ASICMINER: Entering the Future of ASIC Mining by Inventing It

[necro_nemesis]

Possibly due to you're oversimplified conclusions being incorrect judging by your statements.

[1714022415]

1714022415

[1714022415]

1714022415

[1714022415]

1714022415

[jimmothy]

Possibly due to you're oversimplified conclusions being incorrect judging by your statements.

My "oversimplified" conclusion that nobody has yet to demonstrate an asic working significantly beyond advertised efficiency is why you brought up an example of bitmain not surpassing advertised efficiency? Thanks for supporting my point.

Again if I'm incorrect ease provide a single shred of evidence that suggests it is possible to undervolt every asic significantly below advertised limit without diminishing returns.

[klondike_bar]

Either you cant read a chart or you are assuming current asics run at a point to the extreme left of that chart

Yes this is exactly what I am assuning due to the fact that no asic manufacturer has advertised below 0.5w/gh (bitfury excluded) and nobody has been able to significantly undervolt below advertised limit.

which would also imply those other asics can be overclocked by factor of 2-3x. Here is hint: almost none can. KnC, HF, CT, bitmain, Bitmine .. you'd be lucky to hit advertised speeds, let alone 2-3x more.

Either way, welcome to my ignore list.

Heres a hint:

knc can be overclocked 1.7 times  (700gh/s vs advertised 40gh/s)
Hashfast can be overclocked 2 times (800gh/s vs 400gh advertised)
And bitmain can be overclocked 2 times (200gh vs 100gh advertised)

All of which have no problem hashing at advertised speeds.

Again, got any evidence that other manufacturers are not on the far left side of that chart?

Bitmain advertises 180GH. it can overclock to to ~205GH

[jimmothy]

Bitmain advertises 180GH. it can overclock to to ~205GH

180gh is after bitmain overclocked the chips.

Since day 1 they advertised 1.5gh per chip at 0.68w/gh before they overclocked to 3gh per chip at 1.2w/gh.

[necro_nemesis]

Again if I'm incorrect ease provide a single shred of evidence that suggests it is possible to undervolt every asic significantly below advertised limit without diminishing returns.

Why am I being asked to go through the exercise of providing evidence to prove something I didn't claim? If you go back to what I stated we don't know the overall performance numbers of the chip and was requesting these values. I didn't claim that for "every asic" and for that matter more specifically the BE200 could benefit from the reduction as we simply don't have the evidence to do so yet. I also didn't ever infer that hash rate remained the same when under volting and under clocking.

[xhomerx10]

Caveat Emptor!  Information asymmetry detected.

ITs not meant as promotion for BFL, Id be the last person to suggest you order anything from there. But the chart is interesting and relevant to earlier discussions about how you can scale power efficiency of every ASIC. Even if you take those numbers with a table spoon of salt, it nicely illustrates my point.  There is about a factor 3x difference in power efficiency depending what voltage/frequency point you pick. That spread is not going to be vastly different for any other bitcoin asic.

 I'm going to pretend it is relevant and probably valid for other ASICs but the fact of the matter is, the price of the devices are set many months before those cute little graphics come out.  Your money has been spent and if the device you spent it on wont make a positive return for you running at the right-hand side of the graph, it certainly wont make money running on the left-hand side.  Ergo,  the plot is meaningless to anyone but the salesman and the scientist.
 Maybe I am missing something.  What do you think your options are now that you have this information?

[Puppet]

Caveat Emptor!  Information asymmetry detected.

ITs not meant as promotion for BFL, Id be the last person to suggest you order anything from there. But the chart is interesting and relevant to earlier discussions about how you can scale power efficiency of every ASIC. Even if you take those numbers with a table spoon of salt, it nicely illustrates my point.  There is about a factor 3x difference in power efficiency depending what voltage/frequency point you pick. That spread is not going to be vastly different for any other bitcoin asic.

