Bitcoin Forum
December 11, 2016, 04:22:58 AM *
News: To be able to use the next phase of the beta forum software, please ensure that your email address is correct/functional.
 
   Home   Help Search Donate Login Register  
Pages: « 1 2 3 4 5 6 7 8 9 [10] 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 ... 81 »
  Print  
Author Topic: 1GH/s, 20w, $700 (was $500) — Butterflylabs, is it for real? (Part 2)  (Read 138978 times)
Inaba
Legendary
*
Offline Offline

Activity: 1260



View Profile WWW
December 12, 2011, 02:46:56 AM
 #181

I put the testing off until tomorrow as I was not able to make it downtown today.  I will be downtown tomorrow so it shouldn't be a problem.

If you're searching these lines for a point, you've probably missed it.  There was never anything there in the first place.
Advertised sites are not endorsed by the Bitcoin Forum. They may be unsafe, untrustworthy, or illegal in your jurisdiction. Advertise here.
1481430178
Hero Member
*
Offline Offline

Posts: 1481430178

View Profile Personal Message (Offline)

Ignore
1481430178
Reply with quote  #2

1481430178
Report to moderator
1481430178
Hero Member
*
Offline Offline

Posts: 1481430178

View Profile Personal Message (Offline)

Ignore
1481430178
Reply with quote  #2

1481430178
Report to moderator
1481430178
Hero Member
*
Offline Offline

Posts: 1481430178

View Profile Personal Message (Offline)

Ignore
1481430178
Reply with quote  #2

1481430178
Report to moderator
Turbor
Legendary
*
Offline Offline

Activity: 1008


BitMinter


View Profile WWW
December 12, 2011, 06:11:16 PM
 #182

I put the testing off until tomorrow as I was not able to make it downtown today.  I will be downtown tomorrow so it shouldn't be a problem.


Awaiting your report, Sir  Tongue

Inaba
Legendary
*
Offline Offline

Activity: 1260



View Profile WWW
December 12, 2011, 07:28:53 PM
 #183

I just sent an email to BFL to see if they can meet up at 16:30 and if they are in a position to allow me to take a unit over night to have my way with it.

Hopefully they are to the point where they can release it to me and I can test it properly.

If you're searching these lines for a point, you've probably missed it.  There was never anything there in the first place.
Inaba
Legendary
*
Offline Offline

Activity: 1260



View Profile WWW
December 13, 2011, 02:04:42 AM
 #184

Just got back from testing. I need to compile some of the results, eat dinner and compose a post.  It will probably be a couple hours.

If you're searching these lines for a point, you've probably missed it.  There was never anything there in the first place.
worldinacoin
Hero Member
*****
Offline Offline

Activity: 658



View Profile WWW
December 13, 2011, 02:37:50 AM
 #185

Wow, finally after a long wait we are getting something.
ssateneth
Legendary
*
Offline Offline

Activity: 1288



View Profile
December 13, 2011, 03:15:50 AM
 #186

Eagerly waiting for results

coretechs
Donator
Sr. Member
*
Offline Offline

Activity: 362



View Profile
December 13, 2011, 03:52:34 AM
 #187

BFL homepage appears to be updated:

Performance: 832 Mega Hash / s @ 80w

Price says $599 again

http://bitcoindoc.com - The Rise and Rise of Bitcoin | http://nxtportal.org - Nxt blockchain explorer
Inaba
Legendary
*
Offline Offline

Activity: 1260



View Profile WWW
December 13, 2011, 04:00:07 AM
 #188

Ok, so let me start off by giving you the environment and setup details.

Late this afternoon, I was sent the latest codeset to the modified Ufasoft miner for review.  After trying to compile it again, with the same results and trying several different scenarios, I was still unable to compile the software.  I dug a little deeper into the error and consulted with BFL and we determined that the Ufasoft codebase is not 64 bit ready, and thus would not compile on my laptop.  At this point, it was either postpone the test or make do with what we can until either a) Ufasoft is fixed to support 64 bit or b) I reinstall and use 32 bit Ubuntu.  Neither of these options were something that was time feasible this evening.

As such, I elected to do the following:

Take the code set and compile it on BFL's laptop, tar up the directory and keep a copy for later review, both source and binary.  With a working binary, albeit on BFLs laptop, we proceeded to conduct the testing.

First, we connected the miner to the pool as one would normally connect, though we were connected to a debug version of the getwork server and I observed the debug output as the miner connected and started mining.  The debug output was rather enlightening in so far as some optimizations to speed up and decrease stales in the Ufasoft code became readily apparent.  I will not go into detail in this post, but I am willing to discuss it elsewhere and/or at a later time.  Suffice to say that the connection information appeared to be perfectly legitimate, coming from the appropriate IP. 

After letting the miner mine for approximately 15 minutes, we disconnected, hooked up the Kill-A-Watt meter, then proceeded to connect BFL's laptop directly to one of the getwork servers on a non-routable IP address.  I then made sure the routing table on BFL's laptop was cleared and no network connections, save for the non-routable ETH0 was operational.  We then connected to the pool and began mining again.  Because we had effectively reduced the latency to virtuallly nothing, some more optimization targets were revealed in some of the Ufasoft threading (or lack thereof) that will likely also yield and increase in average hashrate.

During the course of this, we were examining some of the scope traces of the power draw on the unit while also looking at the At-the-wall draw, which was fairly steady at 82W. 

At 82W, I observed a fairly steady hashrate around 830 MH/s.  The pool, of course, reports higher and lower hashrates based on estimated averages over a 15 minute window, which you can see in the picture.  The other picture shows a peak stable rate on the miner side. As I said, average rate is about 830 MH/s on the miner side and it's fairly locked into that figure with the current configuration.  See below for some thoughts on this as well as efficiency and power usage.

As I said, during this course we looked at some scope output and I noticed the ripple was fairly bad, with spikes well outside of tolerance, which would lead to severe instability... from talking with Sonny and examining both the power brick in use and the power design of the board, I have some conclusions to make.  These are not backed up by BFL and are purely my own speculation:

1. The power brick they were using was rated for approximately 48w, and was being driving at nearly 100% over capacity - the power brick was excessively hot to the touch.
2. The MOSFETs on the board were similarly overheated.
3. The capacitors and power distribution design is not adequate for the amount of power being pushed through the unit under any circumstance.  I have little doubt there would be premature failure on many, many units if they shipped with the current design. Especially if these units were stacked or otherwise restricted in a hot, low airflow environment.

I highlight these three issues to make a point about this development unit.  The unit was stable at 830 MH/s, with underdesigned power distribution and a severely overloaded power brick.  I estimate parasitic losses to inefficiency, heat and underdesign to be at the *very* least 20% on the power side.  I also estimate that instability due to ripple due in a large part to improper power distribution to affect stable hashrates by as much as 10%.  With a redesign of power distribution and a proper power supply, I would say the unit, as it stands right now, would achieve a 60w - 70w or better power envelope, instead of the 82w observed.  I also estimate that the hashrate would increase by at least 10%.  If, during any redesign that happened, the power issues were smoothed out and distribution was beefed up to at least double what it is now, the clockrate could be increased substantially, also increasing the hashrate to at or above initial targets (though power usage would still be 3x original target at a minimum, but still under 100w).  Additionally, with lots of optimization of the Ufasoft code, I can see an effective rate increase and a huge, huge stale/reject decrease being realized.

All that said, I will be reviewing the code we used to mine over the course of the next day or two.  BFL said they would release a unit to me in the next week or two for raping and pillaging at an undisclosed location with just myself and a BitForce box (The undisclosed location is my house.  Shit, now it's not undisclosed.)

