LMAO!

Well done sir.

+1

I got a little MicroATX Diablotek PSU based on good reviews on newegg - it works well for the low-power system it's on, except for the little annoyances like the fact that the power plug socket on the back decided to just pop out when I unplugged it one day.

Doesn't give me much confidence in their build quality when they can't even secure a major connector.

Hey quit derailing the thread!  I want more nerd porn!

Most of the 6950s I've found are non-reference coolers, there's only one or two on newegg.com with the standard turbine fan/closed box design.

Most of the 6970s on the other hand are ref. designs, with only a handful of customized coolers.

My XFX ZDNC 6950s (blow-everywhere style) ran annoyingly hot (often 95C+) even with a case full of fans (7!), they're much happier blowing free in the wind.

That's understandable, you can go ahead and leave it in if you wish.  Perhaps there should be some context for the numbers added, but this could get really cumbersome.

For example, that 540MH / 530W setup is:

2x 5830's @ 270MH/card, clocks 875/1000mhz gpu/ram
6x 1TB HDDs (most are 'low power' versions)
1x Opteron 6128 ( 8 cores @ 2.0ghz each )
4x 4GB DDR3-1333 RAM
4x 120MM 2400+ RPM fans, 4x 60MM high-flow fans lined up in the card gap
Corsair HX650 80+ Bronze 650W 'professional series'

Certainly some highly-pertinent information can be gleaned from there, such as the ~200W for that base system, but doing so could quickly become really tedious

I don't own it, but the manual doesn't seem to indicate any disabling of slots with the 1x slot occupied.  The only notes they give indicate that using all three PCIe 16x slots will restrict bandwidth to 8x/8x/4x for those slots.

http://europe.asrock.com/downloadsite/manual/Z68%20Extreme4.pdf

It's this regular-style exhuast:


Versus XFX's exhaust:


It's hard to say which is more restrictive..  chopping off the logo or the fins is definitely going to hurt resell value if you've ever got that intention

There are a couple of ways to do this.

You can simply short the GREEN PS_ON to any one of the black 'COM' pins (ground), this will render the power supply 'always on'.

If you want it to turn on with the mobo connected supply, wire the same green PS_ON pin from the mobo supply to your second supply's PS_ON pin, along with one of the adjacent COM (ground) pins.  This is how the cablesaurus adapter works.

Their adapter is handy if you're not comfortable soldering on new wires (recommended), but you can try it out just securing it with electrical tape (not recommended, definitely not permanent)

I'd leave my 540MH / 530W off the list, it's a poor example since my base system uses ~200W of that.  

Working on getting a third rig up and running to get those cards out of my server anyway, my UPS hates them!

[edit] This might do well on a wiki-style page somewhere like the HW comparison guide, so that no one person need feel like it's their job to keep it up to date.  That, and these numbers may change as posters realize their mh/w ratio sucks balls and work to improve it

Hooray, I'm average!

I've come to realize that GPU temps are dependent on a huge array of factors - it's difficult to optimize all of them while retaining performance.

I have 4x 6950's, they were all running ~85-90C in a case with exceptional airflow.  I now have them 'open air' with a ridiculous ventilation system setup to push cold air from an adjacent room directly onto the cards, and I'm getting between 75-85C now.

If your cards are the 'two fan' variety, the only 'intake' is the fans themselves.  They will simply expel hot air everywhere, which generally rolls around and gets sucked back into the fans.  This is a fatal flaw of these cooler types.  Blowing a box fan, etc directly at the cards will make no difference, and may infact increase temperatures, as you're fighting against the exhaust of the cooler fans.

What flags are you using to run your miner?  I've found a drastic difference between -f10 and -f30 on my cards with poclbm for example;  With -f10, my GPU runs at 99% load and temps raise at least 5C.  With -f30, I run at 98% and cooler.  I lose about 5mhash/s / card at -f30 (the default).

MSI's website blows and I can't see pics of the card for some reason, so I can't offer more specific suggestions.

For a closed card with turbine fan (reference cooler), intake is at the rear of the card and exhaust is out the back of your case - these are very easy to direct air into.

Dual-fan or non-reference designs with big honkin' heatpipes, etc, are harder as you need to try to prevent the expelled hot air from being recirculated through the card.

If you've re-applied the thermal paste properly (very thin layer spread over the surface of the GPU), and you're getting 83C at full load, you're actually not doing too badly.  These cards can usually take the pain, and don't thermal-throttle until 100C+

I'd do more experimenting before spending a bunch of money on aftermarket coolers.  Don't get discouraged, it's really not easy to find the right balance.