 I'm going to pretend it is relevant and probably valid for other ASICs but the fact of the matter is, the price of the devices are set many months before those cute little graphics come out.  Your money has been spent and if the device you spent it on wont make a positive return for you running at the right-hand side of the graph, it certainly wont make money running on the left-hand side.  Ergo,  the plot is meaningless to anyone but the salesman and the scientist.
 Maybe I am missing something.  What do you think your options are now that you have this information?

You're missing the fact there is/was a discussion going about scaling of power efficiency.  For every asic you could draw a similar plot, and power efficiency can be traded for performance per chip. Something obvious to anyone who knows anything thing about asics, but apparently impossible to grasp for jimmothy.  As for pricing, current market prices are so far detached from variable production cost, that for now, this is a moot point. Vendors will charge whatever fools are willing to pay, there is basically no correlation with production costs yet. That will only happen once price per GH has dropped another order of magnitude or so.

As for my "options". The only sensible option is the same one since the very first asic was announced: stay away from them, just invest in bitcoin. Doing that has yielded me 100% BTC denomnated ROI or ~10000% when expressed in dollar. How many miners can say that?

[jimmothy]

For every asic you could draw a similar plot, and power efficiency can be traded for performance per chip. Something obvious to anyone who knows anything thing about asics, but apparently impossible to grasp for jimmothy.

Of course you can draw a similar plot with every asic.

I never debated that.

You are arguing that every asic is capable of higher efficiency than any point on the plot.

So either you are assuming that the asic manufacturers were too lazy/incompetent to do proper testing to create their plot, or they decided to not advertise their max efficiency.

[Puppet]

"Again you need evidence that lowering voltage to increase efficiency is possible"

 Quod est demonstrandum. Back to ignore.

[jimmothy]

"Again you need evidence that lowering voltage to increase efficiency is possible"

 Quod est demonstrandum. Back to ignore.

You still have yet to prove any asic works below advertised minimum voltage. Let alone undervolted enough to be twice as efficient as advertised.

You also have yet to explain why knc/hashfast/cointerra would refuse to advertise their chips full capabilities.

[necro_nemesis]

At some point you'll realize that your valiant efforts to calculate best value are fruitless. Miner's are being given only what little incentive they need to purchase and there will come a time if you're to stand a chance you must align yourself with ASIC makers if you're to remain in mining at all. You either invest in BTC or you invest in the mining equipment manufacturers. Sooner or later the little guy is out.

[klondike_bar]

At some point you'll realize that your valiant efforts to calculate best value are fruitless. Miner's are being given only what little incentive they need to purchase and there will come a time if you're to stand a chance you must align yourself with ASIC makers if you're to remain in mining at all. You either invest in BTC or you invest in the mining equipment manufacturers. Sooner or later the little guy is out.

bullsh*t. big miners will be taxed. The little guy can run a few kW of equipment without being taxed as a bitcoin business - so the hashrate will always be able to distribute.

even in GPU days there were people with warehouses and attics crammed full of GPUs

[topminingcontracts]

Bitmain advertises 180GH. it can overclock to to ~205GH

180gh is after bitmain overclocked the chips.

Since day 1 they advertised 1.5gh per chip at 0.68w/gh before they overclocked to 3gh per chip at 1.2w/gh.

My two cents, Antminer S2 uses the same ASIC as Antminer S1

Antminer S1 = 2 W x GH = 2.8 GH x Chip = 180 GH with 64 Chips
Antminer S2 = 1 W x GH = 1.6 GH x Chip = 1000 GH with 640 Chips

Regards

Juan

[jimmothy]

Bitmain advertises 180GH. it can overclock to to ~205GH

180gh is after bitmain overclocked the chips.

Since day 1 they advertised 1.5gh per chip at 0.68w/gh before they overclocked to 3gh per chip at 1.2w/gh.

My two cents, Antminer S2 uses the same ASIC as Antminer S1

Antminer S1 = 2 W x GH = 2.8 GH x Chip = 180 GH with 64 Chips
Antminer S2 = 1 W x GH = 1.6 GH x Chip = 1000 GH with 640 Chips

Regards

Juan

S1 was always advertised as overclocked. Chips were  advertised to reach up to 0.68w/gh at 1.5gh per chip before the s1 was even released.

