Bitcoin Forum
December 11, 2016, 08:11:48 AM *
News: Latest stable version of Bitcoin Core: 0.13.1  [Torrent].
 
   Home   Help Search Donate Login Register  
Pages: [1]
  Print  
Author Topic: PC ATX PSUs for FPGA clusters? 5V / 3.3V Rail Issues?  (Read 3346 times)
catfish
Sr. Member
****
Offline Offline

Activity: 270


teh giant catfesh


View Profile
April 13, 2012, 09:03:53 PM
 #1

Quick question. I've decided that the 'cleanliness' of the 12V DC supplied by the decent quality PSUs invested in for my GPU mining farm would be better than using 5 or 6 'black-box' cheap 'router-type' power supplies. Since Phase 3 of my farm is a 25-unit FPGA array, the 600W Cooler Master modular PSUs I have are more than adequate.

Since they're modular, and I have loads of adaptors, I'm planning to use the 12V DC and ground pairs in the PCIe 6-pin power outputs to feed 5 FPGA boards. The power requirement for 5 boards is less than 60W, so with 18AWG cable throughout, I feel safe enough doing this.

However, since the fan system of my proposed design also taps the 12V DC rail, I'll be running the PSU with zero load on the 3.3V and 5V rails - and all the other load on the 12V rail.

Since I'm not introducing the wasted power, space and setup costs of a traditional computer (i.e. logic board, CPU, etc.) into the modular rig design (instead, the entire array will be looked after by a low-power compact Dell netbook, running Mac OS X Snow Leopard - it has its own PSU and is easily replaceable with another netbook / Mac laptop if it fails), I'll use the normal trick of wiring a dummy plug into the 24-pin ATX connector and a user-accessible switch from green to ground.

This should work, but I've not only read a fair number of threads about how PC PSUs don't like zero load on the 5V / 3.3V rails and all on the 12V rail, but I've experienced it myself when using two PSUs to power 5 GPUs on a shonky P8Z68-V board. It was only reliable if I chucked a hard drive on the slave PSU's SATA connector...


Is this a real problem or just a problem with low-quality PSUs? I'm planning to use Cooler Master 600W modular units that have performed like champs for months - even with my insane right-at-the-limit GPU clocks.

I'm not keen on the idea of connecting an unused mechanical hard drive to the PSU in order to put some load on 5V (still, where's the load on 3.3V? Does it matter any more?) - it's wasting power, and insane UK electricity costs are the primary reason I'm switching to FPGAs, then selling ALL my GPU gear and buying more FPGAs.

Does anyone know for sure whether (a) it'll be stable with only 12V rail loading, and (b) it won't be working at the expense of efficiency (i.e. if the Cooler Master *can* operate with no load on 5V and 3.3V, but it's an unexpected case and the lower voltage circuits waste power by heating up)??


I'm not interested in suggestions involving ditching the HacBook for a bare-bones Atom PC running Windows - under ANY circumstances, and don't want to waste more time and money buying bare-bones Atom logic board / components and then struggling to get OS X running on it. The only alternative 'controller' option I'd consider is something like a Raspberry Pi - but these are under heavy demand and I want to get up and running as soon as the FPGAs are delivered.

So PSU-only advice would be most welcome. Are there any easy 'tiny-but-enough-to get-normal-efficiency' tricks to load the unused ATX rails in my design - or is my concern unfounded and only a problem if I use a cheap-ass PSU (which, with £5k worth of hardware being powered from it, I won't be)...

...so I give in to the rhythm, the click click clack
I'm too wasted to fight back...


BTC: 1A7HvdGGDie3P5nDpiskG8JxXT33Yu6Gct
There are several different types of Bitcoin clients. Server-assisted clients like blockchain.info rely on centralized servers to do their network verification for them. Although the server can't steal the client's bitcoins directly, it can easily execute double-spending-style attacks against the client.
Advertised sites are not endorsed by the Bitcoin Forum. They may be unsafe, untrustworthy, or illegal in your jurisdiction. Advertise here.
1481443908
Hero Member
*
Offline Offline

Posts: 1481443908

View Profile Personal Message (Offline)

Ignore
1481443908
Reply with quote  #2

1481443908
Report to moderator
1481443908
Hero Member
*
Offline Offline

Posts: 1481443908

View Profile Personal Message (Offline)

Ignore
1481443908
Reply with quote  #2

1481443908
Report to moderator
1481443908
Hero Member
*
Offline Offline

Posts: 1481443908

View Profile Personal Message (Offline)

Ignore
1481443908
Reply with quote  #2

1481443908
Report to moderator
P_Shep
Legendary
*
Offline Offline

Activity: 924


View Profile WWW
April 13, 2012, 09:07:55 PM
 #2

I suspect the worry is a thing of the past. I'm using just the 12V, and some small load (usb hub) on the 5v and everything is fine.
DeathAndTaxes
Donator
Legendary
*
Offline Offline

Activity: 1218


Gerald Davis


View Profile
April 13, 2012, 09:33:44 PM
 #3

Any high quality modern PSU is a rail to rail design (DC to DC).

