overclocking
 

Hi,

is overclocking generally the preferred way of running machines doing Mersenne number testing?

I thought I could try to learn about it myself but think I'd be better off if people who know how to do this could
explain it to me.

I've got 4.6GHz FX9350 AMD running 4 tests on four of it's eight cores.

Would be it be a good idea to overclock?

If so, can I have some help to get it overclocked?

Overclocking just increases the likelyhood of your Computer to do wrong calculations, whilst speeding up every test you do. So you will get more credit for that work, but since a single error in a whole run will make your work done on that test all useless, you would be better off not overclocking.

A little Overclock may not have that impact on you, but pushing it to its limit will defenitely cause you to loose efficiency.

With regard to overclocking that CPU, I would advise against it, unless you are prepared to deal with substantial increases in heat. In addition, if you undertake it, rigorous testing (built into Prime 95) should be done to verify stability.

I am running an FX-8350, and have experimented with driving it above its stock 4 GHz. Even with liquid cooling, heat quickly becomes a problem. Also consider that power demands not only on the PSU, but also the motherboard itself will increase

An extreme example, which can be found on the Asus ROG site, describes how to push an 8350 to 4.8 GHz. This guide is specifically about the Asus board I have, but the principles are similar, regardless. I tried the instructions, which are kind of scary, and quickly backed off when I saw high CPU temps combined with failed Prime 95 Torture Tests.

I have somewhat different experiences with overclocking GPUs, but all the same details have to be attended to: power, heat, and the noise resulting from cooling hot chips.

EDIT: It is also fairly likely that a later, faster chip in the same line will have less headroom, having already been selected for pushing to a certain limit.

Start out at stock settings and see how things go there. The same suggestions apply regarding running the P95 Torture Tests to make sure your system is performing correctly. If you do pursue overclocking, you should have baselines for power consumption, heat production, and stability.

Thanks very much for your advice guys.

It's nice to understand it a bit better.

I've already got cooling issues. This machine has 8 cores but I can't get them all working without the machine shutting off.

I'll keep my machines as they are.

[QUOTE=wildrabbitt;397422].....
I've already got cooling issues. This machine has 8 cores but I can't get them all working without the machine shutting off.
......[/QUOTE]

Are you monitoring CPU temperatures? Shutting down under load could indicate overheating. What kind of cooler is on your CPU?

Noctua NH-D14 cooler

Overclocking brings no advantage for GIMPS project. Trust me on this. You may get more credit, say, if you increase your speed 10%, you get 10% more credit. But it is enough to have ONE bad residue in 10 tests and you didn't help the project, contrarily, you confuse the things, as it may take years until someone finds your residue was bad. Ye also may [U]miss a prime[/U], which would be, indeed, terrible. Imagine someone finds M52, and when you look to history, you see that you had this assigned five years ago, and reported a wrong residue. This is my worst nightmare :razz:

Talking probabilities, you do 60 millions iterations for a test, that is 600 million in 10 tests. If one iteration in those 600 millions is wrong, because different (see below) issues, then your "effort" to overclock (see below) was useless.

Why things can go wrong? First, the voltages, then the temperature. Overheating is your biggest enemy. To keep the system stable at higher clocks, you will need to dissipate more power, and even more when you need to increase the voltages. Some 10% overclock goes with 20-30% more power, which goes into more heat. You either burn your CPU (and your residues with it), or you invest even more power in cooling stuff to keep it cool.

So, at the end, you can get a 10% more credit, and keep a low risk for errors, but this comes with 50% more expenses in electricity and components (shortens their lifetime, faster deprecation, blah blah).

This is a serious bisiness! I won't be doing any overclocking.

I never have as it happens for the record.

Thanks

[QUOTE=wildrabbitt;397422]I've already got cooling issues. This machine has 8 cores but I can't get them all working without the machine shutting off.[/QUOTE][QUOTE=wildrabbitt;397425]Noctua NH-D14 cooler[/QUOTE]Those coolers are supposed to be very efficient. Are you sure you installed it correctly (some but not too much thermal paste,case well ventilated helping the CPU cooler get rid of the heath...) ?

Jacob

I'm on the other side of the fence on this one. I am a big fan of overclocking but only doing so in a careful way. I have overclocked my CPU by over 30% and have no recorded errors yet. Almost all of my tests have been double-checks and have all eventually been confirmed.

Basically there are three things which we all want. High speed, Low Temperature, High Stability. Unfortunately, you only get to choose two of those things. High speed - High stability is where I want to be, so that comes at the expense of a hotter CPU. If you ever do get into overclocking a machine working for GIMPS, switch to doing purely double-checks until you get a bunch of matching residues to make absolutely sure you're stable.


