Announcement

Collapse
No announcement yet.

Reliability of BGA CPUs vs Socket

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

    Reliability of BGA CPUs vs Socket

    I need a small, reliable PC to go into a public kiosk.

    I'm trying to decide between an LGA 1151 system with a small form factor ATX compliant PSU or a "mobile computing solution" with a BGA CPU and a power brick.

    I have worked in a computer repair shop, and failed BGA joints seem pretty common in my experience (thanks RoHS). I don't think I've ever seen a socket CPU fail without obvious abuse.

    I know mobile computers use less power and generate less heat which would be good for a poorly ventilated kiosk environment, but I'm not sure that will translate into a longer service life compared to a traditional SFF desktop design with a CPU socket.

    Would you choose BGA over socket CPUs for this application?

    Here's my preliminary build plan for the LGA 1151 system:

    CPU: Intel Celeron G3900 or G3920 (Celerons work fine for our current needs, but I may ask for an i3 to get some headroom for the future. Thoughts?)
    Cooler: Noctua NH-L9i or Silverstone AR05
    Mobo: Asus H170I-PRO/CSM
    RAM: Cruicial DDR4 4GB
    SSD: Cheap TLC 120GB until 3D NAND becomes cheaper. Adata, Kingston, OCZ, PNY, or Sandisk.
    RS-232: StarTech COM to DE-9 serial adapter.
    PSU: Silverstone ST30SF
    Case: Silverstone ML05

    Alternate Case/PSU:
    iStarUSA S-21 (This would likely need a case fan)
    SeaSonic SS-250SU
    Last edited by Winston Chadwick; 03-02-2017, 06:34 PM.

    #2
    Re: Reliability of BGA CPUs vs Socket

    I have never had reliability problems with BGA processors but many defective VGA/northbridge chips built with wrong materials/designs, CPU sockets fail too, specially 775.

    Comment


      #3
      Re: Reliability of BGA CPUs vs Socket

      Why not just go with a machine designed for use in a kiosk or an industrial class box? If you roll your own using consumer grade parts you end up with consumer grade failures.

      Comment


        #4
        Re: Reliability of BGA CPUs vs Socket

        I suggest looking at the whole picture, for a reliable kiosk PC.

        Mainly, airflow and cooling are crap designs on kiosk cabinets. Enclosing a LCD monitor and lighting also adds heat. You have to ensure good airflow to get heat out of the cabinet.

        While BGA packages are the least reliable- they can't take harsh thermal expansion or flex in a laptop IMHO- your motherboard CPU support IC's are BGA's anyhow. The motherboard's caps (polymer right?) are more important.

        Least reliable is the PC's power supply. Dirty power, people dumping drinks on them, ESD to the touch screens - all make them not so reliable.

        I'd suggest a PC watchdog to force a reboot if the OS/App hang.
        Use a screen-saver (monitor blank) after hours.

        If you are showing video or fast moving ads on higher-res displays, then a faster CPUs is worth using. Underpowered M/B will just cook trying to keep up.

        Comment


          #5
          Re: Reliability of BGA CPUs vs Socket

          I've built a couple of decent i5 computers to travel around the country in kiosks to demonstrate a well known watch brand and their augmented reality.

          Keep the cooler stock, they are more than capable of cooling.
          I'd go for the socketed CPU. Decent brand SSD, Samsung or Intel.
          Any reason for that choice of motherboard and DDR4 ? Seems quite high spec to then be gimped by a Celeron.
          Building a PC is all about balance.

          @srhofmann, I've been building PCs for over 15 years, I can count on half a hand how many failures I've had. Pick quality parts and build it well and there shouldn't be any reason for failure. Regular maintenance to remove dust from the heatsink and the occasional cleaning of hard drive contacts.

          @redwire. Can you explain your "Underpowered M/B will just cook trying to keep up" comment. It makes no sense.
          Last edited by diif; 03-03-2017, 12:25 AM.

          Comment


            #6
            Re: Reliability of BGA CPUs vs Socket

            Originally posted by diif View Post
            I've built a couple of decent i5 computers to travel around the country in kiosks to demonstrate a well known watch brand and their augmented reality.

            Keep the cooler stock, they are more than capable of cooling.
            I'd go for the socketed CPU. Decent brand SSD, Samsung or Intel.
            Any reason for that choice of motherboard and DDR4 ? Seems quite high spec to then be gimped by a Celeron.
            Building a PC is all about balance.

            @srhofmann, I've been building PCs for over 15 years, I can count on half a hand how many failures I've had. Pick quality parts and build it well and there shouldn't be any reason for failure. Regular maintenance to remove dust from the heatsink and the occasional cleaning of hard drive contacts.

