Pages:
Author

Topic: 1GH/s, 20w, $500 — Butterflylabs, is it a scam? - page 34. (Read 123109 times)

donator
Activity: 1218
Merit: 1079
Gerald Davis
They have emailed me saying that the Rig Box has additional performance improvements, being 32 clustered boards instead of 1 single, that enable it to reach 54 GH/s.

Does that make sense to you?

Do 32 GPU have 54x the performance of 1 GPU?
Do 32 CPU have 54x the perofrmance of 1 GPU?

If 1 board was capable of 1.7GH wouldn't it make sense to just apply those magical unicorn optimization and make the single board 1.7GH too.  They could sell it for 70% more right?  Unless 70% higher revenue per sale is a bad thing.

Yes it makes sense that 32 specialized boards can run together and achieve improvements you wouldn't otherwise get with a single board.

On a problem like SHA-256 hashing which is perfectly parallel?  Really?  Care to hypothesize how that works since you seem convinced.

SHA-256 is a perfectly parallel problem.  A single hash takes only a billionth of a second.  There is no need for boards to work together.  No need for cores to even work together to solve a problem which takes a single clock cycle to solve.   Each hash is independent of the others thus there is no need for cores to even know about each other. 

So to make a cluster faster they aren't working together rather working simultaneously on the same problem.  Like two, or 8, or 32 workers oblivious to the fact that others are also working on the same problem.

32 independent workers don't accomplish 54x the work of a single worker. 
hero member
Activity: 518
Merit: 500
Quote
Still we will know in 2 weeks.  Although their website says 4-6 weeks just like it did 3 weeks ago so my guess is in 2 weeks pre-orders will be shipping in 4-6 weeks.

You know, having been involved in 2 startups,  I can imagine they have other things to do right now besides updating their website 3x per week and keeping a 30 page thread full of conspiracy theorists happy, considering they are only about to actually launch their first product.  It also makes complete sense they dont want to spill all the beans here yet. Cant blame them for wanting to build some anticipation for their official unveiling.
hero member
Activity: 742
Merit: 500
They have emailed me saying that the Rig Box has additional performance improvements, being 32 clustered boards instead of 1 single, that enable it to reach 54 GH/s.

Does that make sense to you?

Do 32 GPU have 54x the performance of 1 GPU?
Do 32 CPU have 54x the perofrmance of 1 GPU?

If 1 board was capable of 1.7GH wouldn't it make sense to just apply those magical unicorn optimization and make the single board 1.7GH too.  They could sell it for 70% more right?  Unless 70% higher revenue per sale is a bad thing.

Yes it makes sense that 32 specialized boards can run together and achieve improvements you wouldn't otherwise get with a single board.

You are going to need to explain that a little further.

Best guess of what he's going for here:

Scenario 1: You're using FPGAs. Each board has one FPGA that you're utilizing 80% of. There's not enough room to fit anything useful in that remaining 20% unless you're using a large array of many FPGAs in which case some logic could be crammed into the extra 20% that might be useful in tandem with other "wasted" space.

Scenario 2: The data on the site isn't entirely accurate and the cards in the larger rig aren't *exactly* the same as the single rigs. What I said in Scenario 1 about wasted FPGA gates also applies to wasted PCB space.
sr. member
Activity: 448
Merit: 250
They have emailed me saying that the Rig Box has additional performance improvements, being 32 clustered boards instead of 1 single, that enable it to reach 54 GH/s.

Does that make sense to you?

Do 32 GPU have 54x the performance of 1 GPU?
Do 32 CPU have 54x the perofrmance of 1 GPU?

If 1 board was capable of 1.7GH wouldn't it make sense to just apply those magical unicorn optimization and make the single board 1.7GH too.  They could sell it for 70% more right?  Unless 70% higher revenue per sale is a bad thing.

Yes it makes sense that 32 specialized boards can run together and achieve improvements you wouldn't otherwise get with a single board.

You are going to need to explain that a little further.
full member
Activity: 180
Merit: 100
mistaken for gribble since 2011
They have emailed me saying that the Rig Box has additional performance improvements, being 32 clustered boards instead of 1 single, that enable it to reach 54 GH/s.

Does that make sense to you?

Do 32 GPU have 54x the performance of 1 GPU?
Do 32 CPU have 54x the perofrmance of 1 GPU?

