SGI: Hardware

kubatyszko wrote: What I have already done with ATX is good enough and allows me to use it as a PSU replacement with env monitoring off.

It's a match made in Heaven - a mainboard that wants env monitoring off and a power supply that wants env monitoring off.

So I made it !!!

I was actually on the perfectly right track, but my math was wrong,
Also the frequency I measured from the NMB PSU must have been wrong.

The mistake I made:

FAN generates frequency of 2 clicks per rotation with 50% duty cycle.
What I did wrong, is that I calculated required frequency per RPM, and not rotation - which is a difference between say 30 RPS and 1800 RPM.

That led me to think I need frequencies in 4000-6000HZ range, while all I needed was anything above 60Hz (>60Hz is universal for all the Fans in Fuel).

And indeed 5KHz was totally right to give me 60-130000 RPM, because this is the correct value.
63Hz should give this much:

(63/2) * 60 # (number of rotations per second / 2 clicks per second) * (60 seconds in a minute)
= 1890 RPM

At roughly 63 Hz, I get 1753 RPM reading which makes it more of 58Hz, but that's fine as long it's above the 1600 threshold.

What led me to finding out my mistake - I bought a couple fan's (to optionally replace the ones in Fuel), and checked their signal from Tacho.

There are multiple ways to create square wave generator, I did mine in two: using NE555 (+ 2 resistors, 2 capacitors and zener diode to limit the voltage), and using 70HC14 (schmitt trigger inverter, + 1 resistor, 2 capacitors and zener diode).

NE555 gives me much more "square" wave but that doesn't matter much.
Parts:

NE555:
1 * NE555
1 * 1800 Ohm
1 * 470 KOhm
1 * 0.022 uF
1 * 0.01 uF
This yields 69.6 Hz at 50.095% duty cycle

(Schematic here: http://www.royalrife.com/555_calculator.html )


Inverter:
1 * 74HC14 (I used Texas Instruments, another one from Hitachi HD7414P didn't want to work for me).
1 * 470 KOhm
1 * 0.01uF
1 * 0.047 uF
1 * 3.9V Zener diode

(Schematic here: http://www.national.com/assets/en/board ... rguide.pdf - this is actually schematic of the whole evaluation board of LN81 - which has the same function as DS1780)

The Inverter-based solution has an advantage (5 of them in fact), with 1 chip you can generate 6 independent square waves - just in case if somebody wanted to fool the Fuel in regards to other fan's , another minor advantage of inverter-based is that you don't need to worry about duty cycle, it's always 50%.

The generators are powered from 5V (NOT 12V), I used 5V from pin 17.

I DID NOT bother using FANC pin - this is NC on my adapters, maybe in the future I could make the oscillator adjust frequency to the signal from FANC - but that's not necessary for anything - would make sense only if Fuel actually verified feedback while controlling Fan speed...

Also, There is NO FAN MONITORING in the ATX PSU AT ALL - if it fries, you're on your own, (but you have working adapter and can simply replace the PSU with another one ).

While playing, I fried 2 fan's - turns out that tacho logic is very sensitive to sloppy people who forget to swap the pins around - but I had 3 more which worked well.

I bought:

2* Enermax UCTB8 1600 RPM (80mm)
2* Omega Typhoon CFZ-60F 3000 RPM (60mm)
1 * some kind of used noname (80mm)

I actually did replace XIO fans with Omega's but had to revert the process, since those spinned at some 1450 RPM's which was below the threshold (I want to achieve perfect result with no env off/on tricks).
Enermax fan didn't turn out good either, as it spins at 1100 !!! which is whole 20 RPM's below the threshold

I'll keep looking for more suitable fans.

Now a very interesting fact, the reason I did manage to get 2000+ RPM's at over 20KHz frequency - (this should not be possible - as 20Khz would mean some 600k RPM's),
The DS1780 uses internal 22.5KHz clock to drive the counter - somehow I must have managed to get close to sync with that clock and it started to think the periodical going out-of-sync was the real signal .
That means the maximum theoretical fan speed that DS1780 could measure would be < 600000 RPM's - useless for Jet engines
Interesting huh ?

---

Here is a "screenshot"



And below are the pictures.

Awesome!

