Saturday, 14 May 2016

A lesson for us all - why bigger is not always better (or diagnosing a 'non-booting' system)!

If you read my blog regularly, you will know that 2 years ago I built my own system. That previous blog post described how I went about it.

Now over the last few months, I have noticed that this Z87 PC did not always boot when I switched it on. In fact, although the fans whirred a bit, I did not get the familiar 'beep' from the mainboard on these occasions. When this happened, I would just turn it off at the PSU mains rocker switch, wait a minute or so and then turn it on again and it would usually work. That is, until this afternoon,,,

This afternoon, I shut down the PC in order to do a clean boot from a USB HDD (as I have done literally hundreds of times before) and I got the 'no display, no beep' symptom yet again. But this time there was no persuading my Haswell Z87 PC to boot!

Diagnosing a 'dead' system

Now the Asus Z87 mainboard is fitted with a number of diagnostic LEDs soldered to the PCB. These include:



Green Power LED + 4 POST (Power On Self Test) state LEDs comprising of...
  • CPU  LED
  • DRAM LED
  • VGA OK LED
  • Boot Device LED
The POST LEDs were supposed to light up in sequence - i.e.  CPU LED on - BIOS tests CPU - CPU LED off - DRAM LED on - BIOS tests RAM - DRAM LED off - VGA LED on - BIOS tests VGA - VGA LED off - Boot Device LED on - BIOS tried to find bootable device - BIOS LED off - system boots from boot device.

The Green Power LED was lit, but there was no activity from the POST LEDs at all.

This indicated to me that the CPU was not even starting up.

To check this, I removed the two DDR3 DIMM modules - I got no 'beep' - nada! A sign that the CPU was not running.

Next I removed the CPU heatsink and switched it on again. The CPU remained cool (it should have got hot very quickly!).

So either it was a PSU problem, or the mainboard+CPU was broken.

Next, I removed all connections from the mainboard except for the two power leads from the PSU, in case anything else was shorting out or overloading the PSU - still no joy!

I decided to remove the mainboard+CPU+RAM combination and then I plugged it into another PC PSU that was in my old system. This time the Z87 board sprang into life, I got the POST state diagnostic LEDs coming on in sequence finally the Boot Device LED then stayed on.

So something was wrong with the PSU in my new PC!

I tested the voltages using a DVM and they looked OK, but I wasn't sure the 'power good' signal was working. This tells the board it's OK to power up.

Then I remembered that when I was choosing a PSU, I was concerned about the minimum load specification of the Corsair 500W CX Builder 80 Plus Bronze PSU that I had picked, but they did not actually specify a minimum load (when I worked for an OEM system builder, I always checked this with the PSU manufacturer and then measured the power draw from a 'minimum-spec' system to make sure it would work, as well as build several 'minimum-spec' and 'maximum-spec' systems to test).

PSU manufacturers should always state the minimum load requirements of a PSU. Depending on the design, the PC may have to meet a minimum load requirement on the 12V rail and/or 5V rail (and sometimes there may be a dual-rail supply with a minimum load requirement on each rail). It seems the CX500 has a 12V and 5V min. load requirement and the 12V min. load was not being met! Later PSUs derive the 5V supply from the 12V supply, so any load on either the 12V or 5V or 3.3V rail will suffice on this type of PSU, as long as it meets the min. requirement (which Corsair don't publish!).

To meet the 80% efficiency rating, these modern PSUs are designed to have a 'decent' load on them before they will regulate properly. So opting for a 80+ rated PSU is not always a good idea for a low-power system.

I had intended to add a decent sound card to my PC (once you have heard an Asus Xonar sound card, you will be amazed at how good a PC can sound and how crappy the on-board PC audio really is!). I also intended to buy a good graphics card, but actually found the Haswell CPU/GPU perfectly adequate for my needs (I don't do gaming), and so never bothered to buy and fit a VGA card.

Then I remembered that I had recently replaced one of the hard drives with a new SSD drive. What if there was not enough load on the PSU power rails now?

To check this, I connected a couple of very old IDE hard drives to some spare power connectors, then connected the Z87+RAM+CPU combo to the Corsair PSU and Voila! it sprang into life! Just to prove it, when I disconnected both drives, the original problem returned.

So, I reassembled the PC, robbed some drives out of my old PC that I was thinking about retiring to the attic anyway, and I am back in business. A lot effort for something that was actually a very simple fix - just plug in another HDD or two (I don't even need to connect the data cables!).

I found I had to fit two extra HDDs, one modern SATA drive (which did not draw enough power to completely fix the problem) and one very old (and noisy!) IDE HDD (but I could not connect the data cable, because I had no IDE interface on the Z87 board).

It just goes to show that you should never just fit a big PSU, unless the components are actually going to meet the minimum load requirements of the PSU. If there is insufficient load on the PSU, the PSU cannot regulate the voltage properly and so does not supply a 'power good' signal to the mainboard. Corsair do have an article about Haswell CPU boards and low-power states on their website.

It doesn't help that modern hard drives limit the current they take on start-up. This allows you to connect lots of HDDs to one system without getting a terrific power surge on switch-on. So the extra load added by modern hard drives is actually quite small, making it quite hard to find something that will draw enough power. Hmmm..., I may just go and fit that Xonar sound card now... (which also has no power specs that I can find!).

P.S. I got fed up with the noisy IDE HDD, so I soldered two 22 Ohm resistors in series between the yellow 12V lead and the black Gnd lead of a spare Molex power connector and suspended it on the back of my system (outside the case and in front of the rear fan). Now I can disconnect the noisy IDE HDD. I used two 22R resistors because if 44R did not draw enough current, I could reduce it to one 22R resistor, but 44R seems to work fine (12/44=270mA=3.24W). I will screw them down to the back of the case later, but  I am hoping the sound card I ordered will take enough juice so that I won't need them.
P.P.S. My system failed to boot again this afternoon even with the Xonar DG sound card added (after it had got warm by being on all morning), so I removed one of the 22R resistors (the 22R remaining resistor gets too hot to touch now!). Time will tell if this has now fixed it!
P.P.P.S.  2016-06-12 - so far so good! Later I decided to replace the PSU to avoid burning myself on the resistors (after an unfortunate incident followed by much swearing).