Dear PC Industry: Please overclock responsibly

Dear PC Industry:

During the past two weeks we've tested three desktops with ambitiously overclocked Intel Core i7 920 chips. Two of those have failed Prime95, a publicly available benchmark designed to test CPU stability. One desktop last week blue-screened within two minutes of a Prime95 run. This afternoon, a PC that came overclocked to 3.73GHz throttled down to 2.4GHz (below the 2.66GHz stock speed for the 920 chip) after about 10 minutes.

Please overclock with care.

(Credit: Intel)

We've seen the Core i7 920 chip overclocked successfully. A chip bumped up to 3.88GHz in a system from AVADirect passed an overnight Prime95 run last week with no throttling, and at reasonable temperatures. That it's possible to squeeze $700 worth of performance from a $250 CPU is exciting for you and for your customers. However, as should be obvious, overclocking a CPU shouldn't come at the expense of system stability.

Call us foolish optimists, but we won't name names of the systems that failed because both of you had plausible explanations. The blue-screened desktop had been shipped multiple times to multiple review outlets. Neither shipping nor reviewing is a particularly gentle process.

The PC that throttled down came with what you told us was an outdated BIOS. You're not selling this PC yet, so we're willing to give you the benefit of the doubt. We look forward to trying out the new BIOS you said you'd e-mail us.

So, PC Industry, consider yourself on notice. We weren't too concerned about your previous overclocking attempts that went 10 or 20 percent above stock. Now that you're aspiring to 50 percent performance gains, we're going to require two things:

• You must acknowledge on your Web site that you offer CPU overclocking, and that the speeds you can achieve will vary from chip to chip. Most of you already do this.
• Your overclocked PC must be stable enough to survive a 24-hour run of Prime95 in our lab.

If you fail to declare that you overclock and the limits thereof on your Web site, we will decline to review your PC above its stock component settings. If you send it to us anyway, we can send it back or clock it down, your call. We'll take failure to complete Prime95 on a case-by-case basis. Repeated failure will very likely result in a public reprimand.

We're happy to answer any questions you might have, and we know you can sell overclocked PCs responsibly.

Good luck,
CNET

[Maccess]

PCs for sale should never be overclocked. If anything,they could even be underclocked. Leave overclocking to users and enthusiasts--most users need an accurate reliable computng device.

[Random_Walk]

They obviously don't overclock the stuff they sell to the masses - instead the author suspects that the OEMs overclock review units so that it gets better test scores (and likely so that it helps to overcome the shortcomings of a certain current OS, the name of which starts with a "V").

If reviewers aren't careful about checking for that, benchmarks could very easily fall into question as well...

Glad Apple doesn't stoop to such tactics (if indeed it's the OEMs who do it).

[pithenumber]

underclocked is going slightly to far

Apple actually does have a system that with mild overclocking, the Mac Pro auto overclocks itself to 3330 I think
very small bump from 2930 though

[EricJM001]

Hardware overclocking is an interesting phenomenon. It's cool that there is an entire industry of water coolers, and oversized fan/heatsinks to support these amateur computer engineers, but I have to think that they are doing more harm than good.

Intel is willfully limiting the speed of their processors to allow for generous overclocks that get hyped up on websites like tomshardware.com and anandtech.com and it sells lots of chips. Legitimate users end up with slower computers or have to pay hundreds of extra dollars to get faster systems in the first place. Finally, all these overclocked systems are getting sold off on Craigslist or EBAY (yearly I suppose) to unsuspecting second hand users who may not know the chips were run our of spec voiding all warranties.

One friend of mine spent an entire weekend overclocking and benchmarking his computer. I cannot help to think that even if he used his computer 24x7 for the next three years, he would not recoup the 16 hours he lost up front. it's sounds more like a hobby, than a timesaver.

[pithenumber]

now overclocking for 16hr to attempt the fastest stable speed the computer will run at is more like a hobby

but spending 10 minutes bumping a CPU by a couple hundred MHz is a time saver

[cp256]

More harm than good?!?!?!? I have always overclocked my personal workstations if it was possible. I still use a liquid cooled and overclocked AMD XP 2100+ that I built in 2001 (it started as an XP 1900+) every day for work. I have reaped the benefits of the overclocking on it for six to sixteen hours a day, weekends included, for over 8 years. I probably spent 50 to 60 hours on the original modifications, overclocking and benchmarking, but I had a lot of fun doing it. It most certainly is a hobby, but it has practical long term benefits once you establish your maximum stable and safe overclock.

