Investigating a full linear PSU for the XXHE PC and NOS1

[acg]

How to Boot the Asrock X79 Extreme4-M motherboard without an ATX Power Supply

Yep...got it figured out...I can start the Asrock mobo using a three rail linear supply with ZERO help from an ATX supply.

This afternoons efforts have been focused on getting the mobo to boot reliably.  Based on the previous measurements, my 3 rail lab linear supply is not able to provide enough current (from the fixed 5v 3A rail) for the 5v rail of the P24 ATX socket, so I did a little trick to make the BeQuiet SMPS turn on when disconnected from the motherboard thus bypassing its ATX functions.  The BeQuiet SMPS was used to provide power to the P8 CPU socket and the 5v rail for the P24 ATX socket.  The 3 rail linear PSU was used to provide 12v and 3.3v to the P24 ATX socket and 5v to the PPAstudio usb card.

I improvised a wiring loom for the P24 ATX socket that had the three rails (from two power supplies) and grounds distributed as shown in the attached image.  Note that at this stage the grounds don't seem to be too important because everything has worked with my first semi random attempt at spreading the grounds between the three rails.

To get the computer to boot I had to first start the BeQuiet SMPS (12v in P8 cpu, 5v into P24 ATX) and then start the lab linear supply (12v and 3.3v into the P24 ATX socket).  It does not matter how long I wait to turn on the second psu, (I tried from 0.5s to 30s), because the computer fires up every time when I press the power button on the front of the case.  The computer will NOT boot when the BeQuiet SMPS (5v P24 ATX and 12v P8 cpu sockets) is turned on AFTER the 12v and 3.3v rails into the P24 ATX socket.  I am not sure whether it is the 5v rail or the P8 CPU rail or both that must be turned on before the remaining rails into the P24 ATX socket, but that can be determined at a later date.  The important thing is that I can reliably turn the computer on and off with a rudimentary ATX sequencing.

There is one caveat at this stage...for those familiar with the operation of the light on the front power button on the  Silverstone GD06B case...well at this stage I have not spent the time to get that working properly.  It comes on as soon as the 5v rail into the P24 socket is powered regardless of whether the computer is on or not.  The remedy may be as simple as doing something different with the 5v 'Power On' wire into the P24 ATX socket, but I have not spent the time to figure it out as yet.  At this stage the 5v 'Power On' wire is grouped with all the other 5v wires on the 5v rail.

The sequence of events to get the Asrock X79 Extreme4-M motherboard to boot from a non-ATX power supply is as follows:

• Power on the 5v rail into the P24 Socket and the 12v rail into the P8 CPU socket (not sure if both are required to be started first at this stage)
• Following a small delay, power on the 3.3v and 12v rails into the P24 CPU socket as well as other rails to SSD's and USB cards et cetera
• Press the start button on the front of the computer case

Once the computer is successfully shutdown (either by Windows or the power button on the computer case), the order that the rails are turned off does not seem to matter at all for the next boot.  There will be some issue making sure that the computer is fully shutdown before cutting power, so at this stage I think that it will have to be a manual procedure i.e. when windows goes blank after shutdown then turn off the power supply.

Anyway, this took me about 3 hours this afternoon to figure this all out and I am feeling a tad 'special' because I thought it may have taken several weeks on and off, but now we have a way forward for a bespoke FULL linear power supply for the XXHE pc.

Cheers,

Anthony
 

[PeterSt]

Wow again Anthony !

One remark :
At least with Windows 8 (7 I don't know) it has occurred to me that the system can remain active for a second or 10 after all other tells you Windows has shut down (monitor was shut off by the system as well).
Of course this is easy to see by watching the leds inside (but led at the power button shows it just the same). And I must add that I am not so sure I saw this always (read : maybe BIOS/XXHE settings can imply this).

Of course I can't tell what is happening during that stage (because Windows "tells" all has been shut down anyway) but in your (due) case it could be difficult to see when the power really can be cut. I mean, if Windows has not shut down at all really, you will be cutting whatever last process.

