i7-2600K to be used for distributed computing (eg. SETI) - To overclock or not?

[LMF5000]

Hi forum. I've been a regular reader of AnandTech for a couple of years now, but this is my first time writing in the forums.

So far all my computers have been laptops - but now that I've graduated from university and I use my computer almost always at home, I'm considering building a high-end desktop system. I'm currently at the stage where I'm choosing the individual components. CPU will probably be a 2600 or 2600K with an Akasa Venom cooler.

Basically, my questions are these:
1. Is overclocking recommended for distributed computing (see below)?
2. How much of a performance increase can I expect, (preferably expressed as a percentage value) compared to a stock 2600K, if the only overclocking I do is to click "fast" or "extreme" in the ASUS automatic overclocking software (which will come with the new motherboard) - i.e. no manual overclocking whatsoever?

Now, the reason I felt the need to ask on these forums is because of how distributed computing works. You've probably heard of SETI@home or Folding@Home so you know how it works. My particular preference (the one I will be running on the new rig) is World Community Grid (WCG), which runs on the BOINC client. In brief, the way it works is that it basically uses your computer's idle CPU cycles to perform calculations on proteins and such that could help scientists develop new drugs to treat diseases. The project sends your computer a "work unit", the computer spends a couple of hours working on it in the background, and once finished, it uploads the results to the project and downloads a new work unit. You basically get points for every successful work unit your computer completes.

What makes distributed computing so special in terms of overclocking is this:

a. Projects like WCG don't send a particular work unit just to you - they send several identical ones to many different people and thus get several results back, which should be identical. The WCG server then compares the results. If they're identical, everyone gets points for the work. If there are discrepancies, then some or all the people who worked on that unit will get no points. So if an overclock causes the CPU to output incorrect or corrupt data, then it would be useless because you get zero points for every corrupt work unit. I don't know whether overclocking actually affects the precision of CPU calculations in this way? The BOINC help documentation says it might - in fact they discourage overclocking for this reason - but I thought maybe someone from this forum can offer an opinion on this?

b. With the BOINC client, one CPU can process one seperate work unit per logical thread - at the same time. And each work unit is basically run constantly at full throttle, but as a low-priority Windows process, so it takes up all spare CPU cycles. So end result is, each core in the CPU will run at a constant 100% utilisation constantly. So in the case of an i7-2600K, you'll be running all 8 threads at 100% utilisation, for as long as the project is running in the background (i.e. up to 24 hours a day). This means that every piece of performance gained in overclocking is actually going to be 100% useful, but it also pushes the CPU and cooling system to its absolute limit and demands perfect stability, in contrast to more "normal" activities (like gaming and video encoding) - so again, I was hoping some experienced forum readers could offer their opinion about this.

For reference, my current laptop has a Core 2 Duo T7300 (2GHz). Running the project with the stock cooler built into the laptop, and no other cooling systems gives CPU temperature of 92 degrees Celsius, and I used to run it like this for 8-24 hours a day for about a year (before I got fed up of the noise of the little laptop fan constantly at full speed ^_^).

 

[Puppies04]

I would highly reccomend against clicking on any automatic overclocking software, it will overvolt your pc by way to much increasing thermal output vastly. Overclocking a SB CPU is so easy to do manually you might aswell do it yourself. As for the % increase you will see by overclocking that will depend on the software and having no experience with WCG I can't be precise but with a sustainable 4.2GHZ OC i would be surprised if you don't see a 30% increase in output.

 

[mrjoltcola]

Agreed, with nearly any 2600K it is as easy as tweaking the multiplier to 4.3, bumping the voltage to 1.275, running a couple of stability tests, and you'll likely have a decent overclock with no other changes.

Auto-overclock tweaks the base clock, which I think is useless for the most part with Sandy Bridge. Nearly all chips overclock to 44x multiplier with no sweat, and a cheap cooler (anything BUT the Intel cooler).

 

[john3850]

Markfw900 on this form has a has a folding farm up 24-7 x years.
His 2600k is the only pc that gives him trouble with F@H bigadv units what ever that is.

 

[LMF5000]

So if say, I follow your recommendations and take it to 4.3GHz, will that be the frequency of all 4 cores... or will it only go up to 4.3 in "turbo" mode (i.e. when only a single core/thread is being used)?

Are there any bad points about overclocking the 2600K? Like shorter life, chip damage over time, corrupted work units etc.?

