[ActiveMining] The Official Active Mining Discussion Thread [Self-Moderated]

[VinceSamios]

FYI - in addition to the 7 questions I posted before, I asked Ken about the 10.488 TH miner he announced and this is the reply:

One more:

8. you said most recently that the 7-unit (7x 4U units) miner was going to be 10.488TH.  then you released that the boards for the 55nm chips will only hold 24 chips, which would mean 45.6GH per board, so it would need to have 33 boards per 4U box.  is that 10.488TH still the official figure?

The 10.488 TH/s System is 230 cards in 2 42U racks.  1 Base unit with 5 cards and 15 expansion cases daisy chained together.

Are those specs good or bad?

45gh per card seems pretty low density....

[kleeck]

FYI - in addition to the 7 questions I posted before, I asked Ken about the 10.488 TH miner he announced and this is the reply:

One more:

8. you said most recently that the 7-unit (7x 4U units) miner was going to be 10.488TH.  then you released that the boards for the 55nm chips will only hold 24 chips, which would mean 45.6GH per board, so it would need to have 33 boards per 4U box.  is that 10.488TH still the official figure?

The 10.488 TH/s System is 230 cards in 2 42U racks.  1 Base unit with 5 cards and 15 expansion cases daisy chained together.

Are those specs good or bad?

45gh per card seems pretty low density....

So, more than likely the "budget" option that we might see purchased (if we are selling these... I'm still not sure if we're just building the ActM operation with these chips.) will be having an affordable ($1K-$2K) base unit with expansion. This would provide 20 cards of hashing power or 912gh/s before any performance increases. My guess is that we will be able to clock these chips up a bit from 1.9gh/s with good stability and will be able to push this budget option up to and maybe slightly over 1TH/s.

But, yes, I agree. 2 42U racks for a little over 10TH/s seems a bit much. Imagine setting up a 1PH mine with those - you'd need 96 42U racks! Of course, that isn't what the 55nm offering is aimed at.

Good thing we have those 28nm FCICs on deck for bigger operations.

[drawingthesun]

Any specs yet for 28nm? I am hoping for 100GH/s at least, otherwise I'll never make a million from this.

[VinceSamios]

Any specs yet for 28nm? I am hoping for 100GH/s at least, otherwise I'll never make a million from this.

the gh/chip is very much beside the point - you pay per wafer, not per chip (except for packaging)

[minerpart]

But, yes, I agree. 2 42U racks for a little over 10TH/s seems a bit much. Imagine setting up a 1PH mine with those - you'd need 96 42U racks! Of course, that isn't what the 55nm offering is aimed at.

I'm not sure 96 42U racks would be as big a solution as you might think.



Each 42U rack is about the size of an upright fridge. Above you can see what 42 of these units would look like in a room about the size of half of a tennis court. These units can be packed much closer together than this too but that would depend on the cooling solutions we deploy. I'm guessing these chips will run cooler than our 28's so possibly the units could be packed closer than seen here.

EDIT - Ken has said we will have 60 of these 42U racks operating by end April. So about half as much again as what we see above.

[Entropy-uc]

FYI - in addition to the 7 questions I posted before, I asked Ken about the 10.488 TH miner he announced and this is the reply:

One more:

8. you said most recently that the 7-unit (7x 4U units) miner was going to be 10.488TH.  then you released that the boards for the 55nm chips will only hold 24 chips, which would mean 45.6GH per board, so it would need to have 33 boards per 4U box.  is that 10.488TH still the official figure?

The 10.488 TH/s System is 230 cards in 2 42U racks.  1 Base unit with 5 cards and 15 expansion cases daisy chained together.

Are those specs good or bad?

45gh per card seems pretty low density....

So, more than likely the "budget" option that we might see purchased (if we are selling these... I'm still not sure if we're just building the ActM operation with these chips.) will be having an affordable ($1K-$2K) base unit with expansion. This would provide 20 cards of hashing power or 912gh/s before any performance increases. My guess is that we will be able to clock these chips up a bit from 1.9gh/s with good stability and will be able to push this budget option up to and maybe slightly over 1TH/s.

