Topic: Enough Power? (Read 2793 times)

I thought I'd pop in to say I got the Cooler Master GX650 and all is good, at the moment I'm only running a 5850 I ordered a 5830 from EPiSKiNG so I'm waiting on that, however I was using a similar device to a kill-a-watt from Canadian(eh) Tire and originally was reading 1000watts which must be horribly wrong but with the GX I'm getting 200watts which is more reasonable.

Any ideas? Possible witchcraft/sorcery?

I have 3 phases entering the house (120V each). Of course almost everything uses just 2 at 220v, but I have an oddball self-built sawing machine in the garage that needs 3 phases.

Well that one stumped me.  Where do you live that you have 380V mains?

Amateur. I have 380V in my garage!

I just got something similar to that, but I must have the wrong model because it refuses to talk to my KVM switch

No but I will take some this evening if anyone is bitcoin nerdy enough to see a 240V L6-30R circuit (handles up to 7.2 KW load) with high current PDU looks like.  I don't know of anyone putting one in their garage before.

Do you have some pictures of your rigs, D&T ?

I have great respect for you and your knowledge in all things mining !

Even compared to FPGA which do exist the return isn't worth the risk. 

As a tech lover it sucks that cheapo CPU, cheapo RAM, usb stick, decent MB + lots of GPU = winz.

As a capitalist I own 5 such rigs. 

I also am not just thinking of CURRENT FPGA but a year is a long time in silicon world.  FPGA designs will only get more cost effective.  28nm chips likely will be reasonably available this year.  Who knows what the pricing is but FPGA getting 25% or 50% higher efficiency (MH/$ and MH/W) is certainly a possibility. 

All that puts a damper on long payback periods on GPU tech.

Not sure if you chose to notice, but I specifically worded my post to point out that it is obvious that the item does not exist and is difficult to believe that it does exist. This does not mean that I am "dreaming" of such a thing. I was simply using it for comparison to what I actually was dreaming about.

You appear to be deathly afraid of it, if it were true. I think that if we gain such capabilities, miners in general will embrace the technology. Whether or not it exists is a completely different and unrelated matter.

Funny that some of you still think or dream about the 25K top grade unicorn being real. Getting 50 ghash/s with 32 singles at 800 mhash/s each does not compute. Probably will get only around 25 ghash/s in the end LOL and draw 10 000W.

Paying $25k and getting 6 month warranty on a product that is TOTALLY USELESS outside of this community = too much risk for me.

Maybe others can stomach that cost and risk etc. but not me. Good luck with your backplanes.

So I decided to run some more numbers, and found out just exactly how bad of an investment this could potentially be.
So lets assume that IF a 7990 had 2 cores capable of 600 Mhash/s each at stock clocks, and IF they pulled the full maximum of 375 watts for a PCI Express card (2x 8pin @ 150watts each = 300 watts, plus 75 watts from the slot), and IF I could get all 18 slots populated and working with virtual machines or whatever technology, and IF it cost me a conservative estimate of $25,000........

.......that would mean numbers such as the following:
21.6Ghash/s per rig
~7500 watts power draw at the plug(s) (assuming 95% efficiency, which is more common in server-grade power supplies)
6-8 U rack spaces, depending on how it was architected.

Now, if we assume for a moment that the BFL Rig Box exists and works at the numbers stated, that is as follows:
~$25,000
50 Ghash/s
2500 watts
unknown rack spaces

I suppose this serves to illustrate how badly FPGAs have the potential to kick  whomp  BLAST the ass of video cards into next week.

Since my power is 9 cents kw/h, I decided to do an ROI calculation. For the custom monster machine with 18 video cards, ROI is slated to be ~250 days, assuming the difficulty and price remain the same (yeah right ). For the Rig box, it is 95 days. Of course, the video cards and waterblocks have known resale value, but damn.... less than 100 days. I STILL am not sure whether I want it to be true or not.

EDIT: minor spelling corrections.

Wow nice find.  Significantly cheaper than the cheapest I have ever seen.  If it weren't for FPGA I might be crazy enough to try something like that.  I wired a 240V, 30A circuit w/a locking NEMA L6-30R connector and a used ADP 30A PDU.   Nice setup BTW.  Solid, clean power, and no issues w/ overloading circuit in the rest of the house.  It is enough for 20 5970s (40 GPU total) or ~14GH if I wanted to expand.

Still FPGA worry me for anything w/ a ROI beyond a year.

Like I said - pie-in-the-sky - but what made me think about it seriously was this: http://www.ebay.com/itm/260836521449 only $650.
I estimated $2000 for the SHB host card because it wouldn't need to be the top-end model with all the latest stuff.

Oh it certainly IS possible but it will end up costing more than equivelent GH using multiple rigs and virtual machines just add more complexity, part failures means losing an entire hashing farm, etc.

What "should" be cheap is a mATX board w/ PCIe extender to another rack full of PCIe expansion slots.  Sadly those are thousands of dollars more than just building a bunch of rigs.

I spent a good amount of time calculating, and pricing, and trying to think outside the box but

CHEAP CPU
CHEAP RAM
USB STICK
MB w/ ENOUGH EXPANSION SLOTS
3 or 4 DUAL GPU CARD
GOLD PSU

is pretty much unbeatable.  5970 or 7990 it isn't going to change the equation.

One way to look at it is.  Excluding GPU and PSU what is the "overhead".

A 3x PCIex16 MB runs $170.  A Sempron $20, Memory $10, USB stick $10
So that cost is $210 for 3 slots = $70 per slot.  Using Dual GPU makes it $35 per GPU.  That doesn't require any expanders.

With expanders you could go cheaper MB w/ 4 PCIe x1 or x16 slots ~ $120. A Sempron $20, Memory $10, USB stick $10, expanders $40
$200 for 4 slots = $50 per slot.  Using Dual GPU that is $25 per GPU

Now AMD drivers limit to 8 GPU so you could go w/ more slots but then you are talking about single GPU so it doesn't really save much $$$.




To be more cost effective an 18 slot backplane would need to be  (excluding PSU + GPU) $70 * 18 = $1260.  You can't find a backplane + SMB for less than $5000.

Even if you wanted some monster radiator to dump 5KW outside it would be cheaper to just build 5 or 6 watercooled rigs w/ no radiator and a fluid-fluid heat exhanger.  Then get a loop to combine connect the heat exchangers w/ larger tubing, powerful pump, and massive radiator.  I am thinking about doing that this summer.  The problem is the capital costs are huge and AC although inefficient is cheap (relatively speaking).

My power load is ~4KW.  So $3500 per year in power.  However AC requires ~30% more power to remove the heat 6 months out of the year so that is $600 per year.  For the benefit of the doubt lets say $800.

Still to waterblock 15 5970 will cost ~$1800.  5 heat exchangers is another $500.  Pumps, resivours, tubing for 5 rigs is ~$1000.  So we are looking at $3300 for the primary loops.  Lets pretend I get parts for secondary loop for free. 

$3300 / $800 per year = 4 years payback.  It just doesn't make economic sense.

No, I don't have enough capital. But I am still trying to crunch all the numbers and see whether it could be done. Pie-in-the-sky dreaming I guess.

You already started on the backplane project ? Did you buy a backplane already ?

Well my estimates were based on $650 6990s, $110 waterblocks, a $2000 host computing board, a $650 backplane, and six PSUs hopefully being less than $500 each. The unfortunate issue is that the manufacturers of the waterblocks don't publish technical specs such as flow rate at a certain pressure, recommended pressures, etc. so my numbers might be bad estimates.

However, from my calulations, this radiator should be able to dissipate all the heat from 18x 6990s when used between 0 and 1 Bar of pressure, because it is rated at 10 horsepower heat dissipation at that pressure. Yeah I know it is an oil cooler, but there shouldn't be any reason for it not to work with glycol or similar. The heat dissipation table is on the catalog page at this link.

Wow, you're going to have to use radiator surface the size of a football field
Going with so many pumps, water blocks, all that watercooling stuff... I'm afraid even 20k$ might be a low estimate.

My hang-up is the watercooling - it would be a very complex endeavor. How to properly size the pumps, radiators, manifolds, etc, and do it in such a way as to fit in a standard datacenter rack. Air cooling would be all but impossible if I wanted to fill all the slots. I estimate costs per unit to be just under $20,000 and use approximately 35-40 amps of power continuously at 240 VAC. So it would be rather expensive to start with, and could take a while to return any sort of investment.

Funnily enough I investigated these platforms with backplanes but nobody tried it and I don't wanna be the first either

In theory it should make a hell of a mining farm !