Estimated long-term costs of owning and running different ASICs

[crazyates]

I decided to grab the specs for a number of popular ASICs, and see which ones would be the most cost effective, assuming all hashrates were the same. What I mean by that is this: if I wanted to get 1TH/s of hashrate, what would be the cheapest way to do this? What would be the most power efficient? What about long term costs?

If I got 1000GH/s from company A, or 1000GH/s from company B, the BTC that they both earned in a given time would be worth the same in the end. To maximise profit, I would either have to A) decrease the initial investment, B) decrease the operational costs, or C) sabotage the whole network so the difficulty stayed low enough for me to mine away. Sadly, I don't think C) is really a viable option.

So here's what I've come up with:


A few things that stood out at me:

1) Row #9 is the most important for keeping the initial cost of the hardware as low as possible. Those ASICMiner devices are horrible overpriced.

2) Note that the K16 board is just for the hashing board, and nothing else. The Terrahash unit (which is based on the K16) comes with a case, PSU, and controller, so it's more in line with the actual cost of buying a bunch of K16 boards.

3) Those K16/K64/DX setups are priced pretty close to where the BFL hardware is priced at, but both of those were before KNC or BitFury.

4) Power efficiency: Power draw tops all, and that does mean something. I was basing these numbers off of my residential rates of $0.15USD/KWh, so if you get cheaper electric, even better. For me, it would cost over $11,000 USD to power 12 Avalons, yet it would only cost me less than $2,000 USD to power a few BitFury units. Obviously, those add up over the 10 year total cost of operations.

4b) Compare KNC's 400GH/s unit to MegaBigPower's 400GH/s (BitFury) unit. KNC is $6,500 cheaper! Yet using 2.5x the power makes it more expensive over 10 years by over $13,000! When you drop the electric rates down to $0.05, they end up being about the same.

5) I realize only the ASICMiner hardware is available now, but that could very well change in the next 6 months. BFL seems to be getting shit out the door. Those 3rd party Avalon chip boards seem to be coming along quite nicely, and should be available in the next couple of months. I'm interested to see if ASICMiner changes their prices to compensate (EDIT: They have, but not enough). I haven't even seen a pic of a completed product from KNC or BitFury, so we'll see on those fronts.

What do you all think? Again, this is the speculation forum for a reason.

[SGExodus]

Your assumptions are all wrong.

You cannot assume a constant network difficulty to derive your total earnings over a period.

[darklight]

Read the post, it doesn't go into earnings. Its a comparison of set up and running costs for a 1TH cluster.

Its certainly some interesting findings. Probably quite an eye opener to a lot of people

[crazyates]

Your assumptions are all wrong.

You cannot assume a constant network difficulty to derive your total earnings over a period.

[SGExodus]

Ok, my bad.   Misinterpreted the Table.

[rudrigorc2]

Is there a thread where people discuss how dumb move was to fund BFL and Avalon?

[crazyates]

Is there a thread where people discuss how dumb move was to fund BFL and Avalon?

I don't know, but this isn't it.

[superresistant]

Please add KNC miner.

[k9quaint]

The image isn't loading.
If it is a spreadsheet, you could put it in google docs.

[jspielberg]

Asic Miner's mission is maximize BTC profit and not to provide hardware so as to create viable competitors.
It's pricing reflects that.

Metabank's numbers are be interesting... they could potentially deliver earlier than BFL's backlog:
Speed: 120GH
# to get to 1TH/s: 9
Total Speed: 1.08 TH
USD/unit = 2160
BTC/unit = ~20.5
Total USD = 19440
Total BTC = ~ 185
GH/1KUS = 55.55GH!
Power: ~250W (prediction 120; seems unlikely - doubling it)
Total Power Draw = 2250
Draw per TH = 2.083
GH / KW = 432
1yr total Pwr cost = ~3K
5yr Total Cost = ~35K
10 yr Total Cost = ~50K

At this point, I think I would put them at equal to (or slightly behind ) Terrahash, and ahead of KNC.
Both Terrahash and Metabank have chip designed and both are working on board design.
KNC is awaiting chip design and board design.

[crazyates]

Please add KNC miner.