            @redwire. Can you explain your "Underpowered M/B will just cook trying to keep up" comment. It makes no sense.
            So over 15 years you've never had hard drives fail? CPU fans, Case fans, power supply fans? I've been in environments where 5 failures in a week was a good week. I've probably had 5 drive failures in the machines around my house in the last 15 years.

            It depends on what level of reliability and the lifespan that you're looking for is. Consumer grade products are still consumer grade, they don't have the quality of components, testing, reliability and lifespan of something that was designed and engineered for industrial/enterprise use. If you have to regularly clean dust off of a heatsink, it wasn't engineered to be used in a dusty environment. Fans on CPUs are terrible ideas unless they have a dedicated fan alarm. I can't ever recall needing to clean hard drive contacts. A gold contact into a well engineered connector with gold contacts should never have to be cleaned. Filters get cleaned or changed, not connector contacts, not heatsinks. I've never seen a consumer grade PC case that had filters in it.

            My background in engineering spans 4 decades some of which was doing design work that required a minimum of 99.99% uptime including the time to repair. A lot of what was designed and rolled out had better than 99.999% uptime over the first 5 years of use. Off the shelf consumer grade stuff might be fine if you're willing to deal with a lower lifespan, increased downtime and more maintenance costs.

            Why not just go cheap and throw a 350 dollar laptop with a SSD in it. No muss, no fuss and just change it every 2 or 3 years?

            Comment


              #7
              Re: Reliability of BGA CPUs vs Socket

              Originally posted by srhofmann View Post
              So over 15 years you've never had hard drives fail? CPU fans, Case fans, power supply fans? I've been in environments where 5 failures in a week was a good week. I've probably had 5 drive failures in the machines around my house in the last 15 years.

              It depends on what level of reliability and the lifespan that you're looking for is. Consumer grade products are still consumer grade, they don't have the quality of components, testing, reliability and lifespan of something that was designed and engineered for industrial/enterprise use. If you have to regularly clean dust off of a heatsink, it wasn't engineered to be used in a dusty environment. Fans on CPUs are terrible ideas unless they have a dedicated fan alarm. I can't ever recall needing to clean hard drive contacts. A gold contact into a well engineered connector with gold contacts should never have to be cleaned. Filters get cleaned or changed, not connector contacts, not heatsinks. I've never seen a consumer grade PC case that had filters in it.
              Total hard drive failure is extremely rare. When hard drives start to have read errors, that's probably down to oxidised contacts. There are plenty of examples all over the internet. Not something you would think oxidises but they do.
              There have been various hard drive brands/models that are more prone to failure, but then that's goes back to my original point of buying the correct parts in the first place. It's the same with anything.
              You are aware that most of the modern Intel fans blow down onto the heatsink ? That's why they need cleaning and they all come combined, heatsink and fan.

              Regular maintenance ensures better up time. I've personally had one hard drive fail. No CPU fans, maybe replaced a couple of noisy case and PSU fans but I couldn't tell you if they were my machines or others.

              Comment


                #8
                Re: Reliability of BGA CPUs vs Socket

                Originally posted by diif View Post
                ... @redwire. Can you explain your "Underpowered M/B will just cook trying to keep up" comment. It makes no sense.
                I am believing an underpowered Mobo/CPU will have a shorter life in a kiosk application. It's not total mobo heat generated- it's chip operating temperature. CPU/GPU/chipset heatsink sizing is never for high load % for full-time.
                I thought, take two CPU/GPU/motherboard chipsets- given equal workload, the less-powerful combination will run at a hotter temperature.

                This is (my) huge generality and I tried to back it up but Intel's tech data on i3 22nm heat is very detailed with no mention of %CPU load and temps. OP's Celeron G3900 LGA1151 2.8GHz 14nm TDP 51W, with 65W heatsink. Similar to 7th generation i3 for heat.

                But for overall reliability, I find keeping things cool helps.
                Strange that Google's HDD tests showed no difference in HDD reliability with temperature.

                Comment


                  #9
                  Re: Reliability of BGA CPUs vs Socket

                  As a rule mechanical devices run better at a warmer temp because the lubricant protects better. Most of the wear on a gas engine happens at start up, very little when it's up to temp.

