Re: [Soekris] net6501 (Red Light Of Death) debugging
On Thu, Jan 31, 2019 at 03:30:01PM -0500, Michael Stone wrote: > On Thu, Jan 31, 2019 at 07:04:02PM +0100, Harald Welte wrote: > > The problem is that there's basically no alternative to the net6501 on the > > market > > if you require 1U, fanless, small (short 1U enclosure), and a single PCIe > > slot. > [ ...snip... ] Sorry about the double reply the Fan issue deserves some light. In short, if noise is an issue, spend money on Noctua fans. For gamers, fans that move air quietly are a huge consideration. I was amazed at how quiet my son's gaming PC was in normal usage. When I could hear it, it was always the GPU fan making the noise. the Noctua case fans don't make a peep. If you are building a server, you may have to go to 2U. But quiet to silent can be done. My NAS is an Intel Core i5 with 35W TDP and a Noctua NH-L9i CPU cooler and a pair of Noctua 90mm case fans. The cooler + cPU fan is rated for up to 95W TDP. The server is in my basement so ambient never gets above 80F. Because the CPU TDP is so low, I may not have mounted the CPU fan. But I probably did. With the case closed, this PC is close to silent. -- -- Chris __o "All I was trying to do was get home from work." _`\<,_ -Rosa Parks ___(*)/_(*)_ Christopher Sean Hilton[chris/at/vindaloo/dot/com] ___ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech
Re: [Soekris] net6501 (Red Light Of Death) debugging
On Thu, Jan 31, 2019 at 12:42:38PM -0700, Bryan wrote: [ ...snip... ] > > Soekris is really missing out on this I think. As you mention, there is > still demand for the net6501 and nothing really available to replace it. > I think that if the demand were there, someone would have bought the assets from Soren and continued the line. But In this case, I think that our requirements were actually quite specialized. That's why no one replaced Soren. Further, I think that this market will be completely absorbed by Raspberry Pi over the long haul. The Soekris machines are great if you are looking for a relatively large amount of computing power in a low wattage, small form-factor platform. But they fell short on memory capacity and eventually price. I've replaced all of my Soekris Net65xx units with SuperMicro 1U machines that use Intel Atom D525 based motherboards. I did this because the Net65xxs all succumbed to the "Red Light of Death". Used, the D525 units are cheaper than the Net6501 and you can have up to either 8G or 16G of ram at "Best Buy" prices. The machines I have are in Front access 1U chassis that are three times as deep as the Soekris. Stock they don't have the same network density as a Soekris but the motherboard is dual port Intel Gigabit and Intel Gigabit cards cost $20 / port on Amazon in 1x, 2x, and 4x varieties. I don't mean to disrepect the Soekris stuff. In it's day it was awesome but now that the Net6501s have begun to hit the "dying limb of the bathtub curve", there are other options so I'm not going to cry for them. [ ...snip... ] -- -- Chris __o "All I was trying to do was get home from work." _`\<,_ -Rosa Parks ___(*)/_(*)_ Christopher Sean Hilton[chris/at/vindaloo/dot/com] ___ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech
Re: [Soekris] net6501 (Red Light Of Death) debugging
On Thu, Jan 31, 2019 at 07:04:02PM +0100, Harald Welte wrote: The problem is that there's basically no alternative to the net6501 on the market if you require 1U, fanless, small (short 1U enclosure), and a single PCIe slot. There are actually a lot more options these days for those requirements, just look for mini-itx motherboards & rackmount cases. Nobody's making an APU-specific enclosure for that because if you need something that generic then an APU is probably the wrong tool. ___ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech
Re: [Soekris] net6501 (Red Light Of Death) debugging
