BFL Power efficiency argument fallacy

[CoinHoarder]

Everyone knows by now that BFL has more power efficient ASICs than their competitors (or at least, they think they do). Therefore people say they will be profitable for longer, which is true.

However, I'm not sure that this is as big a deal as people claim. I don't think that most people are taking into account 2nd generation ASICs.

BFL, and other ASIC manufacturers, are going to see sales diminishing as difficulty rises throughout the year 2013. They will be looking for a way to increase or maintain profit. One of the most obvious ways of doing this will be to release 2nd generation ASICs.

If BFL puts a cutoff date on trade ins for 2nd generation ASICs (lets be honest, what company would not put a cut off date on such a thing?), like they have on the 1st generation, then all BFL customers will need to pay more capital to trade in their ASICs for the 2nd generation unless they want to quit mining, or be stuck with a previous generation ASIC.

The sooner they announce/preorder/deliver 2nd generations ASICs, the less important 1st generation power consumption rates will be because you will have less time to reap the benefits of the lower power consumption and you will need to spend more money to upgrade.

I'm just thinking aloud here...  whats your opinion?

[1715367410]

1715367410

[1715367410]

1715367410

[1715367410]

1715367410

[1715367410]

1715367410

[1715367410]

1715367410

[bitcoindaddy]

Everyone knows by now that BFL has more power efficient ASICs than their competitors (or at least, they think they do). Therefore people say they will be profitable for longer, which is true.

However, I'm not sure that this is as big a deal as people claim. I'm not sure that most people are taking into account 2nd generation ASICs.

BFL, and other ASIC manufacturers, are going to see sales diminishing as difficulty rises throughout the year 2013. They will be looking for a way to increase or maintain profit. One of the most obvious ways of doing this will be to release 2nd generation ASICs.

If BFL puts a cutoff date on trade ins for 2nd generation ASICs (lets be honest, what company would not put a cut off date on such a thing?), like they have on the 1st generation, then all BFL customers will need to pay more capital to trade in their ASICs for the 2nd generation unless they want to quit mining, or be stuck with a previous generation ASIC.

The sooner they announce/preorder/deliver 2nd generations ASICs, the less important 1st generation power consumption rates will be because you will have less time to reap the benefits of the lower power consumption.

I'm just thinking aloud here...  whats your opinion?

I don't think the power improvements of later generations of ASIC kit will be as dramatic as the shift from GPU to ASIC. Probably on the order of 10-15 percent? In fact, I would be surprised if there was a second generation at all.

[firefop]

Really the only thing they'll be able to do second gen (or third gen) is use smaller gap silicon - which allows for a speed increase in the same heat/watt space. As such, I wouldn't expect many people to upgrade in general.

Now a smarter second gen product would be a same hash power, lesser power consumption unit (in a smaller more modular enclosure)... for a very minor price increase. say 5%. That might actually get people to buy them instead of first gen... but the company could continue to sell both. So then you've got the "refurbished generation 1s" going for 5% or 10% less than a new gen1, and a gen2 device going for 5% or 10% more.

I'd want to see upgrades in the form of better board design or innovative cooling instead of following the gpu model and going "faster, hotter, more power needed, in the same space". Space shouldn't really be a concern imo.

Now give me a rolling racked enclosure with slots for the gen2 singles where the enclosure uses a direct contact copper heatsink and I just plug it into a slot... some management metrics displayed on the outside of it... and I'd be all about upgrading to that.

[CoinHoarder]

Now give me a rolling racked enclosure with slots for the gen2 singles where the enclosure uses a direct contact copper heatsink and I just plug it into a slot... some management metrics displayed on the outside of it... and I'd be all about upgrading to that.

They are not stupid, they will release a product worth upgrading for IMO. That's what smart businesses do, they adapt, change, and upgrade their original big money making ventures to make more money. They will listen to customer complaints and wishes, and then develop a product that meets them to sell more units. (Example: Every product Apple sells... Ipod, Ipad, etc... What version Ipod are they on now?)

[SgtSpike]

Everyone knows by now that BFL has more power efficient ASICs than their competitors (or at least, they think they do). Therefore people say they will be profitable for longer, which is true.

However, I'm not sure that this is as big a deal as people claim. I don't think that most people are taking into account 2nd generation ASICs.

BFL, and other ASIC manufacturers, are going to see sales diminishing as difficulty rises throughout the year 2013. They will be looking for a way to increase or maintain profit. One of the most obvious ways of doing this will be to release 2nd generation ASICs.

