Failure rate statistics

[time]

Les taux de pannes des composants

and the translation:

The failure rate of components

Highlights

• On average, Seasonic power supplies are more than 2.5 times more likely to fail/be returned than Fortron Source power supplies, which in turn have twice the failure rate of Antec units.
  
• Some Corsair power supplies are hovering around the 10% return rate mark.
  
• Kingston RAM is an easy 3 times more reliable than their nearest competitor, and a staggering 22.5 times more reliable than OCZ. Come on now, you're all shocked, aren't you?
  
• Hitachi is now the least reliable HDD vendor (according to the return rates for this particular supplier), averaging 3% returns. Samsung averaged 2.5%, WD 1.5% and Maxtor only 1%.
  
• The Hitachi Deskstar 7K1000.B and 7K1000.C both have return rates over 5%. This compares to the Samsung F1 with 3.4% and the Samsung F3 with just 1.6%.
  
• 2 terabyte drives suck, with Samsung F3 EcoGreen at 4.2%, Hitachi Deskstar 7K2000 at 6.9%, and WD Caviar Black WD2001FASS with an impressive 9.7%.
  
• Intel SSDs are about 4 times more reliable than the SSD norm. Which is about the same as for HDD ... I wonder what happened to all those MTBF hours?

 

[Santilli]

Good link, and post. Thanks.

The page listing Power supplies under the heading,
"food"
certainly gave me pause, and wondered who, or what wrote the article;0

 

[Bozo]

Very interesting, Thanks!

 

[Mercutio]

I would be vastly more likely to return and expensive Antec, Corsair or Seasonic PSU than I would an FSP Group cheapie.

 

[ddrueding]

I would be vastly more likely to return and expensive Antec, Corsair or Seasonic PSU than I would an FSP Group cheapie.

An excellent point.

 

[LunarMist]

Maxtor hard drives? :drunk: The article appears woefully outdated.

 

[Mercutio]

Maxtor hard drives?

It's possible. How many years has it been since you last bought one? I think it's been eight years for me.

The numbers on 2TB drives sort of confirm my experiences as well.

 

[time]

From page 1:

The failure rate reported is for parts sold from 1 October 2009 to April 1 2010, for returns created before October 2010. The statistics are based on brands with a minimum sample of 500 sales and those models with a minimum sample of 100 sales.

Maxtor external drives were still in the sales channel in 2009.

 

[Santilli]

"Food:
- 6,16% : Fortron Blue Storm Bronze 500 - 6.16%: Fortron Blue Storm 500 Bronze
- 2,63% : HIPER New S S625 - 2.63%: HIPER New S S625
- 2,26% : Antec EarthWatts EA650 - 2.26%: Antec EarthWatts EA650 "

Hmmm. Perhaps this article is computer generated? Hal lives?;-)

 

[timwhit]

Hmmm. Perhaps this article is computer generated? Hal lives?;-)

It's machine translated from French.

 

[time]

Update: returns made before April 2011 for products sold between April 1st 2010 and October 1st 2010.

My take is that the numbers look entirely credible, and there have been enough of these surveys to draw some conclusions with extreme cases.

I'd also like to ask what people consider an acceptable AFR (Annualized Failure Rate)? Over 5 years, a 20% AFR means that you will have 100 failures from 100 samples (probably not once each, obviously). That's ridiculous.

10% AFR means 'half' over 5 years, i.e. 50 from 100. Lawsuit time?

Even 5% equates to 1 in 4. I think this is about the threshold of acceptance, i.e. you would be unhappy but not necessarily surprised if it's just one component out of many.

The trouble is, a PC is made up of multiple components, all of which may fail prematurely. Excluding the CPU, I count 4 or 5 that will take the PC out of service (I'm ignoring laptops ...). Clearly, the total failure rate will be higher than that of the worst component, but even if component rates are all low, it still adds up.

2% AFR across 5 components translates to about (I'm simplifying the probability calculations) 10% for the PC. So we're back to 50 from 100 over 5 years.

Bear this in mind when you look at the real statistics.

 

[ddrueding]

That sounds about right. I tell users who have gone 3 years that they are beyond the average and those whose machines have done 5 without failure to count themselves lucky and start planning for the worst.

 

[jtr1962]

I've been using the same system since June 2006. Thus far, I've never had anything except a CPU fail in all the machines I've owned. I tend to think my usage patterns are why. The biggest source of stress for electronic components is being turned on and off. Besides creating current surges and spikes, it also creates thermal shock. I tend to leave my main machine on 24/7, except when I know I'll be away for more than about 8 hours. The only thing which I turn off when I'm not using it is the monitor (and that's going on 5 years as well).

Another thing I do about twice a year is clean out the dust. Overheating is another source of stress. I also reapply thermal grease to my CPU when I see temps start to creep up from where they normally are.

That said, I'm under no illusions that I should probably start thinking about my next system, if for no other reason than to preemptively replace the current one before catastrophic failure sets in.

 

[Mercutio]

Holy crap. Are you still using that A7N8X?

 

[jtr1962]

Holy crap. Are you still using that A7N8X?

Yep-still in service after all these years, and still fast enough for most of my day-to-day needs. This is the longest run I've ever had on the same system. And I've been using a 200 GB Maxtor as the boot drive from day 1. The drive already had I think 2 years on it when I installed it in this system.

Next system I build will definitely have an SSD for the boot drive, perhaps only SSDs. RAM has gotten cheap enough that I can probably start off with 8 GB as well. My ideal system would be entirely passively cooled. If I can lose the mechanical hard drives completely then it should be completely silent.

 

[jtr1962]

Speaking of systems going a long time, didn't Greg know someone who was still using something like an 8086 (and this was only a few years ago IIRC)?

 

[CougTek]

My ideal system would be entirely passively cooled. If I can lose the mechanical hard drives completely then it should be completely silent.

An i3 21xx should be easily passively cooled by any huge tower heatsink, even with the fan removed. You could use a Seasonic X400 power supply. It's easily feasible. However, I would still put a very low-rpm fan on the heatsink tower for safety. Many 12-14cm fans are inaudible below 800rpm and summers can be very hot in New York. When the surrounding temperature reach 42C, a completly passively cooled system may not be that great of an idea.

 

[LunarMist]

Could you even hear that over the HVAC?

 

[snowhiker]

Holy crap. Are you still using that A7N8X?

Yep. Me too. Check my sig. A friend bought it in early '05 (I think) and I got it as a hand-me-down in Aug'07.

I was actually going to get a Sandy Bridge system but then the Intel recall happened so I've decided to wait until Ivy Bridge.

 

[CougTek]

At least you have a nice monitor hook to your aging noisy PC.

 

[ddrueding]

...and your printer is older than all of it. I had a 4L until I got the current CP2025.

 

[BingBangBop]

How often do you all need to re-thermal paste your CPU's (if ever)? I just re-did a q6600 that had been running several years. It used to run all four cores in the low 50's@100% but it had slowly creeped up till it was in the high 80's when at 100%. I re-did it and now it is back down to the mid-50's. I used Arctic Silver 5 previously but this time I used Nocturna NH-1 (because that is the tube I had avail).

 

[ddrueding]

I really don't do it ever, though I know it is a good thing. I'm lucky if I can get in the system just to blow it out. Considering how tricky it is to remove most of the heatsinks I use (requiring removal of the motherboard in most cases), I just don't.

 

[LunarMist]

I don't think it should hardly ever be necessary to remove the CPU cooler and reapply the paste unless the unit is overclocking. Sometimes replacements are needed if the fan is dying.

 

[Tannin]

The paste deteriorates over time, Lunar. This is why the CPU manufacturers recommend (or certainly used to recommend) one-time pads, and why they ship new product with a pad, not with paste.

The pads are not quite as efficient (about 2 degrees less effective) but they maintain their function pretty much without deterioration for the life of the part.

So: if you want maximum cooling and are prepared to clean and replace the heatsink every three years or so, use paste. If you just want it to work, use a pad. If you can't find a pad, use paste anyway - it's not as if there are too many parts these days that run anywhere near their thermal limit.

 

[LunarMist]

Retail HSFs appear to me to have some kind of paste, but I don't

 

[Tea]

Retail HSFs appear to me to have some kind of paste, but I don't

^ quite right. Humans rarely come with any sort of paste. Well, not as-delivered. But when you put milk into one end of a brand-new human, aftera while you get a sort of paste out of the other end. Lazy of me, I know, but I have never enquired into its thermal properties.

 

[mubs]

I routinely use my systems for a long time.

Built my dual P3-800 (slot) in Dec. 1999 and used it till May of 2006. It is retired now, but a spare and still works fine.

My current system was built in March 2006; AMD 64 X2 4400 on a DFI LanParty UT.

Nothing in either machine ever failed - except a case cooling fan. I suspected failure in one of the Hitachi 250GB drives in my current system, and being paranoid, swapped it and its sibling out for 2 x 320 GB Samsungs. This was about a year or year and a half ago. Still got the two removed drives. Plan to wipe and check.

I plan to build a new m/c this fall. Time is always the problem (not referring to our forum friend, but am too busy). Always in the planning stage! I bought 2 sticks of RAM from Clocker in the fall of 2007 intending to put them into this machine. They're still in a box somewhere

 

[CougTek]

New RAM sticks in 2007 were DDR2. Your new system should use DDR3. You've wasted your RAM by waiting too long.

 

[mubs]

Naw, they'll work in my current system (which is 5 years old). I just need to put them in. They obviously will not work in the new system I plan to build.

 

[LunarMist]

Retail HSFs appear to me to have some kind of paste, but I don't

^ quite right. Humans rarely come with any sort of paste. Well, not as-delivered. But when you put milk into one end of a brand-new human, aftera while you get a sort of paste out of the other end. Lazy of me, I know, but I have never enquired into its thermal properties.

