Trying to find minimum voltage for CPU

[Lonyo]

I've got an overclocked E5200 and overshot a lot on the voltage (went into the 1.3's) when overclocking.
Now I'm backing down, and I've gone all the way to 1.2v BIOS (1.185v CPU-Z) and usually I run Prime95 (small FFT because that claims to focus on the CPU only) for ~1hr to check, and then knock it down another notch (0.00625v) and start again.
RAM is basically in spec (2% overclock) so I am not worried at all about that, I am just trying to find the minimum voltage my CPU needs, without running P95 for 8 hours each test (yet).

 

[Idontcare]

When I went looking for Vmin I went the other way...kept dropping Vcc until I fail small FFT in just a 10min run. This got me to the bottom quickly. Then I invested more time in finding the stability point as I slowly bumped up the Vcc while doing longer and longer small FFT runs (basically run until it errors, if no errors in 24 hr then I didn't bump up the Vcc any more).

 

[Lonyo]

Originally posted by: Idontcare
When I went looking for Vmin I went the other way...kept dropping Vcc until I fail small FFT in just a 10min run. This got me to the bottom quickly. Then I invested more time in finding the stability point as I slowly bumped up the Vcc while doing longer and longer small FFT runs (basically run until it errors, if no errors in 24 hr then I didn't bump up the Vcc any more).

My mistake was thinking it would need more than it does.
My OC is 3.4GHz, and Realtemp says VID is 1.1875v (I assume that's stock voltage? I never checked), and it's currently running at about 1.18~1.185v according to CPU-Z at the overclocked speed.
I expected it to hit the minimum at least 0.02v ago, but I still haven't hit the bottom.
At least my methodology seems sound enough though.

 

[katank]

Try using IntelBurnTest (wrapper around Linpack). It generates failures way faster than prime95. 10min of that is worth more than 1 hour of prime.

 

[nevbie]

well my method was:

1. Lower vcore in BIOS one or more steps
2. Boot to windows xp
3. prime95 for a short while (~2min), then back to 1.

eventually when i reached instability points it was almost always the windows booting that gave black screen and reboot, only once did it successfully boot xp without being 2-minute prime95 stable. still i could always get into BIOS to change the setting, even in an instable system. i had to overclock to get any instability though, the lowest available voltage was high enough for default clocks. no wonder with a 1.6GHz conroe single-core though.

 

[BonzaiDuck]

Originally posted by: katank
Try using IntelBurnTest (wrapper around Linpack). It generates failures way faster than prime95. 10min of that is worth more than 1 hour of prime.

Ditto from me.

Anand published an article in late 2007 on how Penryn changed the over-clocking status quo, referring specifically to the QX9650. Implicit in the article is a strategy for over-clocking that assures us that stress-test failure is not due to memory or chipset.

Intel publishes specs for CPUs at their web-site, and they print a "maximum voltage" spec on the retail box. I use this information as a starting point.

With my E8600, I pushed the processor 100Mhz beyond a 20% over-clock @ 1.30V, and once certified stable PRIME95-16 hours and IntelBurnTest 2 hours, I dropped the clock back by 100 Mhz -- which translates to a 10Mhz final adjustment for caution. The retail "maximum was 1.25V; the monitored idle value is about 1.27 or 1.28, and the load value pegs at 1.25V.

In plumbing for a target clock's voltage setting, I do a crude adjustment and a fine adjustment. The crude adjustment depends on preliminary failures with PRIME95 which occur within 30 to 60 minutes. I further prove these results with a 10 to 30 minute IntelBurnTest run. Once I've found the "safe and reliable" voltage to test, I run PRIME95 for a marathon number of hours and then IntelBurnTest for the 2-hour run.

This may be "over-kill" for the time it takes, but it gets pretty simply once you've gone through the preliminary short tests.

 

[BTRY B 529th FA BN]

What does/is the VID number? I just read it's the voltage needed to start the cpu? that sounds like a car analogy.. hah

 

[Idontcare]

Originally posted by: BTRY B 529th FA BN
What does/is the VID number? I just read it's the voltage needed to start the cpu? that sounds like a car analogy.. hah

VID is the Voltage Identification. There is a VID to correspond with each power state (sleep state) of the CPU. Penryn's for instance have 6 VID's.

The VID most folks are referring to when we just off-hand refer to "VID" is the VID for the corresponding fully-loaded power state. Coretemp reads and reports this specific VID.

VID is what the BIOS reads from the CPU when it determines what the Vcore (Vcc) is supposed to be set to. It is factory set by Intel and VID's have a large range even for CPU's binned to operate at the the same clockspeed.

VID is NOT what CPU-Z reports. CPU-Z reports Vcc (Vcore) as it appears to be at any given point in time (so it differs from VID for Voffset, Vdrop and Vdroop reasons).

See more here: http://www.anandtech.com/cpuch...howdoc.aspx?i=3184&p=5

http://processorfinder.intel.c...aspx?label=CoreVoltage

http://processorfinder.intel.c...tails.aspx?sSpec=SLB8W

And see pages 15, 16, 17, and 20 and the note on top of page 21, and the comment at the top of page 73 in this: http://download.intel.com/desi...or/datashts/318726.pdf

 

[BTRY B 529th FA BN]

My Q9650 has a VID of 1.2500v from what Core Temp says...

 

[Lonyo]

Um, I've kept going and did Intel Burn Test 10 runs at 1.14v then OCCT for 1 hour and both passed at 3.4GHz.
Then I dropped to 1.13v (actual, 1.15v in BIOS) still at 3.4GHz and 10 runs of Intel Burn test without error.

Did I get a good chip in terms of low voltage operation or is this normal-ish?