All of the AC voltage is converted to 12VDC.  Then if needed some of the 12VDC output is converted to 3.3V and/or 5V.

Like p_Shep said it is a thing of the past.  The one exception might be some "non 80 Plus" generic POS power supply.  Anything from a solid brand should be fine.

As far as loading only "one" 12V rail most (all?) DC to DC PSU only have 1 REAL rail.  They then use current limiters to split that into a bunch of virtual rails (so someone doesn't try and pull 102A+ through a single 12V wire).

Loading only 1 "rail" shouldn't be a problem.

The main thing is to ensure good efficiency (and poor efficiency is hard on the PSU switching circuity) you want the load to be at least 20% (and 35% to 50% would be better) of peak.  So don't go buying a 1200W PSU to power 2 FPGAs. Smiley

Lastly while 80-Plus Gold may not be a huge need for FPGA (given their already high electrical efficiency) the higher end units tend to be built better with "beefy" Japanese solid state caps, heavy busbars for power distribution, and higher end fans.  Then tend to have 5-7 year warranties so the manufacturer is really designing them to perform well for very long time (or their profits get eaten up by RMA costs).

IF it were me to power $2000+ in FPGA I am looking at a short list with SeaSonic at the top.
MrTeal
Legendary
*
Offline Offline

Activity: 1246


View Profile
April 13, 2012, 09:49:52 PM
 #4

Which PSU is it?

If you do have an older group regulated one, you could always just toss a couple 5V and 3v3 fans on there to keep the rails happy.
DeathAndTaxes
Donator
Legendary
*
Offline Offline

Activity: 1218


Gerald Davis


View Profile
April 13, 2012, 09:54:59 PM
 #5

Which PSU is it?

If you do have an older group regulated one, you could always just toss a couple 5V and 3v3 fans on there to keep the rails happy.

This.  If you do have an older group regulated PSU I would simply replace it with one of the well built modern DC to DC designs.  It is a non-brainer given the cost of the hardware you are looking to power.

If you need to provide a load on 3.3V & 5V rail I would go with a fan.  Hard drive's load isn't going to be consistent like a fan is. 
marked
Full Member
***
Offline Offline

Activity: 168



View Profile
April 14, 2012, 09:19:57 AM
 #6

This.  If you do have an older group regulated PSU I would simply replace it with one of the well built modern DC to DC designs.  It is a non-brainer given the cost of the hardware you are looking to power.

How modern is modern?

marked
CA Coins
Donator
Sr. Member
*
Offline Offline

Activity: 306


View Profile
April 14, 2012, 09:40:35 AM
 #7

Hey Catfish, I think you're looking at running 25 1.15x if I am not mistaken.  I have been running 60 1.15x off just the 12V (4x PCIE cables) using a Corsair AX1200 for the last month or so with no problems.  No load on the 5V or 3.3V.  Like others have mentioned, I think it should work just fine for you.
Jaryu
Member
**
Offline Offline

Activity: 91


View Profile
April 18, 2012, 03:08:22 AM
 #8

This.  If you do have an older group regulated PSU I would simply replace it with one of the well built modern DC to DC designs.  It is a non-brainer given the cost of the hardware you are looking to power.

How modern is modern?

marked

probably any 80 plus should be good enough for the modern requirement, gold or platinum if you want the best of the best of the best.

Hey Death question, since you're so familiar with power supplies, do most (modern) power supplies output have a diode to prevent current back flow and to allow you to link all the 12 v rails into a single busbar or even going further by ganging a couple of such power supplies without causing resonance and blowing either 1 or both?
Pages: [1]
  Print  
 
Jump to:  

Sponsored by , a Bitcoin-accepting VPN.
Powered by MySQL Powered by PHP Powered by SMF 1.1.19 | SMF © 2006-2009, Simple Machines Valid XHTML 1.0! Valid CSS!