 

[gorydetails]

some of those project can tell you if you are sending corrupted results, because they used redundant processing...keep an eye on that if you plan to oc...

Edit oops I seeyou already mention it...anyway I would stick to redundant processing and check for invalids...at least you will not impact result because of a bad oc

 

[PlasmaBomb]

A modest bump should be OK provided you properly test the stability of the overclock.

By the way, in case you didn't realise we have a DC sub-forum -
http://forums.anandtech.com/forumdisplay.php?f=15

Though Markfw900 will probably see your post at some point.

 

[BD231]

Markfw900 on this form has a has a folding farm up 24-7 x years.
His 2600k is the only pc that gives him trouble with F@H bigadv units what ever that is.

Unstable overclocks, which makes me wonder what a 2500k/2600k is REALLY good for as far as stable overclocks go. I see some 4.4/5's/6's around but who's testing for error's?

edit: pegging your CPU at 100% load/voltage 24/7 is a heavy task and you should consider the lifespan of the chip over the added mhz in situations like that. If it were me I'd OC as far as default voltage or .500 over stock voltage and that's about it. Its not that hard at all to degrade a chip these days and running max load like that on high OC's is a good way to do it.

 

[LMF5000]

By the way, in case you didn't realise we have a DC sub-forum -
http://forums.anandtech.com/forumdisplay.php?f=15

Though Markfw900 will probably see your post at some point.

Whoops. Didn't realise there was a forum for that . In hindsight it might have been more related to my question...

edit: pegging your CPU at 100% load/voltage 24/7 is a heavy task and you should consider the lifespan of the chip over the added mhz in situations like that. If it were me I'd OC as far as default voltage or .500 over stock voltage and that's about it. Its not that hard at all to degrade a chip these days and running max load like that on high OC's is a good way to do it.

That's what I was thinking. To me, overclocking is like modifying your car with aftermarket performance parts. Sure you make them faster, but you increase the risk of blowing something when you drive them hard. So I leave my laptop and my car at stock speeds but then I'm never afriad to run either to their limits because I know the manufacturer set those limits with a decent margin of safety.

So, is there any compelling reason to go for the 2600K instead of the 2600? I mean, besides the overclocking potential of the K? Can I really make it 30% faster with the same reliability, stability, and longevity of the CPU at stock settings?

 

[Ken g6]

So, is there any compelling reason to go for the 2600K instead of the 2600? I mean, besides the overclocking potential of the K?

Nope. If you choose not to overclock you might be able to save even more by inserting a barely-slower Xeon 1235 or 1230 in your board. Anecdotally, they work in most boards. (The 1230 works only if you have a discrete GPU.)

Can I really make it 30% faster with the same reliability, stability, and longevity of the CPU at stock settings?

And another nope. But you can get close. First, as you can see from those charts, getting a good cooler helps.

Second, my preference is to never exceed 1.35V and/or 65C on any core, with LLC completely off, when running the Prime95 Small FFT torture test. I'm conservative - some would say very conservative - but that'll help your CPU last longer. So take it up to 1.35V at stock speed, start the Prime95 Small FFT torture test, and see what temperature it reaches. If it's over, say, 60C, after a few minutes (because overclocking will raise the temperature) drop the volts until it's not. Then raise the multiplier until Prime95 says it's unstable at that voltage (or it gets too hot); and finally back down a notch or two on the multiplier.

 

[mrjoltcola]

Can I really make it 30% faster with the same reliability, stability, and longevity of the CPU at stock settings?

30%? No. 10%, yes, 15%, probably (in my opinion) if you use a good cooler on it.

What's worse: running 4.4Ghz @ 1.28v @ 50C around the clock or 3.8Ghz @ 1.20v @ 72C-80C. Heat degrades like voltage, and I've seen those sort of temps out of a few 2600Ks with the stock cooler. I'm no hardware engineer, though, just have my opinions.

Idontcare (IDC) made a good post somewhere, basically making the point that for his purposes, a $300 2600K is easy enough to replace that theoretical lifespan reduction of a few years wouldn't make a hill of beans difference to him, and the risk was worth the reward of the extra speed for the effective life, being until the next couple of generations (Ivy Bridge, Haswell, etc.). I'm paraphrasing, and hope I didn't misquote him, but I agree with him. But to each his own.