If BFL puts a cutoff date on trade ins for 2nd generation ASICs (lets be honest, what company would not put a cut off date on such a thing?), like they have on the 1st generation, then all BFL customers will need to pay more capital to trade in their ASICs for the 2nd generation unless they want to quit mining, or be stuck with a previous generation ASIC.

The sooner they announce/preorder/deliver 2nd generations ASICs, the less important 1st generation power consumption rates will be because you will have less time to reap the benefits of the lower power consumption and you will need to spend more money to upgrade.

I'm just thinking aloud here...  whats your opinion?

My opinion is that electric usage under 100w is largely a moot point anyway.  Even at $0.25/kwh, 100w would only cost $18/month to run.  And if ASICs are only marginally profitable due to electric costs (i.e., it is profitable to operate a BFL @ $0.10/kwh but not at $0.20/kwh), then I don't think many people are going to be interested in running them anyway.  If it is profitable to operate at $0.10/kwh, but not at $0.20/kwh, then the maximum profit a 100w device could possibly be making is $7.20/month.  Who wants to bother with maintaining a device to only make $7.20/month?

Likewise, if a 400w device is profitable to operate at $0.10/kwh, but not at $0.20/kwh, then the maximum profit it could be making is only $28.80/month.  Again, hardly worth even maintaining the operation of it.

So, in my opinion, electric costs aren't going to matter until ASICs drop steeply in price, to the point where someone could buy dozens of units for the same price as buying one of them today.  When they have dozens of units consuming many kw, electricity costs will certainly make a much larger difference in profitability.

[firefop]

I also think then next step is alternate use for asics as a heating element.

Lets build a rig that vents its heat to dry clothes.

"Honey, I need you to do more laundry, we need that dryer be mining for us"

[SgtSpike]

I also think then next step is alternate use for asics as a heating element.

Lets build a rig that vents its heat to dry clothes.

"Honey, I need you to do more laundry, we need that dryer be mining for us"

I don't.

ASICs are expensive and fragile.  Heating elements are pennies and durable.  Different items for different purposes.

For warming a room, perhaps, if they get really cheap or for people who already own them.  Not for drying clothes though!

[firefop]

I don't.

ASICs are expensive and fragile.  Heating elements are pennies and durable.  Different items for different purposes.

For warming a room, perhaps, if they get really cheap or for people who already own them.  Not for drying clothes though!

Yes I can see your point. But that's where I see this going if bitcoin does become the giant I think it will. Why not integrate it into standard every-day application? Sure there are potential problelms, but why hardware has generations, to build more sensible and robust solutions.

Anyone in a very cold winter region could for example, use a mining rig to heat a garage instead of paying thousands of dollars an embedded heater. While at the same time making back some of the operating costs of that unit.

What about floorboard heating that uses asics to make the heat... basic electric baseboard heating, that's very cheap to run.

[SgtSpike]

I don't.

ASICs are expensive and fragile.  Heating elements are pennies and durable.  Different items for different purposes.

For warming a room, perhaps, if they get really cheap or for people who already own them.  Not for drying clothes though!

Yes I can see your point. But that's where I see this going if bitcoin does become the giant I think it will. Why not integrate it into standard every-day application? Sure there are potential problelms, but why hardware has generations, to build more sensible and robust solutions.

Anyone in a very cold winter region could for example, use a mining rig to heat a garage instead of paying thousands of dollars an embedded heater. While at the same time making back some of the operating costs of that unit.

What about floorboard heating that uses asics to make the heat... basic electric baseboard heating, that's very cheap to run.

Time will tell, I guess.

It is just much less complicated and less expensive to generate heat from simple $0.50 coils of metal than from an ASIC that requires an internet connection, setup, and maintenance (blowing out dust, etc).  I don't see it ever being practical or desired by the "average joe".

Embedded heaters are cheap.  I could stick a 2000w in-wall heater in my garage for $60 if I wanted - a far cry from thousands of dollars.  Not to mention that anyone concerned with saving money or energy efficiency would be looking at a heatpump, or perhaps natural gas, as a way to do it instead.

[dserrano5]

So, in my opinion, electric costs aren't going to matter until ASICs drop steeply in price

Or bitcoin value plummets to $1, which is not impossible.

[SolarSilver]

I'm just thinking aloud here...  whats your opinion?

if the hardware is written off after 2 years (where it will eventually start to fail due to overheating issues), then the
hardware cost per day is about $US 1.8. Energy if you have to pay for it in Europe is about EUR 0.7 / day at 100 Watts.