If 1 board was capable of 1.7GH wouldn't it make sense to just apply those magical unicorn optimization and make the single board 1.7GH too.  They could sell it for 70% more right?  Unless 70% higher revenue per sale is a bad thing.

Yes it makes sense that 32 specialized boards can run together and achieve improvements you wouldn't otherwise get with a single board.
donator
Activity: 1218
Merit: 1079
Gerald Davis
Then again, if this where a scam, why would they come up with numbers that dont make sense on first sight? Lets just 2 weeks.

Well sadly they seem to be "good" to some people.

Still we will know in 2 weeks.  Although their website says 4-6 weeks just like it did 3 weeks ago so my guess is in 2 weeks pre-orders will be shipping in 4-6 weeks.
hero member
Activity: 518
Merit: 500
Then again, if this where a scam, why would they come up with numbers that dont make sense on first sight? Lets just 2 weeks.
donator
Activity: 1218
Merit: 1079
Gerald Davis
They have emailed me saying that the Rig Box has additional performance improvements, being 32 clustered boards instead of 1 single, that enable it to reach 54 GH/s.

Does that make sense to you?

Do 32 GPU have 54x the performance of 1 GPU?
Do 32 CPU have 54x the perofrmance of 1 GPU?

If 1 board was capable of 1.7GH wouldn't it make sense to just apply those magical unicorn optimization and make the single board 1.7GH too.  They could sell it for 70% more right?  Unless 70% higher revenue per sale is a bad thing.
hero member
Activity: 518
Merit: 500
I stumbled upon his:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.148.7900&rep=rep1&type=pdf

It goes way above my head, but it proposes some mechanisms to allow rather extreme performance gains compared to traditional FPGA for SHA hashing.  The abstract:

Abstract—This paper presents a new set of techniques for hard-
ware implementations of Secure Hash Algorithm (SHA) hash func-
tions. These techniques consist mostly in operation rescheduling
and hardware reutilization, therefore, significantly decreasing the
critical path and required area. Throughputs from 1.3 Gbit/s to
1.8 Gbit/s were obtained for the SHA implementations on a Xilinx
VIRTEX II Pro. Compared to commercial cores and previously
published research, these figures correspond to an improvement in
throughput/slice in the range of 29% to 59% for SHA-1 and 54%
to 100% for SHA-2. Experimental results on hybrid hardware/soft-
ware implementations of the SHA cores, have shown speedups up
to 150 times for the proposed cores, compared to pure software im-
plementations.


Can someone with a higher clocked brain than me glance over that, and tell if its feasible BFL did a similar thing?

edit: seems they refer to FPGAs as "hybrid".  So scratch the 150x, but I still wonder if these kinds of optimization have already found their way to our  DIY "garage FPGA builders" (no offense guys, I have the greatest respect for you!).
full member
Activity: 180
Merit: 100
mistaken for gribble since 2011

Wich additional questions regarding the upcoming Bitforce Box and Rigbox would be interesting for you ?


I have a big one...how does 2+2=5? What system would see a 70% increase in performance through repackaging? What will the price points be on the units? Why won't they let the guy that lives across the street come and see one in action? I doubt his goal is industrial espionage, and I am sure he is happy to sign an NDA (as long as he is allowed to tell us if it actually works) to reassure them.

They have emailed me saying that the Rig Box has additional performance improvements, being 32 clustered boards instead of 1 single, that enable it to reach 54 GH/s.
sr. member
Activity: 448
Merit: 250

Wich additional questions regarding the upcoming Bitforce Box and Rigbox would be interesting for you ?


I have a big one...how does 2+2=5? What system would see a 70% increase in performance through repackaging? What will the price points be on the units? Why won't they let the guy that lives across the street come and see one in action? I doubt his goal is industrial espionage, and I am sure he is happy to sign an NDA (as long as he is allowed to tell us if it actually works) to reassure them.
sr. member
Activity: 410
Merit: 252
Watercooling the world of mining
And maybe to bring this thread back onto a more reasonable path.

Let us just asume the development by BFL is no scam.( personally I think its technically mostly possible)

Wich additional questions regarding the upcoming Bitforce Box and Rigbox would be interesting for you ?

Maybe we will be fed with some more details here.
sr. member
Activity: 448
Merit: 250
So one company is SCI  http://www.sciengines.com

They use high numbers of standard FPGA's.