I specifically only added those ones, as they seemed the most likely to be legit. Can you link to a thread or post where KNC can be deemed as 100% legit? I've heard of them for a while, but I don't know anything about them. I've also not added BitFury's in there, although I will once we get more info.

[crazyates]

Please add KNC miner.

I specifically only added those ones, as they seemed the most likely to be legit. Can you link to a thread or post where KNC can be deemed as 100% legit? I've heard of them for a while, but I don't know anything about them. I've also not added BitFury's in there, although I will once we get more info.

I've added KNC. I'll add Bitfury a bit later. I also included power costs for $0.05/kW and $0.10/kW, on top of the original $0.15/kW.

[ryszardzonk]

It must like the nicest looking spredsheet I've seen so far but I have two questions. Where is the download link  and what is "cost drop"?

[superresistant]

Please add KNC miner.

I specifically only added those ones, as they seemed the most likely to be legit. Can you link to a thread or post where KNC can be deemed as 100% legit? I've heard of them for a while, but I don't know anything about them. I've also not added BitFury's in there, although I will once we get more info.

I've added KNC. I'll add Bitfury a bit later. I also included power costs for $0.05/kW and $0.10/kW, on top of the original $0.15/kW.

thx

[crazyates]

It must like the nicest looking spredsheet I've seen so far but I have two questions. Where is the download link  and what is "cost drop"?

The original 10 year cost estimate was based off $0.15/kWh. I made 2 revisions, at $0.10 and $0.05, with updated 10 year costs. The "cost drop" is the savings over the 10 years compared to the $0.15 price.

I'm also going to post another revision, with 2 additions: Jalapenos that are flashed to 8GH/s, and Bitfurys ASICs. Does anyone have a link to a thread with all of Bitfury's specs? It's almost 4am here, and I'm a lil tired.

[crazyates]

I'm also going to post another revision, with 2 additions: Jalapenos that are flashed to 8GH/s, and Bitfurys ASICs.

Updated OP with those 2 revisions. How does BitFury plan on getting 3x the power efficiency as BFL at the same 65nm? He's claiming even 50% more efficient than KNC, which is supposed to be using a 28nm process. Weird.

[-ck]

I'm also going to post another revision, with 2 additions: Jalapenos that are flashed to 8GH/s, and Bitfurys ASICs.

Updated OP with those 2 revisions. How does BitFury plan on getting 3x the power efficiency as BFL at the same 65nm? He's claiming even 50% more efficient than KNC, which is supposed to be using a 28nm process. Weird.

At this stage it remains a claim then, much like earlier BFL claims...

[rudrigorc2]

One thing that surprises me, BFL claims to have 65nm and its amazing that the chip runs hotter than 110nm avalon. Are they overclocking that much to fit in that box? Sounds like a timebomb...

[crazyates]

One thing that surprises me, BFL claims to have 65nm and its amazing that the chip runs hotter than 110nm avalon. Are they overclocking that much to fit in that box? Sounds like a timebomb...

BFL consumes more power (heat) per chip because it's a physically larger chip (56mm²) that does 4GH/s, rather than the tiny little chip (15mm²) that does 300MH/s.

If BFL wanted to make a chip that was comparable to an Avalon or ASICMiner chip, it would be 3.5mm² (yes, less than 2mm x 2mm), have a single engine (rather than the current 16), hash a little higher than 250MH/s, and consume 0.2W. But that's kind of a stupid design for a chip, so obv that wouldn't make any sense.

[rudrigorc2]

One thing that surprises me, BFL claims to have 65nm and its amazing that the chip runs hotter than 110nm avalon. Are they overclocking that much to fit in that box? Sounds like a timebomb...

BFL consumes more power (heat) per chip because it's a physically larger chip (56mm²) that does 4GH/s, rather than the tiny little chip (15mm²) that does 300MH/s.

If BFL wanted to make a chip that was comparable to an Avalon or ASICMiner chip, it would be 3.5mm² (yes, less than 2mm x 2mm), have a single engine (rather than the current 16), hash a little higher than 250MH/s, and consume 0.2W. But that's kind of a stupid design for a chip, so obv that wouldn't make any sense.

Oh I forgot they are much faster, but still ~90 celsius seems a lot