Hi Harold, Yeah, I have access to a decent rework machine but I'm not sure I want to tackle it. I do have a handful of 6501's to practice on though. LOL! Soekris is really missing out on this I think. As you mention, there is still demand for the net6501 and nothing really available to replace it. Cheers -Original Message- From: Harald Welte [mailto:lafo...@gnumonks.org] Sent: Thursday, January 31, 2019 11:04 AM To: Bryan Cc: 'Chris Cappuccio'; soekris-tech@lists.soekris.com Subject: Re: [Soekris] net6501 (Red Light Of Death) debugging Hi Bryan, On Thu, Jan 31, 2019 at 06:53:04AM -0700, Bryan wrote: > Is that even possible without wave technology? It's called "BGA rework" and it is possible with a BGA rework station [to which I have access to]. However, it's rather tricky and time-consuming, and you'd also have to find the exact part number of the CPU that's used on the boards. It's not like you can normally buy the solder-type Intel CPUs at Digikey or Mouser ;) So no, not something I would want to venture in. At that point it makes more sense to build an enclosure for the PC Engines APU3 and an external PCIe card. The problem is that there's basically no alternative to the net6501 on the market if you require 1U, fanless, small (short 1U enclosure), and a single PCIe slot. The APU3 board exists, but nobody seems to be offering an enclosure to mount the board with an actual PCIe card adjacent to it. Regards, Harald -- - Harald Weltehttp://laforge.gnumonks.org/ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6) ___ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech
Re: [Soekris] net6501 (Red Light Of Death) debugging
Hi Bryan, On Thu, Jan 31, 2019 at 06:53:04AM -0700, Bryan wrote: > Is that even possible without wave technology? It's called "BGA rework" and it is possible with a BGA rework station [to which I have access to]. However, it's rather tricky and time-consuming, and you'd also have to find the exact part number of the CPU that's used on the boards. It's not like you can normally buy the solder-type Intel CPUs at Digikey or Mouser ;) So no, not something I would want to venture in. At that point it makes more sense to build an enclosure for the PC Engines APU3 and an external PCIe card. The problem is that there's basically no alternative to the net6501 on the market if you require 1U, fanless, small (short 1U enclosure), and a single PCIe slot. The APU3 board exists, but nobody seems to be offering an enclosure to mount the board with an actual PCIe card adjacent to it. Regards, Harald -- - Harald Weltehttp://laforge.gnumonks.org/ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6) ___ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech
Re: [Soekris] net6501 (Red Light Of Death) debugging
Harald Welte [lafo...@gnumonks.org] wrote: > hi chirs, > > On Wed, Jan 30, 2019 at 06:37:16PM -0800, Chris Cappuccio wrote: > > > Solder up a new CPU and you will likely have a working unit... > > Do you know something that I don't? From following the list archives > and other resources, it seemed that nobody really knew what was going > on. Sure, there was some speculation (I would almost call it conspiracy > theory) about this bug being similar to the "Intel C2000" bug that hit > Cisco and others. However, net6501 uses E6xx.. > Yeah it's just a half assed guess based on the known issues with the C2000 line and the determination from Soekris that the boards are unrepairable. ___ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech
Re: [Soekris] net6501 (Red Light Of Death) debugging
Is that even possible without wave technology? -Original Message- From: soekris-tech-boun...@lists.soekris.com [mailto:soekris-tech-boun...@lists.soekris.com] On Behalf Of Chris Cappuccio Sent: Wednesday, January 30, 2019 7:37 PM To: Harald Welte Cc: soekris-tech@lists.soekris.com Subject: Re: [Soekris] net6501 (Red Light Of Death) debugging > That's it for now. Maybe this is useful to share. At this point I'm > not sure yet, if this will lead anywhere. > > When I find another timeslot, I want to compare a not-yet-broken board > with one that exhibits the problem. > Solder up a new CPU and you will likely have a working unit... ___ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech ___ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech
Re: [Soekris] net6501 (Red Light Of Death) debugging
hi chirs, On Wed, Jan 30, 2019 at 06:37:16PM -0800, Chris Cappuccio wrote: > Solder up a new CPU and you will likely have a working unit... Do you know something that I don't? From following the list archives and other resources, it seemed that nobody really knew what was going on. Sure, there was some speculation (I would almost call it conspiracy theory) about this bug being similar to the "Intel C2000" bug that hit Cisco and others. However, net6501 uses E6xx.. Regards, Harald -- - Harald Weltehttp://laforge.gnumonks.org/ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6) ___ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech
Re: [Soekris] net6501 (Red Light Of Death) debugging
> That's it for now. Maybe this is useful to share. At this point I'm > not sure yet, if this will lead anywhere. > > When I find another timeslot, I want to compare a not-yet-broken board > with one that exhibits the problem. > Solder up a new CPU and you will likely have a working unit... ___ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech
[Soekris] net6501 (Red Light Of Death) debugging
Hi! The first of several net6501 units bought only 3 years ago are starting to show the "Red Light of Death" symptom. Being somebody who designs embedded electronics and software for a living, I thought I'd invest a (limited) amount of time into investigating this. The obvious thought is of course to re-cap the board an as dried-up electrolytic capacitors are a common aging problem of any electronics, particularly switched voltage converters. I ordered suitable replacement parts and replaced all aluminium caps, both THT and SMT ones. Testing the old ones however showed clearly that there's nothing wrong with them: positionnominal actual C406820 uF 816 uF C397820 uF 819 uF C30 150 uF 148 uF C31 150 uF 155 uF C380220 uF 225 uF C427680 uF 748 uF C418680 uF 716 uF So everything is well within normal expected tolerances. The interesting part is also that this unit has not been used much. It's been sitting in storage unpowered for most of its life, hence any thermal aging that might have dried up the electrolyte would have had to happen at room temperature ;) While looking at the board and probing around with a scope I found the following noteworthy observations: 1) there's *very* intense chirping coming from the board. Trying to locate it is a bit difficult, but using a "paper roll" between my ear and sections of the board it appears to come more or less directly from the CPU ?! 2) the chirping sound clearly correlates to the load of the point-of-load DC converter implemented using Q19. It changes either by load of the consumer (I think it generates a 1V rail for the CPU?) or by the input voltage of the external DC power supply. During successfull boot, the chirp is gone for the first 2s after reset, and then present for 5s, after which it is gone again. Shortly later, the RED led is switched off. 3) changing the DC input voltage has a significant effect on the chirping. A peak of chirping is at 12.3V, and at least another one at 11.1V. 4) if the board is powered up, the current consumption very early on gives away if the unit is booting correctly or not. Is the current consumption at ~ 470mA @ 15V input, it's not booting. Is the current consumption about 600mA @ 15V input, it is starting up correctly. 5) the chirping also gives away if the board is booting correctly or not 6) using just the reset button to power cycle, even many dozens of times will not recover, i.e. not trigger a good boot. That seems only possible when actually cycling the DC input voltage 7) DC input voltage seems to have some kind of impact on the probability of a boot succeeding or not. My best results were at 8.7V DC input. 8) The voltage at the aluminium cap charged by Q19 is very stable. No significant switching noise is present, i.e. the filtering of the voltage rail works as it is supposed to. Excess switching noise is thus unlikely causing this problem. 9) if you look with a scope at the gate of Q19, you can observe that if there's regular / periodic switching at ~ 180 kHz when the board boots successfully. When the board boots unsuccessfully, there's only irregular switching pulses visible, so the converter is probably in some mode where it stops/pauses at no/little load on the output As a summary, I couldn't find anything wrong with the onboard DC/Dc converters so far. The bug is likely somewhere else, and the difference in current consumption is just reflected by different operating points of the DC/DC converters That's it for now. Maybe this is useful to share. At this point I'm not sure yet, if this will lead anywhere. When I find another timeslot, I want to compare a not-yet-broken board with one that exhibits the problem. Regards, Harald -- - Harald Weltehttp://laforge.gnumonks.org/ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6) ___ Soekris-tech mailing list Soekris-tech@lists.soekris.com http://lists.soekris.com/mailman/listinfo/soekris-tech