My current hot setup is based on a QX9650 that will run up to 4.3 ghz (almost 70% over :) on air that I usually run at 4 ghz (50% over) and it will grind away all day long on anything I feed to it. I also have a 9 year old 1.4 ghz P4 that has been running a 20% overclock 24/7 for nine years, it's my freebsd firewall server and router for all my non-unix boxen. I even have some other overclocked freebsd boxen that host customer websites 24/7 with 99.9+% uptimes. It's virtually free horsepower, why NOT take advantage of it?

Hobby? Yes! Daily drivers? Of course.

Sure extreme overclocking is not for clooless noobs and vendors have no business misrepresenting overclocked systems, but it is absolutely not a harmful thing on its own.

[streamline35]

I agree with CP - sure some people make it a hobby, but it is actually a very practical application, and if you do it correctly, as cp said, it is basically free horsepower.

I knew nothing about overclocking, so I went online and spent a couple hours reading up on it, then another hour playing with it, and I very easily got my q6600 from 2.4 ghz to 3.0 ghz on air, with no significant rise in temperature and only a tiny fraction of a volt added (1.250 to 1.300 volts I believe). Especially being a quad core, it makes a huge difference while costing nothing but a couple hours of my time to learn something valuable.

[c|net Reader]

You must understand that overclocking varies from chip to chip because the manufacturers bin their components. They find a clock speed at which a significant number of CPUs will run and make that their top end speed. They brand the chips, advertise, document, and sell them accordingly. Of those that fail the high speed tests, another group will pass a somewhat slower top speed. Those are handled separately, and the process repeats until the vendor has found a group for most of the chips they've made. As you can surmise, some of the chips in each group are almost good enough to be in the next higher group, so they can be overclocked beyond the rated speed, so long as they aren't overclocked beyond their capacity. Other chips in a given group were just good enough to be in that group and have no capacity for being clocked any higher.

The result of this is that some chips can be overclocked safely and others cannot. How do you know which is which? Try it.

[monkeyfun14]

Well obviously any OEM overclocked machine may have issues all chips are different even of the same model some go higher then others its a matter of chance.

The OEM just slaps them a chip on a preset mobo and moves the machine down the line.

[Suny Buffalo]

Obviously. Responsible overclocking means more than just acquiring the chip and overclocking it overnight. Shamefully, I've meet many gamers who talk the talk but I know that they just boasting.
The proper way to overclock is to do it gradually. It's like if you were given a cruise ship, you'd run it at the proper speed, until the engines had been "properly run", then after a time say a week, you can stretch a bit more and so on.
The same with the CPU. Run it at its properly designed speed for a week or two (so as to acclimatise itself to temperature, loads, etc..) Ten raise (overclock) at 1.5X (times) and let it run for a week or so. Do so until it reaches a point where noticeable depreciation, degradation or sluggishness and performance begin to manifest itself. This tells you where you can decrease at marginal increments to reach total focal, optimal performance.
Try this. You might be surprised at how the CPU has been "properly run" and has accommodated itself to its environment and perform optimally. ( I should know, I was overclocking since the inception of Packard-Bell with CPU at 33 MHz and pushed to 66 MHz (actually, reaching 60-61 MHz) up to today's most CPU [AMD or Intel]) Good Luck!!
Suny

[pithenumber]

sorry
my Ph2 720 is running @ 3300MHz with stock volts right now
far cry from 66MHz

[streamline35]

I believe the standard stability test is 16+ hours of prime95 being run without any crashes or errors. If that won't crash the system, I doubt a week of normal use will.

[c|net Reader]

Sorry, but chips don't need to acclimate to "temperature, loads, etc." What matters is whether the chip has been clocked beyond what the slowest component can handle -- the weak link in the chain, so to speak -- and whether the chip is running too hot to be reliable. When a chip works but suffers from heat, you can add better cooling, but you can't run it faster than the slowest component inside.

[gbelk08]

All I have to say is that Rich Brown is a badass. Not many journalists have the balance between the responsibility of telling it like it is, and the integrity of taking stock of the situations of the said two machines and deciding not to name names without a warning first.

Thank you so much for helping to inform those of us that don't know much about over-clocking, but feel like we deserve the truth.

[Notoapplefanbois]

He doesn't known that much, if he knew enough then he'd know that the companies are being cheap ***** and using C0 chips instead of the new D0 chips which OC to 4.6ghz stably.

[spencer_tennant]

You shouldn't over clock Intel because they are usely already overclocked.

[pithenumber]

huh?
you can lower volts on a q660 and still oc it a fair bit

lots of headroom

[streamline35]

A processor is by definition not overclocked when you buy it from intel or AMD. Overclocking is pushing the clock higher than they specify. And that's a silly thing to say anyway - as spencer pointed out, the q6600 overclocks like a dream. I know because I have one.