If the power consumption halves in the next generation, the proportion of the hardware cost is still too significant to have much inpact on my total profitability. Unless the next generation drops half in price or better, which is a realistic scenario, due to competition (margins are huge, remember the performance jump from 30 to 60 GH/s?)

We'll end up with the same power consumption per unit, but much higher performance probably.

Power consumption is for most of us not much of an issue, as we already live with power hungry GPU setups.

It will be an issue if you want to jam many machines into a small space, but if profit arguments means more old gen machines in a bigger room, I'll settle for that

[PuertoLibre]

Everyone knows by now that BFL has more power efficient ASICs than their competitors (or at least, they think they do). Therefore people say they will be profitable for longer, which is true.

However, I'm not sure that this is as big a deal as people claim. I don't think that most people are taking into account 2nd generation ASICs.

BFL, and other ASIC manufacturers, are going to see sales diminishing as difficulty rises throughout the year 2013. They will be looking for a way to increase or maintain profit. One of the most obvious ways of doing this will be to release 2nd generation ASICs.

If BFL puts a cutoff date on trade ins for 2nd generation ASICs (lets be honest, what company would not put a cut off date on such a thing?), like they have on the 1st generation, then all BFL customers will need to pay more capital to trade in their ASICs for the 2nd generation unless they want to quit mining, or be stuck with a previous generation ASIC.

The sooner they announce/preorder/deliver 2nd generations ASICs, the less important 1st generation power consumption rates will be because you will have less time to reap the benefits of the lower power consumption and you will need to spend more money to upgrade.

I'm just thinking aloud here...  whats your opinion?

That you are spot on.

What you just described is basically a money pit. All the money goes out of your pocket and into the ground (the money pit). In order to recoup your loss you'd have to figure out a way to pass the loss on to someone else who is hopeful and naive.

If you don't, then the only one making all the money in this is the vendor. The only way out of this is by the vendor selling second gen hardware that is several times more efficient than the first gen. At which point you can recoup the money.


-----------------------

But...

Then it becomes a Chicken and the Egg situation. Everytime they sell better hardware, they make the difficulty increase at a faster rate. So you will probably never beat this game. Only if there is exclusivity do you make back your money quickly and then profit.

Exclusivity therefore is the major problem....

Difficulty is just a secondary problem.

---------------------------------

Best you can hope for is that a vendor implodes and takes out a big chunk of the competition with it. Thereby increasing the exclusivity and lowering the difficulty.

A manufacturing defect is good enough. Or bad vendor policy. Whichever comes first.

[makomk]

BFL, and other ASIC manufacturers, are going to see sales diminishing as difficulty rises throughout the year 2013. They will be looking for a way to increase or maintain profit. One of the most obvious ways of doing this will be to release 2nd generation ASICs.

Of course, another way - depending on what their margins are on ASICs - would be to drop the price. Remember that a huge proportion of the cost of ASICs is up-front NRE costs to design the things and get masks made; once they've paid off those costs they can probably reduce prices a fair amount and still make money. Then:

So, in my opinion, electric costs aren't going to matter until ASICs drop steeply in price, to the point where someone could buy dozens of units for the same price as buying one of them today.  When they have dozens of units consuming many kw, electricity costs will certainly make a much larger difference in profitability.

[PuertoLibre]

My opinion is that electric usage under 100w is largely a moot point anyway.  Even at $0.25/kwh, 100w would only cost $18/month to run.  And if ASICs are only marginally profitable due to electric costs (i.e., it is profitable to operate a BFL @ $0.10/kwh but not at $0.20/kwh), then I don't think many people are going to be interested in running them anyway.  If it is profitable to operate at $0.10/kwh, but not at $0.20/kwh, then the maximum profit a 100w device could possibly be making is $7.20/month.  Who wants to bother with maintaining a device to only make $7.20/month?

Finally! Someone who gets it....!

Likewise, if a 400w device is profitable to operate at $0.10/kwh, but not at $0.20/kwh, then the maximum profit it could be making is only $28.80/month.  Again, hardly worth even maintaining the operation of it.

Exactly.

So, in my opinion, electric costs aren't going to matter until ASICs drop steeply in price, to the point where someone could buy dozens of units for the same price as buying one of them today.  When they have dozens of units consuming many kw, electricity costs will certainly make a much larger difference in profitability.

The vendor who figures out a way to modularize their components and add (cheap) upgrade-ability will be the one to succeed. The upgrade path will have to exceed firmware and simple overlcocking changes.

