Puget Systems print logo
Read this article at https://www.pugetsystems.com/guides/872
Article Thumbnail

Most Reliable PC Hardware of 2016

Written on December 30, 2016 by Matt Bach


At Puget Systems we record a huge amount of data from every system we sell including benchmarks, BIOS screenshots, thermal images, and system photos. In fact, much of this data is published on our website and can be accessed through our part information pages. Simply scroll to the bottom of a part information page and you can view this data from recent systems we built.

While this data is nice, one of the most important things we track is something we don't have publicly available at all times: the failure rates of individual components. Reliability is one of our primary goals, so this data invaluable for tracking both individual component, product line, and overall brand failure rates. With 2016 coming to a close, we thought we would run some reports and share what hardware we found to be the most reliable in the last year.

One thing we want to make clear is that we can only comment on products that we have sold significant quantities of. There may be an even more reliable product out there, but either we have not sold it or we have not sold it in large enough quantities for us to have enough data to call it reliable or not. With that said, let's get started!


A reliable motherboard is essential in a high quality computer. Not only is a motherboard very difficult to swap out, but the effects of a poor quality motherboard can be far reaching and difficult to troubleshoot. This is complicated by the fact that motherboards are one of the most complex components in a computer. There are SATA, USB, fan, and network controllers as well as the physical ports, audio chips, and everything else that is needed to interconnect every component in your system. This is a huge number of delicate parts that have to work perfectly together and any one of these could potentially have a problem. If there is a single dead USB port, slight static over the audio, or the voltage levels are measured outside of norm, it does not meet our standards and is considered to have failed.

Because of this, motherboards have one of the highest overall failure rate of any core component with about 1 out of every 18 motherboards (5.5%) failing for one reason or another. This may seem like a high failure rate, but the silver lining is that nearly all of these failures we catch in-house before the system is shipped to the customer. In fact, motherboards as a whole only have a 1% failure rate (or one out of every 100) when you only look at issues that occurred in the field.

We like to look at reliability in two ways: first, in terms of overall reliability rates which includes both DOA (dead on arrival) failures and failures that happen in the field. Second, we look at only failures that happened in the field or failures we were not able to catch in house before the system shipping. Sorting with these methods, there are a number of boards we found to have an above-average history of reliability in 2016:

Best overall reliability:

(Lowest combination of DOA and field failures)


Asus H170I-Pro/CSM
(2.1% overall failure rate)

Asus X99 Deluxe II
(2.27% overall failure rate)


Best long-term reliability:

(Lowest field failure rate)

Asus H170I-Pro/CSM
(0% field failure rate)

Asus P10S-M WS
(0% field failure rate)

Asus X99-M WS
(0% field failure rate)

CPU (Processor)

2016 was a bit of a rocky year for CPUs in terms of failure rate. Last year, we only saw an overall failure rate of .33%, but this year it jumped up to .76%. This is actually still very good, just not as good as it was last year. We did not sell enough AMD CPU/APUs in 2016 for us to make a call on their reliability, but for Intel CPUs we can give numbers divided up between all Intel CPUs, Core i3/i5/i7, and Xeon E3/E5:

Intel CPUs Total Failure Rate Field Failure Only
Overall .76% .28%
Intel Core i3/i5/i7 1.0% .34%
Intel Xeon E3/E5 .32% .16%

One interesting thing to point out is that we saw less than half the failure rate with Intel Xeon CPUs versus the Intel Core i3/i5/i7 CPUs. Due to the jump in Core i3/i5/i7 failures compared to last year, we are going to make the call that if you want the most reliable CPU, we recommend using a Xeon CPU if possible.



RAM (Memory)

The brands and models of memory we offer in our workstations change based on a number of factors including reliability and availability, but in 2016 we primarily offered Crucial and Samsung RAM. Separating standard RAM from ECC and Reg. ECC RAM lets first take a look at the one model of standard RAM with no failures in 2016:

Desktop DDR4 RAM Failure Rate
Crucial DDR4-2133 4GB (CT4G4DFS8213) 0%

Interestingly, this exact model made our most reliable list last year as well which makes it extra good! We offer both 8GB and 16GB version of this stick as well, but unfortunately both of those models had a number of sticks that failed. They were still decent overall (and had very low field failure rates), but nothing like the Crucial 4GB stick.