[Tod Smith]

Please END the false PC OC hopes!

OEMs from CPUs, GPUs, and Motherboards have higher 90 day failure rates.

[thallwyn]

Rich, thank you for doing this. I hope you lead others to be as vigilant in their reviews.

[techman21]

Overclocking by definition puts you at risk for instability and CPU meltdown.

[Htos1]

My AMD 4200+ box has been 10% overclocked for three years now,no anomolies,reboots,meltdowns,slowdowns,bsod's,anything.However,it does double duty as a space heater during winter months. :)
Good article,and as I recommend,give the intel systems to business clients and Mom and Dad.

[thebigsmann]

I agree, I have the 4200 x2 overclocked to 2.57 (up from 2.2) and use it daily for solidworks modeling. Haven't had a problem with it in 3 years. I found the best method is to have nTune overclock it, then take those settings- apply them to the bios, and uninstall nTune. Works well and has been reliable for me so far!

[pugetsys]

Great post, Rich! Its great to see reviewers fighting back against the industry hype.

If you really want to make sure cooling is adequate, you need to stress the video card(s) as well. I personally recommend running Prime95 (make sure you're stressing all CPU cores) and Furmark at the same time. Just a warning though, Furmark is brutal! All of our PCs have to pass that test before we ship, but you'll be surprised how many other units out there will overheat and fail.

Jon Bach
President - Puget Systems
http://www.pugetsystems.com

[RobWilliamsTG]

Good read. I find it interesting that PC builders will sell overclocked PCs that actually throttle down to the CPU's stock speed after being stressed... it's a little foolish. I don't think Prime95 is a good method of determining a stable overclock, however. We're kind of limited with Core i7 at the moment where stress-testing is concerned, but Prime95 is far outdated and is in no way able to push Core i7 processors to their breaking-point. Prime95 used to be the standard with single-core processors, and it's never been updated. We're able to run an instance on each core, but that doesn't negate the fact that Prime95 is unable to tap the extra CPU space past a simple math calculation.

Take LINPACK on Core 2 Quad, for example. Running a LINPACK-based stress-test tool on those processors can heat the CPUs up to 20C hotter over what Prime95 can manage. The problem, of course, is that Intel hasn't released LINPACK for Core i7, so we're out of luck until then. Still, if a Core i7 processor passes a Prime95 test, it in no way should delcare a stable overclock. That benchmark is far too mild on today's processors.

[rhbrown]

A reader of the Tech Report emailed me recommending Linpack as well. I'm looking into it. I actually asked four or five different vendors how they stress their overclocks before shipping to customers and the majority recommended Prime95. Perhaps that indicates they're not stressing their own overclocks enough. If we decide to switch, we'll update accordingly.

[posershadow]

Lesson #1: Build your own PC, way cheaper and if you are buying an i7, you probably know how to build a computer.
Lesson #2: Overclock it yourself, you'll probably do a much better job with voltage optimization.
Lastly, why are you still using prime95? linx and IBT are much better at testing stability much faster.
I run my 920 on a classified at 4.6ghz under water, stable 24/7. I am able to leave most of my voltages on auto except vcore and vdimm, which I enter myself.

[solblack]

Stop!
There are plenty people who overclock and never have a problem usually gamers who know what there doing.

One friend of mine spent an entire weekend overclocking and benchmarking his computer. I cannot help to think that even if he used his computer 24x7 for the next three years, he would not recoup the 16 hours he lost up front. it's sounds more like a hobby, than a timesaver.

Yes it can be a hobby, but never a timesaver most do it for performance most are gamers.

Obviously. Responsible overclocking means more than just acquiring the chip and overclocking it overnight. Shamefully, I've meet many gamers who talk the talk but I know that they just boasting.
The proper way to overclock is to do it gradually. It's like if you were given a cruise ship, you'd run it at the proper speed, until the engines had been "properly run", then after a time say a week, you can stretch a bit more and so on.

Real Gamers don't boast!

[cp256]

It's been a great timesaver for me. Whether it's rendering or grinding out batched volumes of PDFs or chewing on one of my horribly recursive perl utilities, I have saved countless hours of work time over the past 20 years with overclocking. I would never look a gift horsepower in the mouth :)

I used to game years ago before the graphics cards really started to get hot, but even though my overclocked QX9650/SLI/RAID box is a great gaming rig, I have never gamed on it. I even have a wireless gamepad for it, just no desire I guess.