But, yes, I agree. 2 42U racks for a little over 10TH/s seems a bit much. Imagine setting up a 1PH mine with those - you'd need 96 42U racks! Of course, that isn't what the 55nm offering is aimed at.

Good thing we have those 28nm FCICs on deck for bigger operations.

Try 200 racks.  And about 3 MW of power and cooling.

[kleeck]

But, yes, I agree. 2 42U racks for a little over 10TH/s seems a bit much. Imagine setting up a 1PH mine with those - you'd need 96 42U racks! Of course, that isn't what the 55nm offering is aimed at.

I'm not sure 96 42U racks would be as big a solution as you might think.



Each 42U rack is about the size of an upright fridge. Above you can see what 42 of these units would look like in a room about the size of half of a tennis court. These units can be packed much closer together than this too but that would depend on the cooling solutions we deploy. I'm guessing these chips will run cooler than our 28's so possibly the units could be packed closer than seen here.

Oh, I'm very aware of the dimensions of a 42U rack  . My point is that the space it would take is significant in comparison to it being used to mine just 1PH. In my company's major datacenter you would need a little over a full "room" to run this operation (10 rows with 9 42U racks per row). I'm not sure what the cost would be, so perhaps you are right and that setting up 96 42U racks for 1PH is actually cost effective.

There is also an option we may be overlooking. That of sytematically swapping the 55nm boards for 28nm boards for a fairly streamlined tech up. That'd be neat, but would require logistical analysis.

[mainline]

But, yes, I agree. 2 42U racks for a little over 10TH/s seems a bit much. Imagine setting up a 1PH mine with those - you'd need 96 42U racks! Of course, that isn't what the 55nm offering is aimed at.

I'm not sure 96 42U racks would be as big a solution as you might think.

https://cdn2.content.compendiumblog.com/uploads/user/c80ab324-8c02-491f-b1f1-5baeecc7e6e8/947b3f7b-5628-4a6d-a1c9-8180f0c37044/Image/8bee6249f5f013cee3492148f5433fc9/datacenter.jpg

Each 42U rack is about the size of an upright fridge. Above you can see what 42 of these units would look like in a room about the size of half of a tennis court. These units can be packed much closer together than this too but that would depend on the cooling solutions we deploy. I'm guessing these chips will run cooler than our 28's so possibly the units could be packed closer than seen here.

EDIT - Ken has said we will have 60 of these 42U racks operating by end April. So about half as much again as what we see above.

Well, considering what Ken has accomplished thus far...

http://s24.postimg.org/3kfmeyqbp/product1.jpg

[minerpart]

There is also an option we may be overlooking. That of sytematically swapping the 55nm boards for 28nm boards for a fairly streamlined tech up. That'd be neat, but would require logistical analysis.

I forgot you were the data centre guy, I guess you know more about this than me! If it's cost-effective then it certainly looks achievable over the next 3months. Your above suggested solution would make the swap-over super-quick. Someone should ask Ken if this is a possibility - is Vince compiling a list of questions for todays update?

[wasubii]

I'm guessing these chips will run cooler than our 28's so possibly the units could be packed closer than seen here.

Depends entirely on the specs of the 28nm chips and the skill set and experience of those designing it.

I would hope the 28nm are more efficient and run cooler (per GH) than the 55nm

[kleeck]

There is also an option we may be overlooking. That of sytematically swapping the 55nm boards for 28nm boards for a fairly streamlined tech up. That'd be neat, but would require logistical analysis.

I forgot you were the data centre guy, I guess you know more about this than me! If it's cost-effective then it certainly looks achievable over the next 3months. Your above suggested solution would make the swap-over super-quick. Someone should ask Ken if this is a possibility.

Try 200 racks.  And about 3 MW of power and cooling.

To clarify, Entropy is correct. I forgot to account for the fact that Ken said:

The 10.488 TH/s System is 230 cards in 2 42U racks.  1 Base unit with 5 cards and 15 expansion cases daisy chained together.

So the math would be:

Code:

1000/10.488               //1PH div by 1 55nm HashFast unit
=95.347                   //# of 55nm HashFast's needed
*2                        //#of 42U racks needed for each 10.488 HashFast miner
=190.694                 //#of 42U racks needed for 1PH operation)

So, in my last example it would take a little over two full rooms since each room maxes out at 90 racks.

[shaofis]

Kleeck: It won't require 190 racks... 10.488TH solution won't use 2Full racks. The question is can you make efficient use of the space.

Ken: I'd suggest you consider 1 Head + 9 Expansion per rack and keep it to a system per rack. This would be easier to manage. It would give you 140 cards per rack or 6.3TH.

Or roughly 160 racks.

[kleeck]

Kleeck: It won't require 190 racks... 10.488TH solution won't use 2Full racks. The question is can you make efficient use of the space.

Ken: I'd suggest you consider 1 Head + 9 Expansion per rack and keep it to a system per rack. This would be easier to manage. It would give you 140 cards per rack or 6.3TH.

Or roughly 160 racks.

I still think that's a lot of racks for 1PH, but as I said before, if we can keep all of cases in place while upgrading units to the 28nm boards then there could be a lot of benefit to ramping up to this size, if the price is right.

[VinceSamios]

There is also an option we may be overlooking. That of sytematically swapping the 55nm boards for 28nm boards for a fairly streamlined tech up. That'd be neat, but would require logistical analysis.

I forgot you were the data centre guy, I guess you know more about this than me! If it's cost-effective then it certainly looks achievable over the next 3months. Your above suggested solution would make the swap-over super-quick. Someone should ask Ken if this is a possibility - is Vince compiling a list of questions for todays update?

Not this time round - most of the most important questions were answered and ken has given a lot of substantive information. I might start collecting questions again if things get out of hand again.

[minerpart]

Kleeck: It won't require 190 racks... 10.488TH solution won't use 2Full racks. The question is can you make efficient use of the space.

Ken: I'd suggest you consider 1 Head + 9 Expansion per rack and keep it to a system per rack. This would be easier to manage. It would give you 140 cards per rack or 6.3TH.

Or roughly 160 racks.

Excellent suggestion, I'm sure his engineering guys will be thinking about how to use the least space (especially if we are using a third party data centre). But if you see no reply from Ken about this think about putting it into a question for him for the update next week/PM'ing him. We should all be giving Ken our best advice right now as he needs to get the execution of our farm right first time.

[mainline]

...
Not this time round - most of the most important questions were answered and ken has given a lot of substantive information. I might start collecting questions again if things get out of hand again.

Excelent.  Give people a chance to digest the awesome news.
BTW, how's that verification process going?  Has Ken uploaded the RAM to his server yet?

*Lol:

http://s3.postimg.org/lyb6e95c3/Capture.jpg

[minerpart]

Vince I have one for you for next week:

Does the ACtM engineering team have previous experience of deploying a high power-draw data centre of substantial size (100+ 42U racks)? If not, would you consider employing a project manager with such experience to manage the complicated process of getting our farm online?

[EduardoDeCastro]

@equipoise:  Just laughing at the flippers.  I have no money in this.  If it gets to be as fun as Labcoin was for a while, I might trade.

*Notice the last buy.  wut

[minerpart]

equip - please don't quote the Trolls, if you do you will be ignored also. This one has just posted under one name (mainline) and mistakenly answered under another (eduardo) - it is the same guy. Thanks.

[EduardoDeCastro]

equip - please don't quote the Trolls, if you do you will be ignored also. This one has just posted under one name (mainline) and mistakenly answered under another (eduardo) - it is the same guy. Thanks.

My idiot friend:  EduardoDeCastro account was made for you - you insisted that I was Eduardo de Castro of HashFast.
Got to keep tinfoilers happy, amiright?