By the way, did you install the heatsink yourself? The NH-D14 is a really good cooler that should be more than capable of handling your CPU under load at stock settings.

[QUOTE=TheMawn;397463]I'm on the other side of the fence on this one. I am a big fan of overclocking but only doing so in a careful way. I have overclocked my CPU by over 30% and have no recorded errors yet. Almost all of my tests have been double-checks and have all eventually been confirmed.

By the way, did you install the heatsink yourself? The NH-D14 is a really good cooler that should be more than capable of handling your CPU under load at stock settings.[/QUOTE]

I'm with TheMawn on this one. Mild overclocks of Intel CPUs are often fine for years, and error rates can remain fine if temps are kept reasonable. Older chips had massive headroom- I still run a Core2-6300 1866 @2400 8 years after purchase, 24/7 LLR (generalized Prime95 for other numbers) the full time. If you find a machine/overclock unstable, back off by 5% speed and a few % voltage, and run some double-checks after the torture test. There are safe overclocks, "I can play my favorite game faster!" overclocks which produce errors, and "I saw the windows screen at 50000Mhz!" overclocks. The first category is useful for number crunching, and if you fix your heatsink installation temps will be lower than they are now.

2 OC's here....
 

I have an i5-750 that I am using the standard GigaByte Fast6 Software to OC slightly (2.8 - 3.2).... just over 3 years old
An an i5-3570K that I have OC'd from 3.4 to 4.2...I let the installer do it for me who build the PC....just over 2 years old.

In both cases I have after market Coolers (Noctua for both I think).

Almost NO LL errors (knock on wood) a couple 3 years ago for the 750 but I seem to recall that was during a power outage.
None yet on the 3570K.

However, both have more than their share of "unexpected" shutdowns.

Up to 1 per month on the 3570K; just a couple on the 750.

about the Noctua NH-D14
 

[QUOTE]Those coolers are supposed to be very efficient. Are you sure you installed it correctly (some but not too much thermal paste,case well ventilated helping the CPU cooler get rid of the heath...) ?
[/QUOTE]

I've just upped it to all 8 cores. The problem might have been that the machine was in a cupboard (hence it was overheating), or else like you say I might not have installed it correctly.

2 cores got work.
The other 2 of the cores say they are waiting for work. I've never seen that before. Anyone familiar with this?

[QUOTE=wildrabbitt;397490]
The other 2 of the cores say they are waiting for work. I've never seen that before. Anyone familiar with this?[/QUOTE]

Maybe try to "Contract Server Now"

I am with VBCurtis and TheMawn on this OC subject.
Mild and well controlled OCs may be worthwhile. I have been running an i5-750 with a 20% OC for nearly 5 years, 24/7, and it has been perfectly stable and accurate. I use a Noctua NH-U12P cooler, and the temperatures are in the high 40s - low 50s (C !...) depending on room temperature and on how clean the cooler is. Mind you, it´s important to dismount the fans and clean it thoroughly every year or so to keep it in top notch performance.
Also, I second the advice of running a fair amount of DCs in order to get confident in the stability of the system upon OCing it.

[QUOTE=lycorn;397510]Also, I second the advice of running a fair amount of DCs in order to get confident in the stability of the system upon OCing it.[/QUOTE]

That's the root issue.

It's fine to OC if know what you're doing.

Many choose to under-clock to ensure sanity.

[QUOTE=petrw1;397483]However, both have more than their share of "unexpected" shutdowns.[/QUOTE]

While I haven't had any of that on my 3570K my 5930K does shut itself down for no reason on occasion. There are a lot of real life applications where a catastrophic failure is more desirable than its slower counterpart. A pressure vessel that leaks slowly will slowly poison everyone and be a PITA to find (when detected) whereas one that bursts open will set off enough bells to shut the systems down and, barring any immediate casualties, be much better in the long run.

For GIMPS, the catastrophic failure is very noticeable and it at least alerts you that something isn't quite at 100%. Also, because of the 30 minute (default) delay between checkpoints, a sudden "GAAH 1 + 1 = 3 ERROR ERROR BEEP BOOP BEEP BOOP SHUT IT ALL DOWN!!!!" error actually prevents an erroneous residue from being saved, whereas the "slow leak" will produce errors without any other symptoms manifesting.


That being said, it probably behooves us both to actually look into what is causing these problems. Every once in a while I'll see that one of the cores of my 5930K has spiked to 100C although I can't see how often or for how long (my guess is a single sampling time) so I think it's a thermal shutdown as opposed to a BSOD.