The least I theoretically could do for you is watch the disk light suring that last stage, but I am reluctant to do so reliably because as you will know this is not easy to see through the same space and same led color of the power button. So, I already dropped my ever and always looking at the disk activity for this PC (which I always regard the most important) because it can't be seen anywhere when flashes are very short.

Hope this helps ...
Peter

[acg]

Thanks Peter,

The power down thing may be a more difficult issue than the power on difficulties.  To do it right some transistor logic will be required using the signal from the motherboard that is transmitted to the power supply via the 'Power On' wire in the P24 ATX socket.  Providing this logic is far beyond my current knowledge so I would have to defer that to others.

A simpler way may be to wait ten seconds or more before turning off the linear power supply or even using a short piece of optic cable to show the status of the mobo lights on the outside of the case so that we can be certain that the mobo has powered down before we cut power to the power supply.

Anyway, I think this is a hurdle for later on down the road.  First we have to figure if the LPSU into the XXHE will make a sonic difference.

Peter, have you got any ideas about good linear regulators to use when building a power supply?  The obvious ones to me are the Paul Hynes regs but they are very expensive and a very long wait time.

Cheers,

Anthony

[acg]

In the sharing spirit, with Paul (Scroobius) having just purchased a PPAstdio USB3 card like mine, I decided to try the lab linear psu feeding said card to see if it makes a difference.

Well it does make a nice difference.  The NOS1 has already outperformed some really good dacs here at home, but it is even better with the PPAstudio card and even a cheapo lab linear psu.  Blacker backgrounds, better 3D imaging, I can hear more of the room in acoustic recordings, or the 'ambience' as Peter likes to put it...the reverb, the echo et cetera.

So mark that up as one rail that is essential to be powered by the linear psu.

Anthony

[PeterSt]

Peter, have you got any ideas about good linear regulators to use when building a power supply?  The obvious ones to me are the Paul Hynes regs but they are very expensive and a very long wait time.

Anthony,

No, not really (at all). But this is merely because I never looked into ready products. Okay, not true : http://www.tentlabs.com/Components/Shuntcomp/index.html
I have some 300 of these around which is for a reason.
Just doesn't work. Well, not through my eyes.

Coincidentally I very recently designed such a thing in module form and the PCBs for it should have arrived this week but nothing yet (stolen in China ). This should show noise in the nV range and it should be able to replace the input caps on the NOS1 boards (+/- 4.9V). But this is very low current (say 300mA max) and I don't think it can be useful for in the PC anywhere.
To make something general is not easy because it depends on the environment and the knowledge of that. So for example, when it is to take 3A then the configuration (of chips used) is different and e.g. 1mV of noise comes from it (just making up some number). However, while this 3A regulator might be applicable throughout (assuming 3A is max anywhere), it is oversized and also "over-noised" for the situation that only 300mA is needed which could show that nV range of noise.

The funny thing is that when you approach it from this angle of "better regulators" you find yourself the PSU to be not important anymore, as long as the power supply rejection (PSRR) of the regulator is very ok. So all is one big mixed bag of design issues (or challenges) and when you'd believe in this approach you can even think of one 12V out of which all 12, 5, 3.3 and beyond voltages are created. Now that one 12V needs to carry all the Amperes together so probably it will be a no-go for that reason (and additional noise), but the whole point is that you only know for real when you sit down for a week at finding chips to do the work needed and with the notice that only after this "research" you will have a consistent design that may show it failed (against the other theoretical solution, also to work out).
May take weeks or months to do it well.

Peter

[acg]

Thanks Peter,

Now that one 12V needs to carry all the Amperes together so probably it will be a no-go for that reason (and additional noise), but the whole point is that you only know for real when you sit down for a week at finding chips to do the work needed and with the notice that only after this "research" you will have a consistent design that may show it failed (against the other theoretical solution, also to work out).
May take weeks or months to do it well.

Wow...righto...I am quite sure that I am not up to the task of designing a linear regulator, I had just assumed that I would be using an existing design and assembling them (at worst) or using an off-the-shelf linear regulator (at best).