I'd try testing them not one-at-a-time, but in multiples.

IE, boot with 2 cards and make sure they respond, then try 3, 4, and 5.

My guess is that you'll get up to 4 cards with no problem, but the 5th won't show up - this would be a limitation of the motherboard, disabling the 5th slot when the other 4 slots are occupied.  I have a similar issue in my server with 7 slots - if slot 1 is occupied, slot 5 is disabled.  There may be BIOS options to configure this behavior.

Which motherboard (brand + model) are you using? Some quick Google searches will tell you if this is the problem.

[edit] Just re-read and realized you said your PC 'dies' with 5x cards - this could be a power issue if your extenders don't have MOLEX power adapters.  Each card is going to be drawing 75W from the PCIe slots directly, and 375W might be too much for it to handle.  The shutdown could be over-current protection.

I ordered one of these last week at the $59.99 USD price, got it DOA and had to ship it back for $9.09, turned out to be an expensive drive.

When I dropped the RMA at UPS, there was an *identical* box on the counter already, I had to do a doubletake and make sure I wasn't in the twilight zone!  I don't know for sure if it was the same drive being RMA'd by someone else, but chances are fairly good.

I've got two of these in my server already and they're great drives, but be warned of potential for DOAs.

The 80+ certification means that the PSU is at least 80% efficient over all loads - in other words, to get 100W out it may pull 120W at the wall.  Generally these PSUs are more than 80% efficient, with the highest efficiency between 50-70% of their maximum load.  80+ Gold PSUs are often 89%+ efficient at those loads for example.

Your non-80+ PSU should still deliver the quoted maximum wattage, but may be pulling significantly more at the wall.  If it's rated 750W, it should deliver 750W without a problem.  Of course maxing it out is definitely not good for a cheap/generic PSU, it will eventually burn out on you.

If that 700W 80+ PSU is running all 5 of those 5850's, you're really close to max, if not exceeding it, which is bad for efficiency and probably worse for the PSU.

I can estimate it for you - 6970's pull ~160-200W, so 5x would be somewhere between 800 - 1000W

Google + simple math = your friend.

[edit][disclaimer] I don't own these cards (I do own 6950s with similar draw), and this is not taking OC into consideration.

2 6950x2 rigs getting 720MH @ 410W each, so 1.4GH for ~820W

I've also got a pair of 5830's in my Opteron server, for 540MH @ 530W - obviously this is a poor ratio, but pre-mining I was living with the server pulling 200W on it's own so I'm OK with these numbers.

Looks like I've got some optimizing to do

It's probably a pipe dream unless I suddenly win the lotto and can get back to hobby projects.

  |
  |
 \ /
 

I worked quite a bit porting Froyo (Android 2.2) to the Nook Color in it's early days (I'm also cicada on the XDA forums), and helped a bit getting Cyanogenmod running on it.  It's an amazing little tablet.

Someone was working on a full Ubuntu port, but there are serious hurdles there, so I don't expect to see a 'real linux' on it anytime soon.

NC has a pretty powerful DSP chip, which is really a glorified FPGA, what'd be really cool is if someone could write an android app that takes advantage of this, might get 20-30mhash or so from it

Unfortunately programming for the DSP chip requires proprietary toolkits from TI, and doesn't look like a walk in the park.

The extenders with molex power are, imo, almost necessary if you're going to be using more than 4 cards.

PCIe slots provide up to 75W of power to the card directly.  With 4 cards connected, you're pulling 300W through your motherboard.  Attempt more than that and your board, PSU or ATX connector will likely 'release the magic smoke'.

Depending on the motherboard, 300W may even be too much.  Eg. the manufacturer of a budget board with 2 * 16x slots may assume nobody will ever connect more than two high-power graphics cards, that nothing in a 1x slot could possibly require 75W, and skimp on those traces.

There's no good rule of thumb here, but my suggestion is that anything not connected to a 16x slot should probably use extenders with molex power. 

Then again, the MSI BB Marshal and it's 8 * 16x slots doesn't fit this rule - you probably don't want to be pulling 600W through it, though MSI may have taken this into account when designing the board.  Attempting to connect cards to all 8 sockets, however, will probably melt / burn the 12v wires connecting your PSU to the mobo (ATX24 pin) if they aren't of sufficient gauge to handle the current.

The extenders don't require any drivers, they're simply moving the existing socket away from the motherboard.  Your cards should work in any OS as if they were directly connected.