Some new field they apply fore are also Scientific highly paralled computing.

I have to look up the other companys name,but i guess SCI is the major one.
 


 'fore' is a nautical term referring to the front of a vessel (like fore and aft)...not that it really matters

Or what you yell when your golf ball is about to hit someone!

Which is kind of funny...you yell 'Fore!' to get someone to look in front of them, but odds are you are smacking golf balls at someone with their back to you...unless you are like me and putting it in the next fairway on half your shots.
hero member
Activity: 490
Merit: 500
So one company is SCI  http://www.sciengines.com

They use high numbers of standard FPGA's.

Some new field they apply fore are also Scientific highly paralled computing.

I have to look up the other companys name,but i guess SCI is the major one.
 


 'fore' is a nautical term referring to the front of a vessel (like fore and aft)...not that it really matters

Or what you yell when your golf ball is about to hit someone!
sr. member
Activity: 448
Merit: 250
So one company is SCI  http://www.sciengines.com

They use high numbers of standard FPGA's.

Some new field they apply fore are also Scientific highly paralled computing.

I have to look up the other companys name,but i guess SCI is the major one.
 


This isn't grammar trolling, just a friendly correction; 'for' is the word you are looking for, 'fore' is a nautical term referring to the front of a vessel (like fore and aft)...not that it really matters...I just want to see more of these crazy code-breaking machines!
sr. member
Activity: 410
Merit: 252
Watercooling the world of mining
So one company is SCI http://www.sciengines.com

They use high numbers of standard FPGA's.

Some new field they apply for are also scientific highly paralled computing.

I have to look up the other companys name,but i guess SCI is the major one.
 
Edit:Some typos fixed Cheesy
donator
Activity: 1218
Merit: 1079
Gerald Davis
Just to remind :

Here in germany there are two companys exclusivly selling hashing devices in 19" racks from 15000€ up to 150000+ € a piece
for bruteforce codebreaking (military intelligence,Data security research, ....).

So i guess there is actually a market altough i consider it small.
Interesting, do you have a url for that?

However, if that is true that makes this thing even more dubious than it already is...  Huh

Not sure if this is the one he is thinking of but here is one example:
http://www.digitalintelligence.com/products/rack-a-tacc/

240K AES password attempts per second = $19K.

http://www.digitalintelligence.com/products/rack-a-tacc/images/rack-a-tacc_benchmarks_med.jpg

Here is the thing - this device is customizable and has high speed interface (dual firewire 800) so it can brute force many different algorithms to decrypt sensitive documents.  Given SHA-256 isn't used in encryption there is much less demand for SHA password recovery.
legendary
Activity: 1666
Merit: 1057
Marketing manager - GO MP
Just to remind :

Here in germany there are two companys exclusivly selling hashing devices in 19" racks from 15000€ up to 150000+ € a piece
for bruteforce codebreaking (military intelligence,Data security research, ....).

So i guess there is actually a market altough i consider it small.
Interesting, do you have a url for that?

However, if that is true that makes this thing even more dubious than it already is...  Huh
sr. member
Activity: 410
Merit: 252
Watercooling the world of mining
Just to remind :

Here in germany there are two companys exclusivly selling hashing devices in 19" racks from 15000€ up to 150000+ € a piece
for bruteforce codebreaking (military intelligence,Data security research, ....).

So i guess there is actually a market altough i consider it small.
hero member
Activity: 518
Merit: 500
There is no market demand for that level of SHA-256 hashing outside of Bitcoin.  Think about it for a second.  Even a crappy CPU can perform tens of millions of hashes per second.    while SHA is used for things like keeping passwords secure even Google doesn't need the ability to process 1 billion logins every second. 

6 of these boards could (well with enough bandwidth which they lack) validate a Google account login for every single human on the planet in less than 1 seconds.

SHA-256 has other uses (hashing for signatures, resource organization, etc) but likewise there are no applications (military, financial, or otherwise) that need 1 billion hashes per second.

Damn it. Dreams ... crushed. Then this looks to be a scam yet again. Unless they have found some FPGA which improves efficiency, hashrate and cost much over standard FPGAs on here. Or maybe they are using ASICs but much more optimized than GPU ASIC like 5970s etc. ?
Pages:
Jump to: