sandy Bridge is a bad idea as all the system clocks are generated by the processor, change the BCLK and you change a lot of other things - hence multiplier-only overclocking.

You seem to have your Power formula wrong, Power is proportional to the square of the voltage times the frequency.

[sanbeiyan]

Was sleepy and typed a wrong power fomula !!!

Managed to get the RMClock working !!! And here is the result,

Ambient: 18degC

Stock V+MHz (1.285V/2.5G):
idle: 25W from device (pf 0.4)
prime95: 48W
cpu: 55degC

Stock V+MHz, disable EIST in BIOS (1.285V/1.2G):
idle: 25W from device (pf 0.4)
prime95: 39W
cpu: not recorded

RMClock (1.1V/1.2G):
idle: 25W from device (pf 0.4)
prime95: 36W
cpu: 38degC

OK, now, it seems that the board+cpu idle is taking 25W (It seems i5 is the same according to
Tom or Anand site). The CPU taking up another 20W at full load in default setting, while
running at half freq and 85% volt, it saved 10W. It really not a big deal but the cpu temp
dropped more than 30%, which is very good for my setup !!

Seems that Intel is play trick on volt and freq. Recall that the SU2300 is asking for >$100USD (1K)
which is running at ~1V/1.2G. I can get the similar setting with a street price $50 E3300
running at 1.1V/1.2G. PC Benchmark of SU2300 from passmark site is 910 and 1680 for E3300
, while the figure I just got is 1550 at default and 808 at 1.1V/1.2G.

Some questions, the claimed TDP for E3300 from Intel is 65W. But it never happened, the whole system
with default setting at full load is less than 50W.
And, the claimed TDP for SU2300 from Intel is 10W. But from various review web site, it come out that
the idle is ~25W and full at ~33-35W. Seems that its more than the Intel spec !

I'm going for this solution, a 775board + a celeron, perhaps a notebook version which can let me try at
a lower Volt !

Any good and low power consumption 775 board ?

[KTE]

Was sleepy and typed a wrong power fomula !!!

Managed to get the RMClock working !!! And here is the result,

Ambient: 18degC

Stock V+MHz (1.285V/2.5G):
idle: 25W from device (pf 0.4)
prime95: 48W
cpu: 55degC

Stock V+MHz, disable EIST in BIOS (1.285V/1.2G):
idle: 25W from device (pf 0.4)
prime95: 39W
cpu: not recorded

RMClock (1.1V/1.2G):
idle: 25W from device (pf 0.4)
prime95: 36W
cpu: 38degC

Shows either your PSU or your board is the power barrier. E3x00 are very low power to begin with, idling near the 1.4W zone including VRM inefficiency. So if you want lower idle, look at the rest of your setup.

Some questions, the claimed TDP for E3300 from Intel is 65W. But it never happened, the whole system
with default setting at full load is less than 50W.
And, the claimed TDP for SU2300 from Intel is 10W. But from various review web site, it come out that
the idle is ~25W and full at ~33-35W. Seems that its more than the Intel spec !

TDP bin can mean many things... but it's rarely max power, except for mobile CPUs.

SU CULV series are mobile, boutique. ultra-bin CPUs. E3x00 are typical, cheap, low-end desktop CPUs. Don't compare their TDPs. The SU is lower than the E3x00 in power draw, no question, but the E3x00 are still very low in power compared to typical low-end desktop offerings.

SU CULV series and the E3x00 are well established as low in power (in their segments). Those 3 SU processors do not draw any more or even equivalent to any typical desktop processor. That's why SU2300/SU4100/SU7300 ultra-portables generally beat AMD Zacate/Ontorio offerings in battery life (hence power).

You missed something in your power formula BTW. That formula is from the NetBurst and Alpha EV7 engineering days. The silicon around now, and its characteristics, are significantly different. Leakage power is paramount, and it comes down with the voltage although this differs in extent per processor. For the parameters you have access to, P = f(V³ × 0.78) is a good start for anything >45nm based.

On another note - newer boards (and chipsets) are generally more efficient (sometimes far more) causing the net effect of low system power with a newer CPU, even when the processor power consumption itself is generally similar.