Server/Workstation DDR4 RAM Failure Rate
Crucial DDR4-2133 8GB ECC Reg. (CT8G4RFS4213) 0%
Crucial DDR4-2133 16GB ECC (CT16G4WFD8213) 0%
Samsung DDR4-2133 8GB ECC Reg. (M393A1G40DB1-CRC) 0%
Samsung DDR4-2133 16GB ECC Reg. (M393A2G40DB1-CRC) 0%
Samsung DDR4-2133 32GB ECC Reg. (M393A4K40BB1-CRC) 0%

ECC RAM (and REG ECC) is specifically designed to be reliable, and in 2016 it actually turned out to be about 4x more reliable than standard RAM (.2% vs .9%). Because of this, we had a number of different models that we sold a significant quantity of that had absolutely no failures.

Hard Drive

Just like RAM, we typically only use a few brands of SSDs and hard drives that we historically know to be extremely reliable. In 2016 we sold primarily Samsung and Intel SSDs and the reliability for both brands was excellent. In fact, the failure rates were so low that instead of listing individual models we are simply going to list the failure rates for the two product lines we found to be exceedingly reliable in 2016:

Solid State Drives Failure Rate
Samsung 850 Pro (128GB-2TB) - SATA 0%
Samsung 950 Pro (256-512GB) - M.2 x4 0%

If you compare the two drive lines above to our list from last year, you will notice that it is much shorter. The Samsung 850 Pro line is still exceptional with no failures at all in 2016, but the 850 EVO is no longer on the list. The EVO line is still terrific with only a .26% failure rate, but solid state drives in general are so reliable that that is not all that uncommon. Also no longer on the list is the Intel 750 NVMe drives - these had a number of failures in the field this year which knocked them off.

Traditional platter drives - with all their moving parts - did not fare as well in 2016. In fact, the failure rates were high enough that we are not comfortable putting any models onto this list. We primarily offer WD Red, Gold, and RE drives and while none of these were terribly bad in terms of reliability (we wouldn't carry them if they were), at the same time they are not anything special. Platter drives as a whole had about a 1% failure rate overall, with more than half of the failures occurring in the field. So if reliability it a key concern for you, we recommend using an SSD if possible.

Video Card

Like motherboards, video cards tend to have a bit higher failure rate than other hardware components. In addition, with the wide range of video cards we offer (including a mix of different brands), naming the most reliable model is a bit tough as we don't tend to sell incredibly large quantities of any one card. However, of the cards we sold enough of to have a good feel for their reliability, there are a few that stand out with no failures in 2016. In no particular order, these cards are:

EVGA GeForce GTX 950 2GB SC
(0% failure rate)

EVGA GeForce GTX 970 4GB SC
(0% failure rate)

EVGA GeForce GTX Titan X 12GB
(0% failure rate)

EVGA GeForce GTX 1070 8GB
(0% failure rate)

PNY GeForce GTX 1080 8GB
(0% failure rate)

PNY Quadro M2000 PCI-E 4GB
(0% failure rate)

Video cards as a whole have a couple of interesting trends we also wanted to point out:

  1. As a whole, Quadro cards are about three times more reliable than GeForce cards. We saw a low 1% failure rate overall with across the Quadro product line and only had a single card fail in the field. GeForce, on the other hand, had a higher 3.4% overall failure rate - although only .6% of the GeForce cards we sold had issues in the field. This may seem odd since only a single Quadro card made it onto our list, but this is mostly because we don't sell huge quantities of Quadro cards compared to GeForce. In addition, for Quadro we only sell a single brand (PNY) and there are not variations of each model - such as overclocked versions - so there is also higher chance of their being a failure for each specific model. Taken as a whole, however, Quadro cards very noticeably have a lower failure rate than GeForce.
  2. Overclocked (or SC) cards from EVGA are actually very good in terms of reliability. Where standard EVGA cards has a total failure rate of 1.87% (.65% in the field), the SC cards were actually slightly better at 1.58% (or .32% in the field). It is possible that the cards will burn out faster over time, but going back to the start of 2015 EVGA SC cards did not have any higher of a failure rate than their standard counterparts.

Power Supply

Out of all the power supplies we sold an appreciable amount of in 2016, there were two models that stood out as being extra reliable:


EVGA SuperNOVA 550W G2
(.9% failure rate - 1 PSU)

Seasonic M12II EVO 520W
(.76% failure rate - 1 PSU)


Oddly, both of the power supplies are relatively low wattage (for our workstations at least). We use very high quality power supplies from the most reliable manufacturers, but it is interesting to see that none of the higher wattage and higher efficiency power supplies made our list this year. In fact, compared to the 2.6% overall failure rate from 2015, power supplies as a whole increased to a 3.11% failure rate in 2016 (with less than half failing in the field).