Personally, I've never had a processor die, and I still own P3s, P4s and Athlon 64s that were cranked up many years ago. That said, I think guys running their 2600Ks at 1.5v around the clock are asking for it. I have a couple of 5.1Ghz 2600Ks, and put them in my sig because they are stable with no crashing, but I don't run them that way around the clock, I just crank them up for fun, games, or benching. However, of late, I've decided to actually build myself a separate rig for playing stupid, and get back to stock on my workstation.

I certainly run my datacenter servers stock. 10% or even 30% extra speed is lost in the noise when we talk about massively distributed computing, but you can look at it either way. You can just as easily make the caes that 10% overclock is statistical noise, considering the variability of silicon...

 

[LMF5000]

Nope. If you choose not to overclock you might be able to save even more by inserting a barely-slower Xeon 1235 or 1230 in your board. Anecdotally, they work in most boards. (The 1230 works only if you have a discrete GPU.)

So it's settled. No overclocking. The risks just aren't worth it for a 10% increase in performance (which equates to doing an extra 6 minutes of work every hour).

Last question - keeping everything completely stock, will the 2600K be any faster than the 2600? It has a higher benchmark score on cpubenchmark.net , and the price is only €20 more than the non-K version. Xeons are out of the question because the local computer store (of which there are two in all of Malta) doesn't stock them.

P.S. For a CPU with an unlocked multiplier, how do you return the settings to "stock" after overclocking? Is there a "reset settings" button somewhere, or do you write everything down on a piece of paper before you start tweaking and return everything to the same numbers? (can you tell I haven't used a desktop w/overclocking ability for the last 5 years :whiste:?)

 

[Phil L]

If you aren't overclocking, the only real difference between 2600k and 2600 is the iGPU. Therefore it depends on whether you want to use that or a dGPU for display (and maybe some GPGPU power if your software utilize that). If you plan on gaming, then dGPU is preferred, but even then having an iGPU may not be a bad thing either (troubleshooting, QuickSync on supported boards, etc.)

For resetting to stock, there is usually an Optimal or Factory Default option in your motherboard's BIOS/UEFI which you can use. Some motherboards also allow you to save a number of profiles which can be loaded at latter time. In the latter case, it is prudent to make a default profile before start playing with the settings so you can easily revert back if needed.

 

[bryanW1995]

Sorry to be contrarian, but I've been running seti@home and various other DC apps for years now. I currently have an i7 920, x6 1055t, Q6600, and 3-4 c2d dual cores at the office running seti/correlizer. My top cruncher is getting rebuilt, but it is going to be either 2500k or 2600k.

I see an almost linear increase in WU's vs clockspeed, ie, I get nearly a 50% bump from my 50% overclock to 4.0. on my i7 920. It makes sense that you would, too, because the clock speed has much more to do with the actual "work" that the cpu can do than, say, the ram. This isn't like gpus where you can get a 25% overclock but your fps only goes up by 3%.

And markfw900 overclocks the **** out of his cpus as well.

I don't know about WCG, but the gpu has become much more important recently in most DC projects.

I suggest that you post your question in the DC forums, i know that we have some pretty serious WCG crunchers there that will be able to answer specific questions about that particular project.

edit: btw, I've never heard of that cpu cooler that you mentioned. It might be very good, but if you're not overclocking then the stock cooler will be more than enough, and if you ARE overclocking then you'll probably want something a bit beefier.

 

[LMF5000]

For resetting to stock, there is usually an Optimal or Factory Default option in your motherboard's BIOS/UEFI which you can use. Some motherboards also allow you to save a number of profiles which can be loaded at latter time. In the latter case, it is prudent to make a default profile before start playing with the settings so you can easily revert back if needed.

Thanks for the advice.

I see an almost linear increase in WU's vs clockspeed, ie, I get nearly a 50% bump from my 50% overclock to 4.0. on my i7 920.

Makes sense. I mean, operations done per second = clock cycles per second * number of operations per clock cycle... so I would assume doubling the clock speed would give double the performance, all other things being equal. But when a 2600k already runs at 3.4GHz (normal) to 3.8GHz (Turbo) stock, I only see a limited frequency increase possible before air cooling doesn't cut it anymore. BTW, does overclocking a k-series void the warranty?

I don't know about WCG, but the gpu has become much more important recently in most DC projects.

WCG doesn't use the GPU. I would probably be running something else on the GPU. I've always wanted to run gpugrid, but my laptop's 8600M GS is not powerful enough to finish work units before the deadline. Should be no problem for a GTX 570 though

edit: btw, I've never heard of that cpu cooler that you mentioned. It might be very good, but if you're not overclocking then the stock cooler will be more than enough, and if you ARE overclocking then you'll probably want something a bit beefier.

