Topic: Mining rig extraordinaire - the Trenton BPX6806 18-slot PCIe backplane [PICS] - page 17. (Read 169522 times)

State your motivations then, I'm exploring all avenues with bitcoin mining

And you just proved that you don't understand that density is a massively important factor in mining farms...

You don't seem to be able to understand the motivations for this. That's fine, leave me to it.

so you yourself just proved how bad the idea was lol

I said 7990. It is a different card. 7970 does not get 800 mhash either, unless perhaps it is extremely overclocked with something other than water cooling. 7970 gets 550 mhash stock, and 650 OCed on air.

Why would 7970s get you 1 ghs? Everywhere else it says 800 mhs...

Run the numbers again. Sure it is fairly bad pricewise, but it has the potential of being actually quite efficient, both on electricity and more importantly on space. Instead of having a CPU/RAM/mobo for every 8GPUs, I have 2 CPUs (could be 1 if I wanted), some RAM, and a mobo for up to 36 GPUs (including dual-GPU cards). This assumes that I can get them to all run on the platform, which is still in question. Nvidia would likely be less hassle, but they don't make good mining cards (at all).

If we assumed that I waited for 7990s and assumed that it could produce 1Ghash at stock or very mild OC, then we get some insane density. I had been running all kinds of numbers on 6990s with water cooling, but it would be damn difficult to pull off, with just one of the issues being that waterblocks never seem to be designed to run in adjacent slots for some reason. They might fit if you get the right kind, but it is very tight.

You would then need supports built into the case so that the cards wouldn't flop around, a custom power distribution system, and a massive pump and radiator (or 2). My choice of PSU was based on size - they are tiny for how much power they put out. Dell already uses 6 of them side-by-side in their blade servers, so I know that 6 could fit in 2U at 19" wide. Add 3 more U for the backplane and cards, then another U for hoses and fittings, and one more U for sufficient airflow. That last U isn't needed if the radiator isn't attached to the case, but is instead mounted remotely or replaced with a heat exchanger.

Total power draw could be as high as 7 to 8 KW at 90%+ efficiency. Heat dissipation could be as high as 10 horsepower per 8 U, if you use an inbuilt radiator.

a top of the line gpu equals 800 mhs x 18 = 14,400 mhs or 14.4 ghs

this project is not very cost effective

we are hoping for 20GHash/s

in the end it all comes down to how much gigahash it can produce... so where is the final number?

i honestly don't know since my rigs are very small and i have the cheapest power. i am going to check and get back at you

What is the efficiency of the ones you are using? If they are reasonably efficient, I could use them since you would have a pinout for them already. I'm not stuck to the Dell ones that I have, but they are quite high power for their size which is useful.

+1 on the server psu, i've been using server psus ever since i started mining and i wondered why no one else was doing it

Don't know the pinout for that one though, the ones i use are for IBM blade servers

Sorry it was facebook and the open compute project not google.

Open source data center.  

http://opencompute.org/

Anyways here is the all 12V PSU.  


http://opencompute.org/projects/power-supply/

277VAC input (w/ 24VDC for direct connection to DC battery bank for uninterruptable power)
Efficiency is >95% for 50% to 90% load (>90% at 20% load, >91% at 100% load)


Anyways the larger point was there is no real need for a PSU to supply anything but 12V.  The multi voltage PSU is simply an anachanism.  Every device can simply provide its own VRM or DC to DC switching and convert the 12V to whatever they need.

I wish I could just run massive 12V cables to my computers or hell, wire up the whole damn house to have 12VDC outlets in every room!  Unfortunately, 12VDC doesn't carry very far without massive cables and major voltage drop.  That's why DC's like Google and others use ~48V as an intermediate step.  Freaking electricity.

Thats not entirely true. The way most of Googles shit is ran, they use 48v DC power (which is really 36-48v), they take apart el cheapo UPSes for the 48v->48v conversion, and then they step 48 down to 12, plug the 12v rails right into the 12 line, then use a picopsu to get 5v and 3.3v for stuff that legitimately runs on those voltages.

The future of computing most likely will be 12v and 3.3v only. More and more SATA devices are requiring the 3.3v rail to function.

What I'd like to see, really, is a DC/DC converter block that is outside of the PSU that provides legacy connections, such as 5v/3.3v (and pass 12v straight through) for drives. So, say, 4 SATA power plugs that plugs into a 6 pin PCI-E and does the DC/DC conversion in line.

Just a quick update showing +12vdc connected. I wish I had more time to work on this it's going so slow right now.


Now with added win (+12v)

The guy I bought the orange wire from for 3.3v listed it as 14 gauge, but shipped 12 gauge. I might have to get some different crimp connectors or else steal some 16 gauge from another wire bundle.

Oh yeah you definetely should connect all wires I am just saying it is unlikely you need "a lot" of 3.3V or 5V.  Totally seperate from that idea is the fact that FUTURE (as in a spec) computers no longer need anything other than 12V.  Google runs entire datacenters on custom built servers which have only 12V connections and power supplies which only have a single 12V rail.   Multi voltage PSU are an antiquated design from a bygone era.  They have simply stuck around due to inertia and backwards compatibility.

Can't fit a card into the last slot over there? What a waste of a slot :[
You should saw off the metal thing above it and donate the slot to me.

The only thing that pulls 3.3v and 5v are PCI-E cards and on-mobo peripherals. The VRMs for the CPU, RAM, Northbridge, and Southrbidge all pull off the P4/EPS12 plug.