While I am disappointed that we were not able to use my laptop, I am confident that the tests performed were conducted in such a fashion as to minimize any possible outside interference, although not eliminate it entirely.  Those of the more skeptical nature will have to wait until I get a unit that I can take home to due further testing, although the possibility of updated code that will compile on a 64 bit platform and a quick test to verify is not out of the question prior to that.

All tests were conducted in the DC and hooked directly into the appropriate parts of my pool to conduct the various tests.  The pictures provided are of the Kill-A-Watt while the unit is in operation, the miner screen showing the internal hashrate (at a peak, not average) and a picture of the miner as described by the pool.

Summary / take away:

The unit operates at 830 MH/s internally at 82w as configured and as backed up by the pool as an estimated hashrate.  Internal pool debug information confirms operation of the miner to be valid and within expected parameters.  I believe with some redesign of the power distribution system, a new, beefier and high quality power brick and Ufasoft code optimization, substantial hashrate gains can be realized while reducing the power requirement by a significant margin.  All tests were conducted in the DC, with the appropriate tests conducted on a non-routable network segment, hooked directly into the pool.  All tests were conducted on BFLs hardware, but the software as used was taken prior to the test for later examination.  A unit will be released for my at-home inspection / pictures soon and another test with updated Ufasoft code is a possibility.

I apologize for the quality of the pictures, I took them with my phone camera while standing on a ladder, leaning over the top of a rack.  In retrospect, I should have brought my DSLR.









If you're searching these lines for a point, you've probably missed it.  There was never anything there in the first place.
cicada
Full Member
***
Offline Offline

Activity: 196


View Profile
December 13, 2011, 05:05:41 AM
 #189

I think we need some fingerprint analysis on the digit reflected in the kill-a-watt, to be entirely sure that aliens were not involved.

Team Epic!

All your bitcoin are belong to 19mScWkZxACv215AN1wosNNQ54pCQi3iB7
xzion
Member
**
Offline Offline

Activity: 96


View Profile
December 13, 2011, 05:12:15 AM
 #190

certainly sounds like there's room for improvement, holding off for now i think.
thanks for taking the time to do the testing inaba.

Tips: 1xzionJBueq1AkPSmexA7suWkztAkNwSs
CubedRoot
Sr. Member
****
Offline Offline

Activity: 295


View Profile
December 13, 2011, 05:14:15 AM
 #191

Hate to say it, but same here.  I will hold off on buying until they get the power supply and circuitry working normally.  I would bet this increases the hash rate a few percent.
BFL
Full Member
***
Offline Offline

Activity: 217



View Profile WWW
December 13, 2011, 05:18:55 AM
 #192

First off, I'd like to thank Inaba for his time and expertise.   It's much appreciated.

Now that we've come to the end of our development process, I think it might be useful to let everyone know what to expect next.

Our prototype units consume more power than was initially expected.  As someone who has had similar experiences here pointed out, the toggle rates required for bitcoin mining are a surprise.  Simply put, our power system wasn't designed to supply that great a load so we had to tune the chips down to 832 mh/s to meet the available power.

Our delivery units are being built with a sufficiently upgraded power system capable of freeing the chips to explore their true potential.  However, since we can't currently demonstrate greater performance and we don't want to be put in the position of guessing, we've reduced our pricing to maintain similar value even at 832 mh/s.  We're sorry for the hiccup, so until the greater performance potential is demonstrated, our BitForce Single will be reduced to $599. 

Delivery dates will be affected only if you had an early dated pre order and in those cases, you may have already been contacted with your options.

Kind regards,
BFL


Butterfly Labs  -  www.butterflylabs.com  -  Bitcoin Mining Hardware
Turbor
Legendary
*
Offline Offline

Activity: 1008


BitMinter


View Profile WWW
December 13, 2011, 05:58:36 AM
 #193

Looks like beta stage to me. The prize is interesting. The hashrate and power requirements not really. With 5 ztex boards you mine at 950+ MH/s @ 50 watts.

CubedRoot
Sr. Member
****
Offline Offline

Activity: 295


View Profile
December 13, 2011, 06:01:13 AM
 #194

Looks like beta stage to me. The prize is interesting. The hashrate and power requirements not really. With 5 ztex boards you mine at 950+ MH/s @ 50 watts.

Yeah, but those 5 ztex board will run you over $2k.... over 3 times what the BFL costs.  Thats  a LONG time to wait to start making revenues on the same performing kit, even with power considerations.
plastic.elastic
Full Member
***
Offline Offline

Activity: 168


View Profile
December 13, 2011, 06:15:41 AM
 #195

"end of our development process"

Wow, that was not expected!  You are just going to sell these as is? Even with the fail rate that Inaba suspects? (His opinion/guess. Not a fact)

I hope you meant you will redesign this, get the power fixed. Get the temps down...

They cant change much now if the production is already underway. The power brick can be easily replaced with something better.

The unicorn has turned into a fancy horse after all.

Nzhang is working on something interested, His board is 2x more expensive but at 4x less power consumption.

Tips gladly accepted: 1LPaxHPvpzN3FbaGBaZShov3EFafxJDG42
CubedRoot
Sr. Member
****
Offline Offline

Activity: 295


View Profile
December 13, 2011, 06:41:13 AM
 #196

"end of our development process"

Wow, that was not expected!  You are just going to sell these as is? Even with the fail rate that Inaba suspects? (His opinion/guess. Not a fact)

I hope you meant you will redesign this, get the power fixed. Get the temps down...

They cant change much now if the production is already underway. The power brick can be easily replaced with something better.

The unicorn has turned into a fancy horse after all.

Nzhang is working on something interested, His board is 2x more expensive but at 4x less power consumption.


Again, at these relatively low power levels, a board that is twice as expensive is going to take a LONG time to recover the costs.  If as you say Nzhang's boards are about 20 watts, you only have about 60 watts difference. How long does it take you to recover the additional $600 in expenses on a 60 watt difference?
Power in my area is .$09 per killowatt hour.
.095 / 1000 watts = $.000095 per watt hour. 
20 watts running 730 hours (I figured that for a month) at my rate of $.095 = ~ 14.06 KW/hours or total cost of about $1.34 / month
80 watts running 730 hours (I figured that for a month) at my rate of $.095 = ~ 51.10 KW/hours or total cost of about $4.85 / month

So that means it would take you about 170 months to recover the cost differential between the BFL unit and the Nzhang unit....thats assuming they output close the same performance, and we are only figuring power differential.

If my math is off, please let me know...its 2am and I have been drinking a little :O
plastic.elastic
Full Member
***
Offline Offline

Activity: 168


View Profile
December 13, 2011, 07:04:17 AM
 #197

"end of our development process"

Wow, that was not expected!  You are just going to sell these as is? Even with the fail rate that Inaba suspects? (His opinion/guess. Not a fact)

I hope you meant you will redesign this, get the power fixed. Get the temps down...

They cant change much now if the production is already underway. The power brick can be easily replaced with something better.

The unicorn has turned into a fancy horse after all.

Nzhang is working on something interested, His board is 2x more expensive but at 4x less power consumption.


Again, at these relatively low power levels, a board that is twice as expensive is going to take a LONG time to recover the costs.  If as you say Nzhang's boards are about 20 watts, you only have about 60 watts difference. How long does it take you to recover the additional $600 in expenses on a 60 watt difference?
Power in my area is .$09 per killowatt hour.
.095 / 1000 watts = $.000095 per watt hour.  
20 watts running 730 hours (I figured that for a month) at my rate of $.095 = ~ 14.06 KW/hours or total cost of about $1.34 / month
80 watts running 730 hours (I figured that for a month) at my rate of $.095 = ~ 51.10 KW/hours or total cost of about $4.85 / month

So that means it would take you about 170 months to recover the cost differential between the BFL unit and the Nzhang unit....thats assuming they output close the same performance, and we are only figuring power differential.