So there you have it: the most reliable hardware we sold in 2016. While there is too much data to make many broad generalizations, there are a few trends we want to point out:

  • Samsung SSDs continue to be one of the most reliable components within our systems, with a total failure rate of just .16%
  • In the past we haven't seen a large difference in reliability between Intel Core i3/i5/i7 CPUs and Intel Xeon CPUs, but this year Xeon had about 1/3 the DOA rate and just under half the failure rate in the field.
  • ECC and Reg. ECC RAM is much more reliable than standard RAM. Where normal DDR4 RAM saw about a .9% failure rate (with a .14% failure rate in the field), ECC and Reg. ECC RAM had just a .2% failure rate (.04% in the field).
  • Power supplies took a small hit this year, with a combined failure rate (across primarily Seasonic and EVGA units) increasing from 2.6% last year to 3.1% this year. This isn't anything hugely concerning quite yet, but it is something we are keeping an eye on.

Keep in mind that if you purchase a system from us that includes parts not on this list, that does not mean you are at a significantly higher risk of your machine failing. Our internal testing catches the majority of problems before we ship the system so if anything this list is more about the parts that make the production process smoother for us than parts that our customers should specifically use in their systems.

This wraps up the most reliable hardware of 2016. Overall, it was a great year for reliability at Puget Systems with just the normal shifts in reliability as new products and technologies come out. We saw some small increases in failure rates compared to 2015, but the majority of those failures were found in-house. As a whole, hardware appears to be pretty stable in terms of reliability without a significant shift in any direction.

Tags: reliability, reliable, hardware, 2016
Daryl Dixon

Happy new year 2017 Puget. May ur business grow well this year. And btw, Nice review. Thank you.

Posted on 2017-01-02 18:21:52
John Bovay

Hmm... Most interesting and informative. Thank you. One thing though. Does resume-from-suspend and hibernate work correctly with the EVGA GeForce GTX 1070 8GB / 08G-P4-6170-KR under Windows 10 AU and Ubuntu 16.04.x LTS? There are numerous help threads on forums.geforce.com and devtalk.nvidia.com complaining that Pascal cards of all models frequently do not support this function.


GTX 1080 FE random freeze - GeForce Forums

Posted on 2017-01-07 10:55:58
Dolorem ipsum

you should not use Hibernate mode, esp. if you system supports S3 sleep.

Posted on 2017-02-01 19:29:15
Hayder Hussein

I read this interesting discussion about the difference in reliability, heat issues and future hardware failures of the reference design cards vs the OEM design cards:

The opinion was strongly against buying the OEM design cards. Especially for computing and 24/7 working of GPUs.

It seems there is no citation or any scientific study backing up the opinion, but it is claimed that they have a first hand of experience of installing thousands of NVidia cards before.
So what is your company's experience about this? Should we avoid OEM design cards and stick with the original NVidia reference cards in term of reliability and heat issues?

Posted on 2017-01-29 08:23:21
John Bovay

IIRC nVidia graphics cards have a temperature sensor on the GPU chip but none on their VRM. Check out the thermal images of the two Founders Edition (reference) cards compared to those of an assortment of OEM designs in the following article:

28. November 2016
[UPDATE] Pascal-Roundup: GeForce GTX 1070 und GeForce GTX 1080 im Vergleich

As well, here's a video which shows that the much maligned (by PC gaming enthusiasts) reference cooler does the best job of minimizing overall system temperature--and noise:

Dec 28, 2014
Internal vs Rear Exhaust GPU Coolers - YouTube

Posted on 2017-01-29 13:13:27

Interesting question and one that is a bit hard to answer since we don't really track individual cards by usage. I will tell you, however, that we lean towards reference cards if the card is expected to be put under a heavy load or if multiple cards will be in a system. Many of the 3rd party designs like the EVGA ACX and ASUS STRIX series don't have very good rear exhaust so the air tends to stay in the system and you have to vent it with the chassis fans. That is fine for a single card, but as soon as you stack multiple cards into a system it can produce a lot of heat that is hard to get rid of. The Linus video John posted in reply to your comment lines up pretty closely what we have seen in our testing.