They should also diversify the support for various cyrptocurrencies. So you don't have all your eggs in one basket.

[bcpokey]

My opinion is that electric usage under 100w is largely a moot point anyway.  Even at $0.25/kwh, 100w would only cost $18/month to run.  And if ASICs are only marginally profitable due to electric costs (i.e., it is profitable to operate a BFL @ $0.10/kwh but not at $0.20/kwh), then I don't think many people are going to be interested in running them anyway.  If it is profitable to operate at $0.10/kwh, but not at $0.20/kwh, then the maximum profit a 100w device could possibly be making is $7.20/month.  Who wants to bother with maintaining a device to only make $7.20/month?

Finally! Someone who gets it....!

Even if his numbers are wrong, he does get the general idea, which is good.

Likewise, if a 400w device is profitable to operate at $0.10/kwh, but not at $0.20/kwh, then the maximum profit it could be making is only $28.80/month.  Again, hardly worth even maintaining the operation of it.

Exactly.

Again don't mean to be nitpicker, but it can't be "exactly" if the numbers are wrong, even if the idea is right.

The vendor who figures out a way to modularize their components and add (cheap) upgrade-ability will be the one to succeed. The upgrade path will have to exceed firmware and simple overlcocking changes.

They should also diversify the support for various cyrptocurrencies. So you don't have all your eggs in one basket.

Here's where I begin to question. Modularizability sounds great and all, but it's not really the "game-changer" that you make it out to be. As far as I can tell, the BFL singles (only picture of an ASIC device we have currently) is essentially a box around a heatsink and a PCB+chips. In essence, it is almost as bare minimum as you can get. Aside from being able to buy some sort of massive copper-block with slots to insert boards into, you're not really going to get much more efficiency from some sort of modular design (and that's not very efficient really).

As to the cryptocurrencies, there are already merged mining pools, so all the bitcoin forks do not need to covered by vendors, and the non doubleSHA256s can't work alongside bitcoin for ASICs as needs no explanation.

[SgtSpike]

My numbers are correct bcpokey.  If you believe them to be wrong, then please show the math.

[PuertoLibre]

Here's where I begin to question. Modularizability sounds great and all, but it's not really the "game-changer" that you make it out to be. As far as I can tell, the BFL singles (only picture of an ASIC device we have currently) is essentially a box around a heatsink and a PCB+chips. In essence, it is almost as bare minimum as you can get. Aside from being able to buy some sort of massive copper-block with slots to insert boards into, you're not really going to get much more efficiency from some sort of modular design (and that's not very efficient really).

If the ASIC vendors can make smaller daughterboards that are alike and can be hosted on an internal bus or something like a mini-PCIe board, then that would cut costs in manufacturing the device.

http://en.wikipedia.org/wiki/PCI_Express#PCI_Express_Mini_Card

Consider for example the BFL Single. Like you said, it is a square box with a PCB board with a heatsink. Imagine if you only had to produce 1 motherboard and had vertical mounts (like the old BTCFPGA design). This would make it a bit cheaper to produce quantities of add-on boards.

With fabs, with quantity there are discounts. There are also less processes to go through to put one together than say a BFL single which should have a higher overhead because it brings lots of components like the case and power supply.. The metal box is one piece and it has to be created n number of times how many boxes you want.

With add-on daughter card(s) it should be simpler, cheaper and easier to populate a rig with identical daughter cards (as needed).

If I am not mistaken, the BFL mini-rigs are modules but they are separate by connective cabling. They have to be professionally assembled as opposed to an end-user just popping the case open and adding more processing power into one of many east to install slots.

As to the cryptocurrencies, there are already merged mining pools, so all the bitcoin forks do not need to covered by vendors, and the non doubleSHA256s can't work alongside bitcoin for ASICs as needs no explanation.
[/quote]

[bitcoindaddy]

Here's where I begin to question. Modularizability sounds great and all, but it's not really the "game-changer" that you make it out to be. As far as I can tell, the BFL singles (only picture of an ASIC device we have currently) is essentially a box around a heatsink and a PCB+chips. In essence, it is almost as bare minimum as you can get. Aside from being able to buy some sort of massive copper-block with slots to insert boards into, you're not really going to get much more efficiency from some sort of modular design (and that's not very efficient really).