If my math is off, please let me know...its 2am and I have been drinking a little :O

Your math is not off, but your theory is lacking.
Because not everyone only focus on ROI. There is also Opportunity cost which is related to mining capacity.

Your theory is flawed because you assume difficulty stay constant which its not. Hence the mining capacity is mentioned. I doubt the difficulty will stay the same for long.

Tips gladly accepted: 1LPaxHPvpzN3FbaGBaZShov3EFafxJDG42
xzion
Member
**
Offline Offline

Activity: 96


View Profile
December 13, 2011, 09:19:41 AM
 #198

How soon will the unicorn fail? Can it even run for an hour at the current temps? I do not think I will be placing an order any time soon...

Anyway, when will these units be shipped out? Thanks,

they've stated that they've beefed up the power brick. that should solve most of the heat problems, and perhaps a bit of the power/hashrate deficits, as inaba speculated.
just a waiting game now.

EDIT: didn't actually realise they'd knocked $100 off the price, even though the delivery units will ship with the better power block. tough decision.

Tips: 1xzionJBueq1AkPSmexA7suWkztAkNwSs
makomk
Hero Member
*****
Offline Offline

Activity: 686


View Profile
December 13, 2011, 10:36:03 AM
 #199

Our prototype units consume more power than was initially expected.  As someone who has had similar experiences here pointed out, the toggle rates required for bitcoin mining are a surprise.  Simply put, our power system wasn't designed to supply that great a load so we had to tune the chips down to 832 mh/s to meet the available power.
This was actually one of the things that worried me at the time; if you didn't realise that Bitcoin mining would use so much power it seemed unlikely that you would've designed the boards with power circuitry capable of handling its power consumption. (The various Spartan-6 boards have generally been designed with regulators capable of handling the maximum possible power usage.) It may have been a good idea to be more open about the problems you were experiencing.

I actually managed to blow up a wall wart testing the 25 MHash/sec bitstream mentioned in my sig, and that was obviously a lot smaller-scale than your BitForce boards!

Late this afternoon, I was sent the latest codeset to the modified Ufasoft miner for review.  After trying to compile it again, with the same results and trying several different scenarios, I was still unable to compile the software.  I dug a little deeper into the error and consulted with BFL and we determined that the Ufasoft codebase is not 64 bit ready, and thus would not compile on my laptop.
Don't suppose anyone would be willing to give me a copy of the protocol specs and access to a box or VM with a board connected? I already have a poclbm-based FPGA mining client lying around that could almost certainly be adapted. (It does fancy things like pool failover, roll-ntime support and testing found nonces against the previous work unit as well as the current one for increased efficiency.)

Quad XC6SLX150 Board: 860 MHash/s or so.
SIGS ABOUT BUTTERFLY LABS ARE PAID ADS
legolouman
Sr. Member
****
Offline Offline

Activity: 448


Decent Programmer to boot!


View Profile
December 13, 2011, 11:48:07 AM
 #200

Great work Inaba.

I'm not neglecting the use of a working FPGA, but BFL really needs to get their stuff together, the power usage thing, just doesn't seem to add up on their part.

You know what output you want, and how much energy it will take to get there, why was the unit designed with parts that were less than capable of hitting that point?

If you love me, you'd give me a Satoshi!
BTC - 1MSzGKh5znbrcEF2qTrtrWBm4ydH5eT49f
LTC - LYeJrmYQQvt6gRQxrDz66XTwtkdodx9udz
Pages: « 1 2 3 4 5 6 7 8 9 [10] 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 ... 81 »
  Print  
 
Jump to:  

Sponsored by , a Bitcoin-accepting VPN.
Powered by MySQL Powered by PHP Powered by SMF 1.1.19 | SMF © 2006-2009, Simple Machines Valid XHTML 1.0! Valid CSS!