Bitcoin Forum

I plan to run 4 HD5970's on a water cooling, what is the best motherboard for this for the price?

In the time you'll set up the new hardware difficulty will rise by another 40% average. Buying _now_ hardware to mine isn't the best idea ever.

That system with watercooling will easily be over 3000 Euro.

I already have a system running just upgrading cards and motherboard.

http://www.newegg.com/Product/Product.aspx?Item=N82E16813130223

or
Giga-Byte GA-890FXA-UD7 <SAM3, AMD 890FX, 4*DDR3, 6*PCI-E16x, SATA RAID, 2*GB Lan, XL-ATX

As far as I know you can't set 4 5970s in one system since each of them already has 2 GPUs onboard.

It is certainly possible, some people around here have done it.  I think only the linux drivers support 8 GPUs.  Regardless, going with 2 5970's per computer is probably the more cost effective option.

Agreed...

Just what kind of PSU (would one handle it) would you need for quad 5970s?

The cheapest option would be to use two PSUs.  A while back I bought one of these in an attempt to run 4x5970 http://www.newegg.com/Product/Product.aspx?Item=N82E16817256054 (http://www.newegg.com/Product/Product.aspx?Item=N82E16817256054)  Expensive...

for dual 5970s, you shouldn't need more than 950ish watts.

A good cost effective board would be something like the gigabyte 890GPA-UD3H (http://www.newegg.com/Product/Product.aspx?Item=N82E16813128435)

I had one, very nice board for the $$$

We already passed the point where buying hardware was profittable.

I did the computation for myself. Adding a new video card (only the card), a good bitcoin generation/dollar one, will not pay for itself. I didn't count the electricity.

Agreed - as fast as the difficulty is changing now it isn't worth it.

Now - if you were doing this 4 months ago - maybe so...

2 x 750w PS should just fine

Yeah - that would seem best...

I am not wise enough to figure how to get two PSUs to supply a single system  ;)

You can't. A 450W PSU it's around 40$, 2x450W = 900W would be 80$. Why the hell would someone pay over 130$ for something he could get for 80$.

The only thing you _MAY_ try is to use 1 PSU to power 2 video cards. The different load will cause different voltages and most likely system instability, it may even burn the card VRM.

http://blog.zorinaq.com/?e=42

http://blog.zorinaq.com/?e=44

It's doable but you'll need to re-route the mobo power to 2 of the graphic cards to stop the mobo power rails from cooking. If you're going to do that for 2 do it for them all.

http://bitcointalk.org/index.php?topic=3883.0

Looks pretty sophisticated, anyone tried that direct PSU feeding trick yet? I can't seem to understand what connectors he's connecting to and to which PSU plug it is hooked up.

It looks like he's just a grabbed a 12V line from the PSU (off a Mobex connector?) and soldered it to 5 12V broken out ribbon-cable cores of PCI-e x1 extender cable.
http://blog.zorinaq.com/images/flex-pcie-4x-modified.jpg
http://blog.zorinaq.com/images/flex-pcie-before-after.jpg
"...re-route the five 12V lines (PCIe pins A2, A3, B1, B2, B3) to a single 16AWG stranded wire directly connected to one of the 12V outputs of the power supply. ..."

The jumper trick across A1 to B17 on the slot connector to make the down-plug work is necessary also.
http://blog.zorinaq.com/images/pcie-short-schematic.png
http://blog.zorinaq.com/images/pcie-short-photo.jpg

So he's connecting 5 cards, by taking the powerlines of the respective extender cables and soldering them into _one_ molex female plug, and them he plugs it into the power supply? Or does he use a molex from the PSU for each card?

He spells it out in good detail if you read it carefully.

4 hd5970 cards, 1 mobo.
4 PCi-e extender ribbon cables, each one modified to have 5 12V cores soldered into 1 12v (yellow) 16AWG from PSU ... it doesn't really matter what connectors you use if you know what you are doing.

How are you calculating profitability? I came up with the following naive analysis - can you guys poke holes in it and tell me how this isn't profitable?

 - Each server has 2x XFX Radeon 5970 boards @ $900 each (approx)
 - The rest of the server costs $1000
 - Total server cost is $3000
 - At the current difficulty level, each server will crank out 50BTC approximately every 1.66 days (4 x 5970 GPUs)
 - 50BTC is worth 0.83 x 50BTC = about $40 USD
  + Implies daily revenue of approximately $40 / 1.66 = $24/day
 - Power consumption per server is 1.5kW
 - Power cost is therefore 1.5 x $0.10 x 24 = $3.60/day
 - Net after power is about $20/day per server
 - Days to recover hardware investment: $3000 / $20 = 150 days
 - Depreciation rate of hardware: 45%/year -> implies $1350/year loss in value
 - Net profit per year including hardware depreciation is $20 x 365 - $1350 = $5950

How is this not incredibly profitable? I must be missing something.

Thanks!

Yep, you are missing something.
2 5970's (4 GPU) maxes out at around 1.2 GHash/sec
http://www.alloscomp.com/bitcoin/calculator.php
That's 50 BTC every 3.125 days roughly ... assuming 100% uptime.

So you can roughly half everything you wrote after this point.

Okay, this is where I went wrong. The days to generate 50BTC on average is double this. I was assuming that the CPUs were doubly as powerful as they really are. If you cut in half the revenue run-rate, then mining is a lot less profitable than I modeled above. But still, my model shows an annual return on capital after financing, power, depreciation and conversion costs of close to 60%. Very few businesses do that well.

Your other potential issue is in assuming the difficulty rate will be constant for a whole year--in fact, as long as mining at any scale remains profitable, it is likely to increase continually.

Got it - however, as I understand it, as the difficulty increases and profitability decreases, fewer people will find mining worthwhile, eventually leading to a reduction in difficulty if the number of blocks being solved drops below the required rate to keep the market going. There must be some steady state and I wonder how close we are to it. I have heard others argue that it's unlikely we'll see the same pace of difficulty growth in the near future as has been the case for the past several months.

for daisy chain two power supplies in your computer

http://www.kustompcs.co.uk/acatalog/info_1919.html

I'm not sure the ultimate equilibrium will be profitable, because there are always people who are not paying for their electricity and/or hardware--these people will mine no matter what the price/difficulty.

Also figure in the people that don't care or don't realize they're losing a little and a few other cases.