If the ASIC vendors can make smaller daughterboards that are alike and can be hosted on an internal bus or something like a mini-PCIe board, then that would cut costs in manufacturing the device.

http://en.wikipedia.org/wiki/PCI_Express#PCI_Express_Mini_Card

Consider for example the BFL Single. Like you said, it is a square box with a PCB board with a heatsink. Imagine if you only had to produce 1 motherboard and had vertical mounts (like the old BTCFPGA design). This would make it a bit cheaper to produce quantities of add-on boards.

With fabs, with quantity there are discounts. There are also less processes to go through to put one together than say a BFL single which should have a higher overhead because it brings lots of components like the case and power supply.. The metal box is one piece and it has to be created n number of times how many boxes you want.

With add-on daughter card(s) it should be simpler, cheaper and easier to populate a rig with identical daughter cards (as needed).

If I am not mistaken, the BFL mini-rigs are modules but they are separate by connective cabling. They have to be professionally assembled as opposed to an end-user just popping the case open and adding more processing power into one of many east to install slots.

As to the cryptocurrencies, there are already merged mining pools, so all the bitcoin forks do not need to covered by vendors, and the non doubleSHA256s can't work alongside bitcoin for ASICs as needs no explanation.

[/quote]
This has been discussed before. Using PCI instead of USB would make the boards more complicated and harder to design.

[bcpokey]

My numbers are correct bcpokey.  If you believe them to be wrong, then please show the math.

Apologies, I was thinking along different lines, your numbers are indeed correct.

Here's where I begin to question. Modularizability sounds great and all, but it's not really the "game-changer" that you make it out to be. As far as I can tell, the BFL singles (only picture of an ASIC device we have currently) is essentially a box around a heatsink and a PCB+chips. In essence, it is almost as bare minimum as you can get. Aside from being able to buy some sort of massive copper-block with slots to insert boards into, you're not really going to get much more efficiency from some sort of modular design (and that's not very efficient really).

If the ASIC vendors can make smaller daughterboards that are alike and can be hosted on an internal bus or something like a mini-PCIe board, then that would cut costs in manufacturing the device.

http://en.wikipedia.org/wiki/PCI_Express#PCI_Express_Mini_Card

Consider for example the BFL Single. Like you said, it is a square box with a PCB board with a heatsink. Imagine if you only had to produce 1 motherboard and had vertical mounts (like the old BTCFPGA design). This would make it a bit cheaper to produce quantities of add-on boards.

With fabs, with quantity there are discounts. There are also less processes to go through to put one together than say a BFL single which should have a higher overhead because it brings lots of components like the case and power supply.. The metal box is one piece and it has to be created n number of times how many boxes you want.

With add-on daughter card(s) it should be simpler, cheaper and easier to populate a rig with identical daughter cards (as needed).

If I am not mistaken, the BFL mini-rigs are modules but they are separate by connective cabling. They have to be professionally assembled as opposed to an end-user just popping the case open and adding more processing power into one of many east to install slots.

As to the cryptocurrencies, there are already merged mining pools, so all the bitcoin forks do not need to covered by vendors, and the non doubleSHA256s can't work alongside bitcoin for ASICs as needs no explanation.

[/quote]

I see, you meant an entire revamp to the whole design, rather than a simple modification to the existing design. As to what the specifics would be PCI vs. USB vs. ?? I would not want to get into, but I still question the efficiency gained by this method. As I understand the mini-rigs are going to be something along the lines of the size of a large trash can? This obviously would be unattractive for the average consumer. A smaller design such as the FPGA would be more attractive, but lose some of the savings you imagine, by being larger/bulkier to ship, requiring extra superfluous parts (1 board with spaces for daughter boards, power connector slots, etc.), increased cooling solutions for the near-housing of units, and still needing to make a large number of housing units (say 5 daughter boards per housing?). I'm not saying there would be NO savings, but simply that I don't know that it would be game changing enough to make a company that went this direction the sole survivor in the competitive ASIC market.

[SolarSilver]

This has been discussed before. Using PCI instead of USB would make the boards more complicated and harder to design.

You could create a PCI express card that has a USB hub on it and link that direct to the USB port of the existing card design. The power for the card can be drawn from a 6 or 8 pin PCIe power cord. (this is the same principle to hide the complexity of ADSL firmwares in PCI ADSL cards where they show up as a simple realtek network device)

A little bit more complicated then a stand alone USB card but not much. And much easier to 'house' into existing servers.

The only problem is the form factor, the current card from bASIC for example is 4" x 9" (10 cm x 23 cm) and with the
heatsink on the side, it's probably to bulky to fit into a PCI slot