[ikramerica--2008]

very much a time saver. way back in the day I overclocked a marginal iBook G3 using nothing more than a knife and a metalic pen, but the 30% boost in performance made every task faster for the next 1 plus years. Overclocking a G4 offered similar results, and render times vastly improved. (I know some Windows people would scoff here, but for a time, Macs were quite clockable).

The question isn't if the person will "get that weekend back" but whether that weekend was a critical time for them. Obviously it wasn't. But later on, if you use your machine for business, when you are at a deadline, the faster machine makes a difference. You can't transplant the "lost weekend" forward in time when you need those hours. If you have 2 weeks to finish a project 6 months from now, you can't just take those 16 hours from last weekend and add them on then. If only life worked that way...

[thebigsmann]

Unfortunitly you don't need to 'break in' a processor the same way you 'break in' a newly rebuilt engine. Gamers take their pride in how they can overclock their systems for their needy games. I don't game, but have had extremely good luck overclocking my processor to better the use in Solidworks.

[cccmd-2009]

By not reviewing and publishing the unstable systems, you are not serving your readers. You should review these systems and make their builders stabilize them. It is a golden opportunity to review an aspect of boutique PCs which is too often ignored - the post sale customer support. You cited in your AVA review of the poor track record regarding PC vendors and customer support. Yet when presented with an opportunity to shed light on this, you instead protect the vendors (with anonymity) who send you an a broken machine.

[rhbrown]

Reviewing the complete customer service experience (RMA, shipping out and back for repair) typically involves purchasing a system. Our general practice is that we work with vendors directly to acquire review products. We've never made that practice a secret (we're also not alone), so I hope it won't come as a surprise.

Purchasing products to review certainly has its benefits, in particular the ability to report on the full customer experience. However, one of the primary benefits of working with the vendors is that it gets us access to and information about unreleased products.

If we were able to get a vendor-submitted system through that same vendor's regular customer service channel (generally tough without a paper trail documenting a purchase), there's a very good chance the vendor, a boutique especially, would be able to identify that the system came from us. That would then ruin the chance that system would be treated like any other.

Now I'll admit there's a certain double-standard here, in that I'm saying we don't want to report on vendor-filtered customer service, even though the hardware arguably comes through that same filter. The reason is that we can follow up when the product ships to stores or goes on sale online to ensure what we've written about is the real thing. Most vendors realize that, and thus know they run a very real risk of getting busted (Editor's Choice awards can be revoked, review scores lowered) if they submit bogus hardware. Without seeing the return and repair operation for ourselves, though, we have no way to comment on the complete customer service experience.

I've actually tried to respond to this post two or three times, because you open a can of worms (at least for me) with the suggestion that I'm protecting these vendors (note to future commenters: if you want good odds for plaguing me over a weekend, question my integrity at 5:00 PM ET on a Friday.) I thought instead of writing a treatise on review theory, I'd address the customer service issue first. If anyone's still interested tomorrow, I'm happy to address the protection question. Preview: I don't think I am.

[lumpoco]

Wow. I hadn't realized that some of the people that frequent this site have no idea on how overclocking works. Some have said that instead of overclocking the cpu that you should actually under clock it. While others have said that you shouldn't overclock the cpu because Intel has already overclocked it. And there are some that also say that you shouldn't overclock the cpu because it makes your computer unstable and will lead to a meltdown. All three are false statements made by those who don't even know how to take apart their computer case to upgrade a memory bank, hard drive or video card.

The only time anyone should be under clocking their cpu is when they want to lower the heat that's being generated by a faster running cpu. Also, Intel doesn't make every cpu they way they had planned every time. If they are making a 3.5GHZ cpu and it is not stable to run 24/7 without extreme cooling then they sell it as a 3.5GHZ cpu. This is because most people who use computers are stupid. Some think that you turn the computer off by unplugging the power supply cord. Hahaha. While others don't understand why when they save their files at their work or school computer that the files do not show up at home. Hahaha.

Getting back to my example, if Intel can't use a certain 3.5GHZ cpu at a guaranteed speed of 3.5GHZ they will have to sell it at a lower speeds. They find that it is very stable under normal or minimal cooling if the 3.5GHZ cpu is set at speeds from 3.0 to 3.3GHZ. Now if one of you guys happen to buy this "rejected" 3.5GHZ cpu you will be guaranteed to get a cpu speed of 3.0 to 3.3. If you happen to have a setup that allows you to cool the cpu with extreme measures such as employing a cpu cooling system that costs anywhere from $80 to $200 I'm certain that you can reach speeds greater than 3.5GHZ. Lastly, if cpus were not meant to be run faster than normal then perhaps some of you guys out there should not be "speeding" in your automobiles.