Here's the manufacturer page. I've always had a soft spot for Akasa. Decent cooling for a reasonable price. And so far my collection of 4 Akasa fans have collectively run for 2500 hours without self-destructing ^_^. No point paying 2x the price for a cooler that's, say, 25% better*. Same reason why I'm looking at a GTX 570 instead of a GTX 580 (performance increase of 10.1% for a price increase of around 60%) - and why my current laptop is an Acer instead of a Dell or HP :biggrin:.

*Keeping in mind that the only hardware available to me is the handful of products available at the local Maltese computer superstore (not sure if I'd be breaking forum rules to post the name here).

 

[dguy6789]

You can overclock and have very tangible performance gains in DC, perfect stability, and no significant reduction in processor life span. It's a matter of doing it right. There's no shortage of new(to PC overclocking and enthusiast stuff in general) users on these boards who don't have a total grasp on all of the more minute aspects to OCing that come from doing it for a long time. There's an awful lot of people whose total experience amounts to reading a guide or two, changing some bios settings, and running a benchmark. To be blunt, anyone who is telling you it's a just plain bad idea to OC for distributed computing is flat out wrong.

There's plenty of stability tests that can be run that will tell you if the processor is indefinitely stable at full load or not. Running LinX for 24 hours then Prime95 for 24 hours is a good way to see if the machine is rock solid reliable or not. Another factor to take into account is making sure you have a power supply that can comfortably supply that kind of juice. You do not want to run DC on a computer whose power supply is running close to the edge of its specifications to power the PC. That will shorten the life of the power supply greatly.

You must keep the processor voltage and temperature under control. Voltage and temperature are the primary set of attributes that will shorten the life of a processor. A very safe overclock would be to take the CPU as high as you can go(while maintaining absolute stability as indicated by the stress tests) without raising the voltage at all. As long as temperatures were controlled well, the life span of a processor configured like this would be almost identical to the stock configuration.

I definitely agree with seeking advice from Markfw900

 

[Ovven]

You can think about cpus in this way: they are designed to go for 10-15 years at stock setting, but what's the point in keeping them that long? Just think what cpus were 10 years ago compared to now. Or five years. With modest overclock of 20-30% you may (and i stress may) shorten the cpu's lifespan to 5-6 years, but by then it would be obsolete anyway.

 

[LMF5000]

Running LinX for 24 hours then Prime95 for 24 hours is a good way to see if the machine is rock solid reliable or not.

Which of the three modes in Prime95 is best for the task? Small-FFTs, Large-FFTs or blend?

 

[petrusbroder]

I have been running WCG for quite a few CPU-years (27 years 271 days 08 hours 40 minutes). I have crunched, folded, primed or what ever since December 2001.

I have used Pentiums, C2D, C2Q, i7, AMD X2, AMD Phenom and AMD Phenom II. I always OC my CPU as much as I can. Some times it is 15%, often it is much much more (e.g. my i7 920: stock 2.66 GHz, just now 4.0 GHz = 50% OC). I seldom have temps above 80ºC, mostly they are 10 - 20% higher than stock. I use name brand high efficiency coolers: Tuniq or Noctua, some times with an extra washer to increase the preassure between CPU and cooler + high quality thermal interface compound).

I have not had any CPU fail on me unless the PSU failed and took the main board and CPU with it. But the results are quite convincing: there is a clear linear relationship between the points you earn and the how much you OC.

The main point is regular maintenance: blow away all dust in the coolers, get all the dust out from the main board, blow away dust in the PSU or better: keep your computers as dust free as you can. Also a regular maintenance of the software (clean out all trash files, clean up the registry, defrag the HDDs (but not the SSDs), use as much RAM as you can afford, all og this helps.

My hardware failures are 75% PSUs (yeah, they work 24/7 and have some mechanical parts and they are much more exposed to heat and ageing), 20% HDD and the rest is fans and main boards. Never a CPU.
Don't be afraid to try stuff out.

1. Decide on what CPU-tems you feel comfortable with and keep your temps there. Just as a test I have run an i7-860 @95ºC for 1 year and 8 months - and no problems, but I do not recommend that. That is a test. Side by side with that computer runs an identical one with CPU-temps of 70ºC and they are neck-to-neck in the scores when running the same project.
2. Maintain your hardware and software: keep it dust free, renew the termal interface compound (I use Arctic Silver 5) about every 18 - 24 months, keep enough pressure between cooler and CPU, remove all the "trash files", check the registry regularly.
3. Keep the computers well aired.
4. Check speeds and performance regularly: I do it approx each month and that takes a few hours for my computer farm (25 comps), but it should not be a problem for one or two computers.