Sweet I love hardware hacks like this. Now all we need is an adapter PCB with the connectors and oscillator circuits

There's no point making PCB for swapping 24 wires + circuit with whole 4 components.
PCB will also make for bad cooling and it will be hard to place it inside of the chassis.

kubatyszko wrote: At roughly 63 Hz, I get 1753 RPM reading which makes it more of 58Hz, but that's fine as long it's above the 1600 threshold.

This is very cool and I imagine a bunch of Fuel owners owe you a big thanks !

But I have to admit I'm a little disappointed in the lowly 1600 rpm value. 300,000 was a lot more impressive

The Inverter-based solution has an advantage (5 of them in fact), with 1 chip you can generate 6 independent square waves - just in case if somebody wanted to fool the Fuel in regards to other fan's , another minor advantage of inverter-based is that you don't need to worry about duty cycle, it's always 50%.

This might be a good trick for using alternate low-noise fans in other boxes, like the O300 or O350. Of course, one would lose the environment monitoring .... do you feel like another project ? How about a ten percent signal amplifier ? That would give us several other options for fans that are just a tiny bit too slow, especially at startup ...

Good idea, challenge accepted
I could employ the same device in my O200 - making this beast quiet would greatly reduce earthquakes in the region.
But since this is nowhere high on my list it will take some time...
Cheers

kubatyszko wrote: Good idea, challenge accepted
[...]
But since this is nowhere high on my list it will take some time...

Heh - we've waited this long, waiting a year or two longer isn't that big a deal.

This is awesome news!
So in short, my original idea of a converter cable with some additional logic to fool the environmental monitoring wasn't that far off after all?

How generic can the PSU be? What are the ratings?

Technically any ATX PSU will do, but you need certain levels of power to make the Fuel run:

RockPower 550W PSU, ATX2.2, model LW-6550HG (used for ~11$), and another AQTIS 600W AP-600GR (about the same price)
I decided to use the RP - since its power levels are much closer to my Fuel's NMB (lacks a little on the 3V3 and 5Vsb).
(NMB supplies: 27A on 5V, 16A on 12VIO, 18A on 12Vdig, 45A on 3.3V, 3A on 5Vsb, and 0.6A on -12V)

I've never seen ordinary ATX PSU that would satisfy all those levels, so try to get as close as possible.
It of course needs at least 4-pin P4 connector (the 6-pin for PCI-X might also do but I haven't tried).
And don't be deceived by the overall power - my 600W PSU has lower levels than 550W PSU (the former has more on 12V rail though).

If your PSU has tacho signal on the internal fan you can also try feeding it directly into Fuel.

Probably not close enough to ramq to make shipping practical, but if anyone is looking for an ATX PS to mod for use with a Fuel, I still have the generic Sparkle ATX2.2 FSP460-460PFN power supply mentioned earlier in this thread. The generic ATX 2.2 Sparkle carries the same Sparkle part number (FSP460-460PFN) and 460w power rating as the Fuel WTX Sparkle, but would require kubatyszko's adaptation for use in a Fuel.

Good point, the sparkle is one of the very few that would match power levels perfectly.
I think we could try and assemble a list of tested PSU's that work well, bu there are so many of them it may knot make much sense...
I actually asked NMB (minebea) for specs, source and binary image of the PIC, so far all I got is that they are out of pc supply business but they will forward my request to the tech guys

Cool.
Well actually what one would need is a list of all the voltage rails demands just to have a list to compare with.
I know all those ricer gamerguys always keep en eye out on spec numbers when they're shopping for power supplies, so the manufacturers seem to be pretty good at specifying amps.

Demand varies, this is the trick, it will vary mostly depending on CPU - 500 vs. 900 MHz will affect the power reqs. pretty significantly, also, number of HDD's, PCI cards etc.

I have: 1TB SATA, 73GB SCSI disks, 4 PCI cards: Gbit, LSI U320, SATA controller and Audigy SB0090 in a 600Mhz, 2GB RAM box - and my box runs with no issues on the ATX I got (or actually ran some some short time I had it on, now I've installed the NMB back).
When adding more devices or upgrading CPU the setup may still work fine, but as soon as it gets close to capacity some voltage will likely drop - this is a good indicator when to look for better PSU .
So my advice for whoever tries ATX in Fuel would be to keep watching "env" on the L1 periodically.