For instance, this reg from your part of the world that I could pull out a soldering iron and follow a relatively simple build.  Noise seems fine but the current limit is 1A, so it could only be used to power a USB card or SSD.  I reckon I could build these for about AU$75/50euro each.

5A and 2A loads are going to be much more difficult.

Thanks,

Anthony

[acg]

No, not really (at all). But this is merely because I never looked into ready products. Okay, not true : http://www.tentlabs.com/Components/Shuntcomp/index.html
I have some 300 of these around which is for a reason.
Just doesn't work. Well, not through my eyes.

I have had  look at the tentlabs shunts that you linked.  Why do you think this regulator has not worked for your situation?  Noise and output impedance both seem low.  Perhaps the not-published PSSR and transient response figures were too far out of spec.?  I'm just guessing here, trying to figure out what I should be looking for in a voltage regulator.

Coincidentally I very recently designed such a thing in module form and the PCBs for it should have arrived this week but nothing yet (stolen in China ). This should show noise in the nV range and it should be able to replace the input caps on the NOS1 boards (+/- 4.9V). But this is very low current (say 300mA max) and I don't think it can be useful for in the PC anywhere.

Actually, the builder of the PPAstudio USB card recommends a 5V 1A power supply but I can't get my card to draw more than 0.02A into the NOS1, which is 20mA, and is well within 300mA range of the regulator you mentioned above.

I have found an article from Linear Audio Magazine that first measures 15 commercial and diy linear regulators and then does blind listening tests when using them to power a line amplifier.  Although the listening tests are relatively meaningless in the context of the XXHE PC project, the measurements of these regulators could be of some use.  The article itself is not free but the graphs are readily available on their website, so I will reproduce some of them below.  Note the Sjostrom measures well in all three graphs below.  It is interesting to note that one of the 'surprisingly good' regs in the listening tests were the Linear Tech devices which the author of the paper attributed to 'fast transient response'.

Anthony

[acg]

TOTAL DRAW OF XXHE PC

I just realised that I had not published the complete draw of my XXHE PC in this thread.  To do this I plugged my pc into the UPS and let the UPS tell me the draw as the pc performed various tasks.

Whole XXHE PC	Minimised o/s	Maximised o/s
Startup	98w	98w
System Idle	58w	68w
Playing Music	68w at start 58w stable	88w at start 58w-78w in a cycle
Shutdown	78w	107w

This information is useful if I end up going a different route with the linear power supply and using a wide range pico psu (eg. 12v - 24v 160w) fed by a single high powered linear power supply.  The pico psu is a tiny dc-dc converter that manages the atx power functions including derivation of the various rails for the P24, P8 and peripherals.  A single 20V 8A supply would be sufficient to run the 160w wide input pico psu.

As you can see from the table above, a 160w wide input pico psu MAY have enough juice to run my XXHE PC, but does it have enough juice to run yours?  This is especially relevant if you run more HDD's or SSD's than me, and use the higher power i7 cpu that is recommended for the XXHE PC.

Anyone care to measure the total draw of their XXHE PC?  I may be able to rule out the pico psu angle if your computers draw much more juice than mine.

Anthony

[acg]

PERIPHERAL DRAW

I thought that I would also share the draw of the two peripherals inside my XXHE PC: a single Samsung 840 EVO 750GB SSD; a PPAstudio V1 USB3 card.

Computer State	SSD	PPAstudio V1 USB Card
Computer Off	0.25A	0.02A
Startup	0.44A	0.02A
Playing Music	0.40A to start 0.05A stable	0.01A
Shutdown	0.5A	0.01A

The results were identical when the o/s was minimised and maximised.

The SSD is spec'd at 5V 1.4A (written on drive itself) but I never saw more than 0.5A draw at any stage, although it would almost certainly be prudent to be able to supply the recommended 1.4A to the SSD just in case...we would not want it to lose information if it could not find enough juice.