                  I don't know about today, but I was told 20 years ago the different series of processors was because of variations in the manufacturing process. If it would run at 2.6 Ghz before getting too hot it was called, say a 9600, if it ran at 2.5Ghz for the same temp they sold it as a 9500 and charged less. That would suggest temperatures do have a bearing on lifespan. Just like gas engines will vary 10% so would motherboards. The more parts the more variation in performance. Car manufacturers take an average of an engine, you may get one that puts out 300hp, someone else may get one that tops out at 280hp, and yet someone else may get one that does 320hp. Then throw in the varying drag of a gear box and the rest of the drive train, few identical cars will have the same identical performance.
                  sigpicThe Sky Is Falling

                  Comment


                    #10
                    Re: Reliability of BGA CPUs vs Socket

                    Originally posted by diif View Post
                    Total hard drive failure is extremely rare. When hard drives start to have read errors, that's probably down to oxidised contacts. There are plenty of examples all over the internet. Not something you would think oxidises but they do.
                    There have been various hard drive brands/models that are more prone to failure, but then that's goes back to my original point of buying the correct parts in the first place. It's the same with anything.
                    You are aware that most of the modern Intel fans blow down onto the heatsink ? That's why they need cleaning and they all come combined, heatsink and fan.

                    Regular maintenance ensures better up time. I've personally had one hard drive fail. No CPU fans, maybe replaced a couple of noisy case and PSU fans but I couldn't tell you if they were my machines or others.
                    Again I have to ask how does a gold plated connector oxidize? If you have oxidation on a connector it means you didn't get a quality product.

                    Hard drives do fail and they fail catastrophically, head crashes, bearing failures, lubricant issues(remember Quantum Fireballs). That's just the mechanical parts. Do another list for the electrical components. If they didn't fail on a regular basis there wouldn't be hardware and software raid implementations. I guess over the years I've probably purchased or spec'd hundreds if not thousands of times more hard drives than you have because I've personally seen all of these failures and they were without a doubt catastrophic failures.

                    I have to state again, you shouldn't be cleaning the heatsink on a CPU. If you are you haven't cleaned the air sufficiently before it enters the case. That's the difference between a consumer and an industrial/enterprise class product. I've never had to clean a heatsink in a commercial product. The cases are designed to keep dust and dirt out. A machine can keep working while you're changing the filters, but you don't keep it spinning while you're cleaning around the CPU.
                    Last edited by srhofmann; 03-03-2017, 04:07 PM.

                    Comment


                      #11
                      Re: Reliability of BGA CPUs vs Socket

                      I've seen gold plated connectors fail. Not from the gold corroding but more, contaminants in the air.

                      Moisture can bring in calcium and other contaminants until they build up enough to give a bad connection.

                      The effect is worse near the ocean with salty air. In the old days smoking could cause failures. I've seen tar so thick on electronic boxes it would drip like molasses.
                      sigpicThe Sky Is Falling

                      Comment


                        #12
                        Re: Reliability of BGA CPUs vs Socket

                        Originally posted by rhomanski View Post
                        I've seen gold plated connectors fail. Not from the gold corroding but more, contaminants in the air.

                        Moisture can bring in calcium and other contaminants until they build up enough to give a bad connection.

                        The effect is worse near the ocean with salty air. In the old days smoking could cause failures. I've seen tar so thick on electronic boxes it would drip like molasses.
                        I would love to see pictures of that. Full blown shots, through an optical microscope and with an Auger Electron Microscope to see the actual interface layer.

                        I'm having a hard time with a gold to gold connector doing anything but working. I can envision a problem with something like a tin-plated or cadmium-plated connector to gold connector at the interface. There is the potential there for galvanic action. I suppose this is why you usually see gold to gold connectors and rarely see gold to "other" connectors, in addition to the electrical properties of gold. The question is given enough time and the correct environmental conditions could the gold contact actual be plated with enough tin/cadmium oxide to force the connectors apart and/or plate a thick enough layer of oxide to form an insulator.

                        I got a guy at Molex so the next time we have cocktails I'll try to remember to pick his brain a little bit.....

                        Comment


                          #13
                          Re: Reliability of BGA CPUs vs Socket

                          Molex is a joke. Try working on a 40 pin Canon connector that costs $5000. I had a lock get stuck and came out with the pin. Things happen.
                          sigpicThe Sky Is Falling

                          Comment


                            #14
                            Re: Reliability of BGA CPUs vs Socket

                            Originally posted by srhofmann View Post
                            I would love to see pictures of that. Full blown shots, through an optical microscope and with an Auger Electron Microscope to see the actual interface layer.

                            I'm having a hard time with a gold to gold connector doing anything but working. I can envision a problem with something like a tin-plated or cadmium-plated connector to gold connector at the interface. There is the potential there for galvanic action. I suppose this is why you usually see gold to gold connectors and rarely see gold to "other" connectors, in addition to the electrical properties of gold. The question is given enough time and the correct environmental conditions could the gold contact actual be plated with enough tin/cadmium oxide to force the connectors apart and/or plate a thick enough layer of oxide to form an insulator.