Well, looking at whatever photos our fellow members has come up with, these might be useful:

A photo of the label/specifications of the NMB power supply used in the Fuel: download/file.php?id=2823084&mode=view
and the Sparkle power supply used in the Fuel: download/file.php?id=2822491&mode=view

Seems like the 430W PSU is alot beefier on the +3.3V rail.
FWIW, all Fuel PSUs I've seen so far has only been the 460W. Don't know what to judge by that, but there's two options:

A) Increased need for more juice on the +3.3V rail due to later design and customers beefing up their systems
B) Supply vs demand or simply "power effeciency"

Since we're probably never overload these systems I believe specs similar to the 460W PSU might suffice after all.
With that in mind I've come to this little table:

+3.3V 27.0A
+5V 29.0A
+12V I/O 15.0A
+12V CPU 16.5A
+5Vsb 2.0A
-12V 0.8A

So then:

Table for Sparkle (460W):

+3.3V 27.0A
+5V 29.0A
+12V I/O 15.0A
+12V CPU 16.5A
+5Vsb 2.0A
-12V 0.8A

Table for NMB (430W):

+3.3V 45.0A
+5V 27.0A
+12V I/O 16.0A
+12V CPU 18.0A
+5Vsb 3.0A
-12V 0.6A

Table for HIPRO (460W):

+3.3V 28.0A
+5V 30.0A (both 5+3V3 should not exceed 200Watt)
+12V I/O 18.0A (this is 12V1, not sure if this is CPU or IO)
+12V CPU 15.0A (this is 12V2, same as above, the PSU states that both 12V rails should not exceed 30A)
+5Vsb 2.0A
-12V 1.0A


If we were to take the lower bounds of all three, we'd get:

+3.3V 27.0A
+5V 27.0A
+12V I/O 15.0A
+12V CPU 15.0A
+5Vsb 2.0A
-12V 0.6A

ATX meeting those requirements should be fairly easy to find.

I've just installed my ATX in the Fuel and will be using it permanently, see how stable it is over long time,
I've also created cron job:

0,15,30,45 * * * * (date && l1cmd env) >> /var/adm/l1cmd_env

this saves env state every 15 minutes, should be perfect to have good overview of PSU stability.

Now, I only need to solve this annoying vibration issue in the PSU - I bought it used, and the FAN is horizontally mounted, probably it needs replacement (which will cost me more than another such PSU), it might also be related to the PSU being screwed only using 3 screws and not using that hanger hook like it's big brothers....

Appreciate all the hard work going on here. Only comment: If I'm going to buy a PSU rather than use one I have on-hand, I'd take the top specs from kubatyszko's three tables rather than the minimums, then look for at least a Plus 80 Bronze efficiency rating. Idea being to not have to do this over again any time soon, and not have to ever worry about exceeding the supply. I'd guess this would still come in under $100 for a brand new PSU from a reputable seller with a quiet product.

Valid point, we were just trying to establish minimum that should work okay, and finding the top-specs PSU will be not be very easy (unless you go for civilian models of Sparkle or NMB ), plus this is more for people who *need* a PSU ie. don't have one working etc - thus have no choice
The problem with today's top-spec PSU's is that most of them focus on the 12V rails for GFX rather than 5 or 3V3.

One more important thing for the public -- overall power rating is NOT a good indicator whether PSU works or not - my another 600W PSU barely matches the lower bounds, with the exception of 5V (24A), ALWAYS CHECK the detailed specs.

Understood and appreciated! I had figured I might wind up looking at 1kW PSUs to meet all the output requirements and just not putting much of a load on them, but your point about 12V versus 3/5V is well taken. I should look up some of the ATX server PSUs to see what that segment looks like...

Thread lives on!

Small PSA for those experimenting with ATX and Fuel, I just tried a 400W PSU with one of my adapters, and despite voltage level issues it works okay, the Fuel has a V10, 800MHz PIMM, 2G RAM and one IBM 18GB HDD - I'd say it's a rather good configuration, except V10.
The PSU has the following specs:

3v3 : 28A
5v : 40A
12v : 16A (only one rail)

According to our research from this thread, it's fine on the 5v but definitely lacking on the 12v and 3v3 rails - and that's clearly visible in L1 output...

It's clear the PSU has difficulty regulating the voltages, 12V at boot and then 3v3 during operation, but even then the machine runs fine.
I have not tried loading the machine up with any software and I won't (because I'm also missing that blue air-duct - using a regular household fan to improve the heat exchange ).

WARNING - this is just an experiment (because I used up my PSU's and this was the only I had left), and I highly advise against using underpowered PSU in *your* Fuel...