[xhomerx10]

Caveat Emptor!  Information asymmetry detected.

ITs not meant as promotion for BFL, Id be the last person to suggest you order anything from there. But the chart is interesting and relevant to earlier discussions about how you can scale power efficiency of every ASIC. Even if you take those numbers with a table spoon of salt, it nicely illustrates my point.  There is about a factor 3x difference in power efficiency depending what voltage/frequency point you pick. That spread is not going to be vastly different for any other bitcoin asic.

 I'm going to pretend it is relevant and probably valid for other ASICs but the fact of the matter is, the price of the devices are set many months before those cute little graphics come out.  Your money has been spent and if the device you spent it on wont make a positive return for you running at the right-hand side of the graph, it certainly wont make money running on the left-hand side.  Ergo,  the plot is meaningless to anyone but the salesman and the scientist.
 Maybe I am missing something.  What do you think your options are now that you have this information?

You're missing the fact there is/was a discussion going about scaling of power efficiency.  For every asic you could draw a similar plot, and power efficiency can be traded for performance per chip. Something obvious to anyone who knows anything thing about asics, but apparently impossible to grasp for jimmothy.  As for pricing, current market prices are so far detached from variable production cost, that for now, this is a moot point. Vendors will charge whatever fools are willing to pay, there is basically no correlation with production costs yet. That will only happen once price per GH has dropped another order of magnitude or so.

As for my "options". The only sensible option is the same one since the very first asic was announced: stay away from them, just invest in bitcoin. Doing that has yielded me 100% BTC denomnated ROI or ~10000% when expressed in dollar. How many miners can say that?

You're right.  Not many miners can say they have made that kind of return.  I would suspect if they are still in the game, they are simply hoping to get a small percentage of the coins back that they forked over for the latest generation of pre-orders and wondering what to do with all the waste heat when they finally arrive.

[antirack]

What this likely means for mining equipment providers is that in the intermediate term the ability to migrate and maintain beats efficiency. That's why developments by Allied Control are so interesting which focus on ease of deployment.

Here you're opening a whole new can of worms. I think that would depends on your OPEX. CAPEX is high with the tanks but if maintenance (OPEX) is low enough, it would be a decent target for countries where labor and space are expensive (but what of electricity costs, what of heat dissipation power usage?). If OTOH you have cheap labor and space, the tanks aren't looking that good anymore.

Also, unless there is a way to stack them, they won't really help you improve the density of your deployment (they have other issues with connectors, plumbing, etc) vs a regular datacenter (more on that after the end of June, I guess). I'm not really sold on the tanks, I think the idea is neat, but the numbers don't convince me so far (probably need more data).

The OPEX and CAPEX of immersion cooling is lower. Unless you are talking about some home miner of course, not the 1.2MW container mining units.

Isn't immersion cooling incredible expensive?
 
No it's not. 3M fluids and immersion cooling have a reputation of being more expensive. What most people forget is that you only need very little fluid for a lot of heat, and that a tank full with hundreds of boards costs a lot less than dozens of racks with machines with cases or fans.

To give you an example, we cool 4kW (4000W) in a small space of 1 liter volume with only 200cc of fluid. Even heatsinks and fans for hardware with 4000W will cost more than just 200cc of fluid. And that is just one side of the equation. To make a long story short, it is less expensive to build and to run. That's the value proposition of passive 2-phase immersion cooling.

If you factor in device cooling (heatsinks, cold plates, pumps, hoses, radiators, speciality racks, air conditions, etc), it costs less to build a container with a few tanks from the first day. But where immersion really shines is when you start reusing all of this for your next-gen hardware, you basically get everything for free, you already own everything you will ever need. A single person can deploy hundreds of boards within hours, instead of days or weeks.

The reduced electricity cost for cooling is a bonus on top of all that.

Source:
https://docs.google.com/a/allied-control.com/document/d/1zQwlGoDfHgqlPFlLuUviKzkSUepWMTKcwGaNI_nqNjg/edit

they have other issues with connectors, plumbing, etc

A 200kW tank for the AM gen-3 boards has exactly one water hose in and one water hose out. That is for 32,000 AM BE200 chips. No plumbing issues.