Would you believe that somebody stole the remainder of my sentence?

 

[Bozo]

Would you believe that somebody stole the remainder of my sentence?

No :-D

 

[time]

After reading countless PSU reviews where they were deliberately overloaded, desoldered in the quest for better photos, and rated on how flat their oscilloscope traces were, I've been trying to reconcile all this apparent wisdom with actual failure rates.

I think that there are different types of engineering. Junior engineering just increases component tolerances (or in this case, 'perceived quality' such as which brand of capacitor) without any regard for cost, which is what senior engineering is all about. The 80plus certification has made this worse, with units being deliberately overspecified and then under-rated to suit the test profile. That costs 3 ways: increased cost, reduced efficiency at 'normal' loads and reduced reliability.

Yes, reliability. Some of the most reliable power supplies in the stats are the least efficient. Building something to handle a load beyond its rating does not contribute to its reliability, and may even detract from it.

If a component is within spec, it's within spec. That's all there is to it. It means it will work as advertised with other components that expect it to meet that spec. No magic points for meeting some specs while being outside normal operating parameters, which pretty much includes running a PSU at maximum load rating, let alone beyond it. The ATX spec says you can only do this for 17 seconds, for chrissakes.

I have no idea where this leaves me as I contemplate which PSU is most likely to last a full 5 years, particularly 24x7.

 

[Bozo]

Time, the computers I build go into an industrial enviorment; dirt, dirt, and more dirt. Also, because we convert AC to DC (and in some cases back to AC) all the SCR switching and power factors that approach 20%, our computers take a beating. We also pump DC back into the AC line through SCRs when the DC motors 're-gen'. A 'scope trace on our 120v line looks like a wooly worm (hairy catapiller). I have used nothing but Antec power supplies for 10 years now with great success. I have had a few fail, but it is usually after they have been running 24/7/365 for 5-6 years, mostly failing because of the dirt.
And, 95% are not on a UPS.
My hardware failures for this year so far are: 5 Supermicro motherboards, 1 power supply - Antec, and 2 hard drives - Seagate.

 

[time]

Thanks Bozo, that's more what I'm after - except that Antec has several different models made by three different manufacturers. According to the return stats, their most reliable model is the 400W Basiq that's made by FSP. The 650W Earthwatts is from Delta and was (relatively) the least reliable of their bulk sellers. I don't understand why the more common 380 and 430W Earthwatts aren't in the list; either they don't sell in Europe or none fail?

The Neo 400 reported must be a Neo Eco, which is a cousin of the original Seasonic Earthwatts. We know it did well - in the first 6-12 months of its life anyway!

Any idea what your current crop of Antecs is made up of?

 

[Bozo]

I have been using 350W to 650W 80+ power supplies for the last 3-4 years. I keep three spares in stock now. 2-MT350 (or MT352) for the minuete cases, and 3 or 4 Earthwatts 430w and 1 Earthwatts 650W. This covers everything I have installed. ~175 computers.
Also, 99% of my cases are Antec, so a lot of them came with power supplies. Mostly the 380W versions.
But:
The early Minuete cases with the MT300 and the early MT350 power supplies had a lot of failures.
I had an Antec power supply (I don't remember the number, but it had removable cables) that wouldn't work with some Intel motherboards.
Nothings perfect. I bought 15 Supermicro motherbooards and 5 died within 6 months.
I am now replacing some older Intel motherboards with swollen capacitors. I have 33 Intel D915GEV motherboards in the plant that die without warning, and the capacitors look fine.
And, I had one Antec power supply that was dead out of the box.
In the last three years the failure rate on most computer parts has dropped. Probably because of the better capacitors, in the better equipment.

 

[CougTek]

I know I replace Intel-brand motherboards from 3-5 years old on a regular basis. MDG (a canadian national reseller) sold most of them. They were probably manufactured by Foxconn. Like you said, the capacitors look fine most of the time, but the boards just die. Fortunately, those are LGA775 platforms so it's still worth it to replace only the motherboard and charge some ~100-130$ to the customer (including the labor).

These days, I use the Asus P5G41M-SI as a replacement board. It isn't on Asus' website, but it's like the P5G41-M, plus a DVI port and it has four DIMM sockets. It works with LGA775 Pentium 4, something that is often not the case with new LGA775 motherboards. It cost me less than 60$, so it is a good value for what it offers.

 

[CougTek]

I've found a picture of the back panel of the P5G41M SI :

You can see all the ports and the four DIMM sockets (DDR2) behind. As I wrote, not bad for less than 60$.

 

[ddrueding]

That does look like a nice board, Coug. Will it let you use the onboard video and the PCIe 16x slot at the same time (RAID card, etc)?

 

[CougTek]

I've never tried. Can't tell. I know I once used a PCI-E 1x wireless network card in the PCI-E16X slot and it worked, but it wasn't on that board.