Welcome to the Distributed Computing forum to seek advice and ask questions.

Best regards and enjoy crunching.

Petrusbroder
(here are my BOINC-stats)

 

[PlasmaBomb]

Whoops. Didn't realise there was a forum for that . In hindsight it might have been more related to my question...

No worries, it's not meant as a criticism as there isn't anything wrong with your posts or their location, it's just that DC maybe better able to help you.

That's what I was thinking. To me, overclocking is like modifying your car with aftermarket performance parts. Sure you make them faster, but you increase the risk of blowing something when you drive them hard. So I leave my laptop and my car at stock speeds but then I'm never afriad to run either to their limits because I know the manufacturer set those limits with a decent margin of safety.

So, is there any compelling reason to go for the 2600K instead of the 2600? I mean, besides the overclocking potential of the K? Can I really make it 30% faster with the same reliability, stability, and longevity of the CPU at stock settings?

CPUs have pretty long lifespans you should be able to extract some extra performance without it dying before the upgrade bug hits, even if you run it 24/7. Unfortunately I haven't had the chance to play with any SB rigs yet

Certainly plenty of core2 chips have withstood lots of abuse (mine was OCed and ran FAH for years).

I wouldn't play around with the base clocks as it can corrupt data, so if you want to OC the K is your best bet.

 

[Idontcare]

Which of the three modes in Prime95 is best for the task? Small-FFTs, Large-FFTs or blend?

Depends on what you are intending to stress test. See the sticky Overclocking CPU/GPU/Memory Stability Testing Guidelines

 

[LMF5000]

I have been running WCG for quite a few CPU-years (27 years 271 days 08 hours 40 minutes). I have crunched, folded, primed or what ever since December 2001.

---

Just as a test I have run an i7-860 @95ºC for 1 year and 8 months - and no problems

I've been using WCG since the days when they were still called United Devices. Originally found out about them in a PC magazine and started using it because of the screensaver feature because I was tired of the default ones in windows 98 . My laptop runs at about 92ºC so good to see someone else confirm that CPUs can take the abuse. My primary worry about O/C was returning corrupted data and not getting points for it, but if I actually buy and build the new rig I would be able to check the logs to see if my results are validating properly and getting me points - and if not I can always tweak the O/C settings. Thanks for the maintenance suggestions too.

Depends on what you are intending to stress test. See the sticky Overclocking CPU/GPU/Memory Stability Testing Guidelines

Thanks for that. I'll keep the thread for future reference ^_^.

 

[GLeeM]

Welcome to the forum, LMF5000

You have to get the CPU that overclocks best!

Listen to petrusbroder, he knows what he is talking about, 25 computers overclocked running DC 24/7 for years on end.

I've run DC (including some WCG) since 2004 on OC'ed computers. I've had fans and one PSU fail. On one computer, after four years I had to lower my OC a little to stay stable (probably because of the PSU).

A DC computer shouldn't be run OCed as high as it will go because they run at max all day.
My i7 920 will OC to 4.2 but I have been running DC on it for 2.5 years at 4.0.
I watch Real Temp to know when to clean the dust out.

You will be amazed at the increase in your ppd with an OCed 2600K desktop.

If you don't already crunch for a team, you would be welcome at TeAm Anandtech Check out our sub-forum if you want. We have a lot of fun We even have a secret code for smilies

 

[LMF5000]

Welcome to the forum, LMF5000

If you don't already crunch for a team, you would be welcome at TeAm Anandtech Check out our sub-forum if you want. We have a lot of fun We even have a secret code for smilies

I don't currently crunch for a team (or crunch at all at the moment, due to the inconvenience of doing it on a laptop), but that's a good suggestion! I'll keep it in mind for when (or actually, if) I can bring myself to pay the €1700 to get all the components for a complete desktop system! Which projects do you have in mind?

 

[Diogenes2]

Markfw900 on this form has a has a folding farm up 24-7 x years.
His 2600k is the only pc that gives him trouble with F@H bigadv units what ever that is.

My 2600K has no problems with bigadv, so the 2600k is not likely to be the problem ...