I am totally sold

[TheFuneral]

How can dividends not exist if ASICMiner had cooling farms running in China?

[bitcoin.newsfeed]

How can dividends not exist if ASICMiner had cooling farms running in China?

Right? And if Friedcat has chips in hand for 5 weeks+ ? What is he doing again ? So he ordered 850k chips for April ... is May, where are they ?

[antirack]

How can dividends not exist if ASICMiner had cooling farms running in China?

Right? And if Friedcat has chips in hand for 5 weeks+ ? What is he doing again ? So he ordered 850k chips for April ... is May, where are they ?

First of all, where does it say he has them on his hand for 5 weeks? Second, they are in the the hand of the packaging factory and the board designers. Wafers are just that, wafers. They need to be cut into dies, then packaged, then tested, and all that. And of course he is also taking his well deserved nap. Just as any good cat would, at this time of the day. No matter in China or otherwise.

Hope that helps.

[KS]

What this likely means for mining equipment providers is that in the intermediate term the ability to migrate and maintain beats efficiency. That's why developments by Allied Control are so interesting which focus on ease of deployment.

Here you're opening a whole new can of worms. I think that would depends on your OPEX. CAPEX is high with the tanks but if maintenance (OPEX) is low enough, it would be a decent target for countries where labor and space are expensive (but what of electricity costs, what of heat dissipation power usage?). If OTOH you have cheap labor and space, the tanks aren't looking that good anymore.

Also, unless there is a way to stack them, they won't really help you improve the density of your deployment (they have other issues with connectors, plumbing, etc) vs a regular datacenter (more on that after the end of June, I guess). I'm not really sold on the tanks, I think the idea is neat, but the numbers don't convince me so far (probably need more data).

The OPEX and CAPEX of immersion cooling is lower. Unless you are talking about some home miner of course, not the 1.2MW container mining units.

Isn't immersion cooling incredible expensive?
 
No it's not. 3M fluids and immersion cooling have a reputation of being more expensive. What most people forget is that you only need very little fluid for a lot of heat, and that a tank full with hundreds of boards costs a lot less than dozens of racks with machines with cases or fans.

To give you an example, we cool 4kW (4000W) in a small space of 1 liter volume with only 200cc of fluid. Even heatsinks and fans for hardware with 4000W will cost more than just 200cc of fluid. And that is just one side of the equation. To make a long story short, it is less expensive to build and to run. That's the value proposition of passive 2-phase immersion cooling.

If you factor in device cooling (heatsinks, cold plates, pumps, hoses, radiators, speciality racks, air conditions, etc), it costs less to build a container with a few tanks from the first day. But where immersion really shines is when you start reusing all of this for your next-gen hardware, you basically get everything for free, you already own everything you will ever need. A single person can deploy hundreds of boards within hours, instead of days or weeks.

The reduced electricity cost for cooling is a bonus on top of all that.

Source:
https://docs.google.com/a/allied-control.com/document/d/1zQwlGoDfHgqlPFlLuUviKzkSUepWMTKcwGaNI_nqNjg/edit

they have other issues with connectors, plumbing, etc

A 200kW tank for the AM gen-3 boards has exactly one water hose in and one water hose out. That is for 32,000 AM BE200 chips. No plumbing issues.

I am totally sold

Well, OPEX would be about 20-30% cheaper for me in a DC for a 240KW farm (includes electricity and rack costs). It gets slightly worse if you include CAPEX (but I haven't because I'm not sure of the useful lifetime of the hardware).

With 120 racks (2KW per rack), I would also get a better rate than the one I currently have (I could theoretically also build my own, but I would still be smaller than the ones with multiple DCs and they would still have better yet electric rates - a big factor here is the electricity "transport" cost, also there'd be all the associated building management "stuff" to deal with).

Setup would be more expensive in a DC than for a tank, but the OPEX difference is so big that it wouldn't matter.

I'm in a densely populated, high electric cost area. The tanks would look much better if I lived, say, in Washington state.