The builder of the PPAstudio USB card recommends 5V 1A.  The NOS1 does not seem to take usb power at any stage, so perhaps we can get away with powering the PPSstudio card with a higher quality and lower current 5V linear supply.  The USB3 standard gives a maximum load of 0.9mA, and considering that I have been able to measure not more than 20mA draw, perhaps a linear power supply around that 100mA (0.1A) will be perfectly suited if only the NOS1 is connected to the card.

Anthony

[AlainGr]

The PPA would be a good candidate for the iFi iUSB I suppose. I intend to have a cable made with a USB male A at one end and a molex at the other end with the required 5V...

Alain

[PeterSt]

Ok, I hooked up a usage meter. I will try my best to make it interpretable but we will soon see that this is a though job.

I used Minimized OS only (at least for now).

Whole XXHE PC	Minimised OS Anthony	Minimised OS PeterSt
Startup	98W	92W
System Idle	58W	52.5W
Playing Music	68w at start 58w stable	54.5W - 57W

Notice that it is crucial to compare the XTweaks settings; mine you see below. Even more crucial is that by this means I finally found that the Cool when Idle setting makes all the difference in the world. Look here : Re: XTweaks : Cool when Idle and Windows 8.

General remarks :

Mains is ~222V today.

Looking at thee table above, my figures show consistently lower (all sort of the same percentage lower). My conclusion from that is that Anthony can have his processor frequency at 1200MHz while mine runs at 720MHz. And yes, since this board his here anywhere I will create a topic about how to do that.

The 2648L Xeon processor Anthony uses consumes (well, dissipates) 70W max. My 3930 is rated at 135W. Now careful :
Assumed the both processor types (completely different by themselves) use the same "die" technology, all what happens (or may happen to my expectation) is that the max usage occurs at the max demand. Here is already a difference to notice because the 3930 is 3.8GHz in turbo mode while the 2648L is 2.2GHz in that mode. The 3930 is 6 core while the 2648L is 9 core, but turbo mode happens in one core only. Otherwise the 3930 is 3.2GHz and the 2648L is 1.8GHz.

Still there ?
Nothing compares, also knowing that turbo mode won't be used at my XTweaks settings, plus my 3930 won't exceed the 720Mhz by a singly cycle ever. So ...

What I try to say is that probably there is no difference as long as the both processors use the same frequency *and* would have the same number of cores - which they have not.

Music Playback

If it were for me, this never can be stable and which is also what I see happening. To see this well the SFS must be somewhat higher (I set mine to 120 for this) pluf of course the meter must respond fast enough. Anyway, what I see is that loading a trackpart consumes 2.5W.
I tried to see a difference between loading from RAMDisk and loading from the spinning disk, but there is none that I can see.
I also tried vary the Clock Resolution and Q1, but no difference to be seen.

Peripherals

My test was with one 2.5" spinning disk in there (in the removable bay). So given figures are for that. Otherwise the only device connected was the NOS1. So, 52.5W idle.

Connecting a USB3 disk (self powered !) implied an additional usage of 1W (when settled). Otherwise maybe 0.5-1W more (right after plugging it in).

Inserting a 3.5" spinning disk implied a steady 6.5W more, which is quite a lot compared to the 52.5W without it.

Sadly I could not test a 2.5" disk or SSD because the bay for it was occupied by the OS disk. But assumed the usage of that is similar then without disks something like 46.5W would be the result for the whole PC.

Might it be of importance : Idle figures are with LAN running, while Playback figures are with LAN shut off. It is a bit cumbersome to test this better, however, pulling the LAN cable makes a difference of between 0.5-1W.

Inserting a random USB2 stick in the USB2 port (after settling) doesn't seem to make a difference and otherwise it is under 0.5W (my meter has no better resolution than 0.5W). Inserting that in an USB3 port (all front side) I see 0.5-1W.
A USB3 stick same story.

Boot

Thinking this is all about what a LPSU should be able to cope with, I think this is crucial data :

Once the frequency ratio is taken out of the default in the BIOS, in "BIOS stage" (which is the same as during booting) no turbo frequency will occur, while otherwise I'm fairly sure it does. I did not test this thoroughly (I mean, ever back), but we know that when we set up the PC for the first time all the fans should run at max, until we changed that ratio (to any value, also higher than 3.8 ) and when then a reboot is applied all is safe to look at the BIOS for a longer time.
Of course such behaviour should be visible with the usage meter just the same.