                            I got a guy at Molex so the next time we have cocktails I'll try to remember to pick his brain a little bit.....
                            Check out this thread for some close up photos. https://www.badcaps.net/forum/showth...light=oxidized

                            Comment


                              #15
                              Re: Reliability of BGA CPUs vs Socket

                              Originally posted by diif View Post
                              Check out this thread for some close up photos. https://www.badcaps.net/forum/showth...light=oxidized
                              That's not gold, that's copper PCB traces that have oxidized.

                              Comment


                                #16
                                Re: Reliability of BGA CPUs vs Socket

                                Originally posted by rhomanski View Post
                                Molex is a joke. Try working on a 40 pin Canon connector that costs $5000. I had a lock get stuck and came out with the pin. Things happen.
                                Never seen a 5 grand connector from Molex. I've seen them from Amphenol Aerospace and Glenair. Absolute pain in the ass to work with.

                                Comment


                                  #17
                                  Re: Reliability of BGA CPUs vs Socket

                                  Yes they are. The second one I was populating, a tiny piece of the rubber got into the pin lock. When the pin needed to be removed later, the lock came out with the pin, not a good thing.
                                  sigpicThe Sky Is Falling

                                  Comment


                                    #18
                                    Re: Reliability of BGA CPUs vs Socket

                                    You know, you have the AMD A8-7600 (4 x 3.1ghz , 65w tdp, integrated graphics) at 82$ .. it's about 150-200% more powerful than your G3920 (5140 points vs 3477 points in cpu benchmark). The i3-6100 matches or is a bit faster than this a8-7600, but it's about $120, 40$ more expensive.

                                    It has better built in graphics cards. It can work with cheaper DDR3. It has PGA socket, probably better cooler retention, probably better cooler inside the box.

                                    mb GIGABYTE GA-F2A88XN-WIFI , itx wifi ac , hdmi, dvi, displayport, usb3, bluetooth , everything except m.2 but you can probably squeeze a sata cable inside. 20% off today.
                                    or
                                    mb MSI A88XI AC V2 FM2+ same deal


                                    82$ cpu a8-7600 : https://www.newegg.com/Product/Produ...9SIA1N83U90726
                                    0$ cooler stock

                                    90$ mb GIGABYTE GA-F2A88XN-WIFI : https://www.newegg.com/Product/Produ...82E16813128663
                                    100$ mb MSI A88XI AC V2 FM2+ https://www.newegg.com/Product/Produ...82E16813130852
                                    23$ ram 4 gb ddr3 1600 https://www.newegg.com/Product/Produ...74&ignorebbr=1

                                    I'm not sure but the A8-7600 may even support ECC (though it wouldn't be "validated"/certified by AMD) .. it could be up to the motherboard and the bios if the ECC is enabled in bios. So that could be a bonus as well.



                                    47$ cpu g3920
                                    38$ cooler Noctua NH-L9i
                                    122$ mb ASUS H170I-PRO/CSM
                                    27$ crucial ddr4 memory

                                    So while the cpu is more expensive, you end up cheaper as the mb is 10-20$ cheaper, the memory is about 5$ cheaper, you may not have to buy a third party cooler...

                                    Also.. right now on socket AM4 there's only ryzen 7 which is from 320$ and up. There's some non-zen based processors on socket am4 but only on OEMs (a12-9800, bristol ridge, socket am4) .. but this may change in the new future. So then you'd have cheap am4 motherboards and cheap processors with built in graphics.
                                    Last edited by mariushm; 03-04-2017, 10:57 AM.

                                    Comment


                                      #19
                                      Re: Reliability of BGA CPUs vs Socket

                                      Gideon Labs has really cool failure analysis articles of BGAs, IC's, electrolytic capacitors, connectors etc. using electron microscopy. Really neat reading.

                                      I had PC's running in the back shop where cutting, welding, grinding of steel happens. The dust alone killed the PC's.
                                      So I used the biggest air filter from the local auto parts place, a pie plate and a muffin fan to keep positive clean air pressure in the kiosk. Totally worked great. Got good cooling and no dust on the PC's. Used California Closet Co. to make the cabinets, lol.

                                      Comment


                                        #20
                                        Re: Reliability of BGA CPUs vs Socket

                                        Originally posted by srhofmann View Post
                                        That's not gold, that's copper PCB traces that have oxidized.
                                        I never mentioned the word gold.

                                        Comment

                                        Working...
                                        X