I did go ahead and pull some failure numbers from the last two years. This is looking at all the reference cards we sold (EVGA, ASUS, and PNY mostly) versus the EVGA ACX and ASUS STRIX cards (which are the only non-reference cards we tend to sell):

Total Failures: Reference 1.8%, EVGA ACX 5.0%, ASUS STRIX 6.6%
DOA/Shop Failures: Reference 1.0%, EVGA ACX 3.9%, ASUS STRIX 1.5%
Field Failures: Reference .7%, EVGA ACX 1.1%, ASUS STRIX 3.4%

Again, we don't know the specific usage for each card, but this is looking at about 4,000 cards in total so it should average out pretty well. If anything, since we prefer to use the reference cards in 24/7 compute situations this is making the reference cards look worse than they actually are. The most telling is probably the field failure rate since that is where the cards fail over time. In that case, the reference are only a bit better than the EVGA ACX, but quite a bit better than the ASUS STRIX cards.

Overall, I would definitely advise using the reference style cards for anything that is heavy load. We find them to work more reliably both out of the box and over time, and the fact that they exhaust out the rear really helps keep them cooler - especially when you have more than one card.

Posted on 2017-01-30 19:53:52
Hayder Hussein

Hey Matt,

That was an excellent comment. Maybe if you put this information in the article itself. I am sure many will find it interesting.

Alright, it is clear now that the reference design is the best for heavy loads. This will beg this question which is a bit harder to answer.

Recently Nvidia began selling their own cards by themselves (with a bit higher price). What will be your preference? The cards that Nvidia are manufacturing and selling by themselves or a third party reference design cards like EVGA or Asus ?

Posted on 2017-01-30 20:21:09

As far as I know, NVIDIA is only selling their own of the Titan X Pascal card. I think that was just because supply of the GPU core or memory is so tight that they couldn't supply all the different manufacturers so they decided to sell it directly. I believe the goal is to get it to the different manufacturers eventually, but who knows when/if that will happen.

If they start doing that for the other models too, there really shouldn't be much of a difference between an NVIDIA branded card and a reference Asus/EVGA/whatever. Really hard to know if NVIDIA would have a different reliability than other brands but my gut instinct is that the difference would be minimal.

Posted on 2017-01-30 21:28:07
Hayder Hussein

I was interested in either the Titan X Pascal or the GEFORCE GTX 1080.
And it seems the GTX 1080 is also sold by NVIDIA:

There shouldn't be much difference though compared to other 3rd party reference cards, as you said. But with the minimal price difference, maybe the best bet will be to buy from NVIDIA directly.

Posted on 2017-01-31 00:13:04

Interesting, I hadn't realized they were doing that for the other cards as well. We always buy through distribution channels so except for the Titan X (where we have to buy direct) I don't think these generic NVIDIA cards ever really show up. I'm not actually sure which way is the better way to go. Physically, the cards should be identical no matter who you buy from so I think the main difference is going to be the warranty and RMA experience. I can't really speak to the RMA experience between brands though, we have direct contacts with ASUS/EVGA/NVIDIA so we don't go through the normal consumer process.

I doubt there is going to be much real world difference, however, so if you feel better about buying direct from NVIDIA, I say go for it. Worst case you spend a couple extra bucks.

Posted on 2017-01-31 00:44:50
John Bovay

Hi Matt.

What has been your experience re the reliability of resume-from-suspend with the EVGA GeForce GTX 1070 under any OS version?


Posted on 2017-02-02 09:13:19

S3 sleep usually never has any problems - at least ones caused by the GPU. Sometimes we have issues caused by the motherboard, but that is usually due to a buggy new BIOS revision. When that happens, we can usually just revert to an older version, report the issue to the manufacturer, and wait for an updated version.

We usually disable hibernate on our desktops - there isn't really any advantage over S3 sleep for workstations and with the larger RAM capacities more systems use today it makes for very large hibernate files taking up a bunch of disk space.

Posted on 2017-02-02 19:23:40
John Bovay

Thank you for your insight into this issue, Matt.

There's a lot of hair-pulling on various forums re resume-from-suspend / S3 involving GTX Pascal cards of every make & model (and to a lesser degree the GTX-970 and higher Maxwell cards under a range of GNU/Linux distros) emanating mainly from DDR3 and DDR4-era Intel motherboard users along with a handful of AM3+ fans.

In your experience which modern motherboard models are the least troublesome in regard to S3 when used in conjunction with the GTX-1070 & 1080? Also, from a behavioural standpoint, do the Pascal Quadro cards fare better or worse than the Maxwell units?

I'll close by thanking you and your staff for the treasure trove of on-line articles you've made public. Many a time I've pitched a Puget Systems piece to someone seeking aid for a PC problem. Said articles supply an empirical counterbalance to the rumour & lore of misguided info that passes for general computing knowledge on the Internet.