I think it will be so that when for example the ratio is set to 1.2 (less is not possible) the usage will be relative to that when in BIOS mode hence when the PC boots. In my situation (see table) this implies 92W max, but which 92W did not really occur during "POST" but much later.

Vague will be that while this 92W occurs later and will not come from "raw" motherboard behaviour, it will come from the more heavy OS activities BUT which in my case are already done under the 720MHz frequency. Now, to me it seems not important how much the raw usage is during "POST" because this OS usage is just the max occurring. In that situation all cpu cores will be active and thus the most draw is implied at that stage. I just started a FLAC album which also uses all cores (hyperthreaded) and during a short stage the usage is 68W. Notice though that this is I/O intensive and no way a core will be at 100% because of this. This story is more complicated because of (cache) memory to RAMDisk operations here, but the key point is that this 68W never exceeds the 92W during booting.

It seems that 92W is the maximum usage and this is only during a short while (but never mind the short while because nothing should be dangerous.

Current draw

Just because the question may be asked :
I am not sure what I see is reliable; it is so much lower than what Anthony's tables show that I must think about it. Notice it is measured through the same meter. Still what I see is consistent with the Watts used. But for example : 0.265 - 0.280A during Playback which nicely relates to the 52.5-57W I see. The 92W max used compares to 0.44A max used (boot).
So, it may be consistent, but I don't think this can be correct.

Peter

[PeterSt]

Since I had to reboot anyway I watched the time before the power led went off after the monitor went blank and powered down :

A sheer 22 seconds.

[PeterSt]

Yea, I see now where I didn't understand my amperage draw;

I was so smart to mention my mains voltage because this is related when we need the amperage, but ... that is of course not the draw from 12V etc. internally ...

Peter

[acg]

Nothing compares, also knowing that turbo mode won't be used at my XTweaks settings, plus my 3930 won't exceed the 720Mhz by a singly cycle ever. So ...

What I try to say is that probably there is no difference as long as the both processors use the same frequency *and* would have the same number of cores - which they have not.

My Xtweaks are almost identical to yours Peter (my Balanced Load is 45), and my cpu was set to 'Auto' which means it overclocks at will, which it does, so it goes up to about 2.2GHz and was dropping as low as 1.6GHz.  This was the default setting in the BIOS, so that was were the measurements were made.  I have performed no BIOS tweaks in relation to the RAM or CPU.

Current draw

Just because the question may be asked :
I am not sure what I see is reliable; it is so much lower than what Anthony's tables show that I must think about it. Notice it is measured through the same meter. Still what I see is consistent with the Watts used. But for example : 0.265 - 0.280A during Playback which nicely relates to the 52.5-57W I see. The 92W max used compares to 0.44A max used (boot).
So, it may be consistent, but I don't think this can be correct.

I think our amperages are different during playback for a number of reasons, the most important of which is that the amperages that I have measured to date have been inside the computer at DC, while yours is 222V AC before the computer.  Your AC wattages will be much more accurate than mine because I have plugged into a UPS that updates the screen at about 1Hz and seems to have a resolution of about 10 watts (i.e. I see 58w or 68w but nothing in between).  Also, as mentioned earlier, although my Xtweaks are almost the same as yours Peter, until now my CPU ratio in the BIOS has been set to 'auto' which means that it under and overclocks at its own discretion, so my clock frequency varies with load.  Of course I will change this at a later date.  Another thing to contemplate are RAM timings...have you changed yours Peter?

Thanks so much for doing this Peter, it certainly does look like we can go down the pico psu + single linear supply route for the mobo if we need to.

Anthony

[acg]

Since I had to reboot anyway I watched the time before the power led went off after the monitor went blank and powered down :

A sheer 22 seconds.

I also had an unusually long shutdown last night...a solid 20-30 seconds after the light went off.