A little undervolting adventure

[scaryjim]

... I'm surprised about the RAM instability though. The mobo is rated up to 2000oc (mine is 1600) and the RAM itself was designated 8-8-8-24. ...

Firstly, if it's an AMD rig the memory controller is only officially rated to 1333MHz, regardless of what the motherboard claims. It may be able to cope with running the RAM faster than that, but officially 1600MHz is out of spec! Secondly, make sure the RAM is running at its rated voltage: it may well only be rated to 8-8-8- at 1.65V or higher, and the motherboard may be running it at a lower voltage.

[Attila the Bun]

You can buy extension cables, I'm sure. If your motherboard regulates the speed of 3 pin fans, then you'll be able to cut down on the noise a bit (you can buy a resistor to drop the voltage from 12V to 7V, for example)

Extension cable already bought

[Attila the Bun]

Firstly, if it's an AMD rig the memory controller is only officially rated to 1333MHz, regardless of what the motherboard claims. It may be able to cope with running the RAM faster than that, but officially 1600MHz is out of spec! Secondly, make sure the RAM is running at its rated voltage: it may well only be rated to 8-8-8- at 1.65V or higher, and the motherboard may be running it at a lower voltage.

Cheers Jim,

It's running at 1.545v currently

Would the rated voltage be on the packaging? Although I could probably get it from the Corsair site thinking about it. *toddles off to check Corsair*

[scaryjim]

assuming you have these then they're rated to 1.65v so you'll need to work out how to give them +0.1v in the BIOS.

[Attila the Bun]

assuming you have these then they're rated to 1.65v so you'll need to work out how to give them +0.1v in the BIOS.

I have the CMD4GX3M2B1600C8 not the CMP4GX3M2B1600C8...what's in a letter? It's an older version...I am sure that I ordered the "P" version but as I didn't save the basket I will never know.

Anyway this motherboard came with a utility that allows you to up (or down) the voltages of the CPU and DRAM via the desktop...in theory I could simply up the voltage to 1.65v...in practice I'll be doing a lot more reading of the manual before I start casually moving dials.

[Golden Dragoon]

Decided to have a go at this undervolting lark as I recently got an Athlon X3 (unlocks nicely to an x4 as well )
So I followed my usual overclocking method, but lowering volts instead of increasing things, and got it down to 1.15v (1.175 for 4 cores) fully stable on prime95 and OCCT (running together for 4 hours and separately for 2hours each) when you consider that this chip has a default vcore of 1.4 that is quite a drop.

However I encountered a slight problem, anything under 1.28v (3 or 4 cores) will cause folding@home to crash with a client core communication error or F@H core a3 stops responding not sure why as prime and occt are stable much lower voltages, but still a 0.12 drop in vcore is ok, not perfect compared with a 0.225 drop but ok

[scaryjim]

The folding clients seem to hit the CPU much harder than other stability testing tools. A few people on here have mentioned that they can get a 24h Prime stable overclock which borks in minutes under F@H!

Stanford (well, Gromacs, to be precise ) have released a F@H stability tester which simulates the folding load without the work units! Available at their utilities page.

[Terbinator]

That should save people losing WU's after a quick tweak, including myself. Cheers 'Jim.

[Queelis]

Yup, somewhere in the thread I've written that (I think). I've stress-tested my X3 Phenom with OCCT, all went OK for 8 hours, assumed it stable. Few months later I discover folding, and it wraps in several hours

The voltage looks good for an unlocked CPU. My unlocked X3 Phenom would only run folding with 1.275V, current X4 965 makes do with 1.25V. What a difference a revision makes - higher clock speed and lower voltage Still generates more heat than the unlocked Phenom with higher voltage, though.

[kalniel]

Stanford (well, Gromacs, to be precise ) have released a F@H stability tester which simulates the folding load without the work units! Available at their utilities page.

GROMACS is a MD engine, not an organisation Useful link though.

[scaryjim]

GROMACS is a MD engine, not an organisation Useful link though.

Yeah yeah, I kinda know that, but given that:

StressCPU v2.0 is available to download directly from the Gromacs.org web page

I assume there's some kind of organisation that maintains and develops it as a partially (wholly?) independent project...?

Shortcutting it to "GROMACS", rather than "the team responsible for developing and maintaining GROMACS", seems reasonable to me...

edit: shame they don't do a GPU stressing util to go alongside the GPU memory testing utilities though...

[kalniel]

Yeah yeah, I kinda know that, but given that:

I assume there's some kind of organisation that maintains and develops it as a partially (wholly?) independent project...?

Uppsala University, and Royal Institute of Technology, Sweden hold the copyright, and development takes places at Department of Cellular and Molecular Biology, Uppsala University, Sweden and Stockholm Bioinformatics Center, Stockholm University, Sweden

[scaryjim]

Uppsala University, and Royal Institute of Technology, Sweden hold the copyright, and development takes places at Department of Cellular and Molecular Biology, Uppsala University, Sweden and Stockholm Bioinformatics Center, Stockholm University, Sweden

I think I'll stick to calling them "GROMACS"