Posted on 2017-02-03 09:17:31

Most sleep issues we are able to fix in-house and unfortunately we don't currently track the times when we had a problem but were able to resolve it without needing to RMA a part. I went back and over the last two years of failures, however, and of the times we couldn't fix the issue ourselves we have only had to RMA a motherboard for sleep/standby issues three times. Once because it wouldn't resume if a platter drive was plugged in, and twice simply because the motherboard itself was having issues. Both of those two times were actually with Quadro cards (M4000 and a NVS 810) but since a new motherboard resolved the issue that is probably just a coincidence.

On the GPU side, we only had to RMA a card for sleep issues twice over the past two years - one GTX 970 and one GTX 960. We haven't had any sleep issues with either Maxwell/Pascal Quadro cards or the new Pascal GeForce cards that were not fixable.

I know we have seen more sleep issues over the last two years than just those five instances, but this just shows how rare it is to be a hardware problem. Normally the issues we see in-house are caused by things like driver or Windows update bugs, or issues with a BIOS update. Those we can usually resolve with our engineering contacts or by reverting to older driver/BIOS versions. For issues in the field, the problems are usually caused by either those same driver/update bugs or peripherals like USB hubs, USB printers, etc. Sometimes customers have installed software that is causing problems but I don't think that is as common for us. In the DIY community, however, I could see things like fan control software and CPU/GPU overclocking utilities causing sleep issues. Or simply different hardware that we don't sell. We also tend to stick to the highest quality components, so it is entirely possible that some of those people are using off-brand power supplies or something like that due to their budget constraints that is the actual cause of the problem.

Glad the articles are useful to you! We've always considered transparency to be a really good thing, so we're happy to keep publishing this kind of information and answering questions like yours.

Posted on 2017-02-03 21:22:48
John Bovay

Thank you for expanding upon your approach to mitigating S3 issues, Matt.
If you don't mind, I'd like to post a quote of your comments and a link to this page in a dedicated thread on Linux Board - NVIDIA Developer Forums.

Posted on 2017-02-04 01:02:01

Sure, go for it. Just keep in mind that even though we sell and test thousands of systems a year, that doesn't mean that what we see is necessarily 100% correct and what others see is wrong. Computers are incredibly complex pieces of technology so we might just be really lucky or maybe we are doing something different that I haven't even thought of.

Edit: Also keep in mind that most of the systems we sell are using Windows. So for linux-based problems, we don't have nearly the same sample size as we do for Windows.

Posted on 2017-02-04 01:10:23
John Bovay

10-4. In that case I'll post a 2nd thread on forums.geforce.com as well. :)

Posted on 2017-02-04 01:20:03
Dolorem ipsum

Thanks for releasing this information again!

Posted on 2017-02-01 19:26:54

The power supply failure rates in the conclusion are off by a decimal. 2.6% and 3.1%, correct?

I love seeing reliability data, thank you for sharing!

Posted on 2017-04-18 04:56:29

Yep, that was a mistake. Thanks for pointing it out!

Posted on 2017-05-01 17:17:08

About ECC 0.2% and non ECC 0.9%, do you have more data on this? Was the registered ECC better, worst or same than the unregistered, I know the specific percentages are really small but do you have them? Thanks

Posted on 2017-04-26 05:59:28

We actually don't sell a huge amount of normal ECC memory these days compared to the other types of RAM. The actual failure numbers are about .2% for REG ECC and .1% for normal ECC, but since we've sold about 6.5x more REG ECC memory than ECC that isn't really as clear as it appears.. That .1% on the ECC is actually just from a single stick, so if one more stick had failed it would have doubled the failure rate and matched REG ECC. Because of that, I don't think we can really draw any meaningful conclusion comparing REG ECC to standard ECC. My gut instinct is that normal ECC should be a hair more reliable than REG ECC simply because is is less complex (fewer things to fail) but not to a significant degree.

Posted on 2017-05-01 18:18:57

Great, thank you.

Posted on 2017-05-01 23:18:05

That's a ton of bullshit. I wonder how much Asus paid for this.

Posted on 2017-05-15 05:21:38
Bob Washere

I have a dead EVGA GTX 970 and a number of failed crucial products that says your failure rate data is bogus. Let me know if you'd like me to mail it to you.

Posted on 2018-10-04 16:03:35
Bob Cannell

So you have a couple data points and his 8000 or so are bogus. Got it.

Posted on 2019-05-05 17:58:28