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
Topic: 1GH/s, 20w, $700 (was $500) — Butterflylabs, is it for real? (Part 2) - page 72. (Read 146909 times)
December 13, 2011, 01:18:55 AM
December 13, 2011, 01:14:15 AM
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.
December 13, 2011, 01:12:15 AM
certainly sounds like there's room for improvement, holding off for now i think.
thanks for taking the time to do the testing inaba.
thanks for taking the time to do the testing inaba.
December 13, 2011, 01:05:41 AM
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.
December 13, 2011, 12:00:07 AM
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.
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.
December 12, 2011, 11:52:34 PM
BFL homepage appears to be updated:
Performance: 832 Mega Hash / s @ 80w
Price says $599 again
Performance: 832 Mega Hash / s @ 80w
Price says $599 again
December 12, 2011, 11:15:50 PM
Eagerly waiting for results
December 12, 2011, 10:37:50 PM
Wow, finally after a long wait we are getting something.
December 12, 2011, 10:04:42 PM
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.
December 12, 2011, 03:28:53 PM
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.
Hopefully they are to the point where they can release it to me and I can test it properly.
December 12, 2011, 02:11:16 PM
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
December 11, 2011, 10:46:56 PM
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.
December 11, 2011, 07:42:02 AM
For the record, I believe that neither DeepBit nor Inaba are shills for BFL
And I can't imagine what deluded frame of mind is causing other people to think otherwise. It's insane.That's exactly the kind of things that everybody said when people started to question Bruce Wagner's relationship with MyBitcoin.
No one in their right mind said that about BW. I saw some feeling bad for him, etc, as I did but that was about the extent of it...
To hold a claim of the form "no one said xyz", you would need to have read thoroughly all the threads of bitcointalk and SA which you obviously didn't do otherwise you would know that when a few people started to whistle the blow about Bruce Wagner, the initial reaction from the community was that of amused denial.
You know how temptating it is for your average Joe to give in to a little "let's put our tin foil caps" sarcasm. People love that. They think that makes them look smart.
It depends on whether you wear a matching tie or not.
December 10, 2011, 09:40:16 AM
I noticed that many people here are totally lost about what has be told and done, told but not done, done but not told and neither told nor done regarding the third party testing of ButterflyLab BitForce Single product.
With all the stuff scattered over two lengthy threads and weeks of endless debate, it is geting so confusing that even Inaba seems to be lost about what he has said and done.
This is why I have written for you the little pocket guide of "ButterflyLabs 3rd party testing for Dummies".
Thread : https://bitcointalksearch.org/topic/1ghs-20w-700-was-500-butterflylabs-3rd-party-testing-for-dummies-54169
All the quotations are taken from the "is it a scam" and "is it for real" threads (courtesy of BFL and Inaba).
The emphasis in bold and comments are from me.
I have tried to remain objective, but I am of course totally biaised, so do not base your opinion solely on my comments.
Please base your opinion on the facts, quotations, and linked messages.
This is a recommanded reading for :
- People waiting for their pre-order to be delivered
- Inaba, to help him sort out what he has said already
- Everyone else who wonders what the heck is going on with that ButterflyLabs testing nonsense.
With all the stuff scattered over two lengthy threads and weeks of endless debate, it is geting so confusing that even Inaba seems to be lost about what he has said and done.
This is why I have written for you the little pocket guide of "ButterflyLabs 3rd party testing for Dummies".
Thread : https://bitcointalksearch.org/topic/1ghs-20w-700-was-500-butterflylabs-3rd-party-testing-for-dummies-54169
All the quotations are taken from the "is it a scam" and "is it for real" threads (courtesy of BFL and Inaba).
The emphasis in bold and comments are from me.
I have tried to remain objective, but I am of course totally biaised
, so do not base your opinion solely on my comments.Please base your opinion on the facts, quotations, and linked messages.
This is a recommanded reading for :
- People waiting for their pre-order to be delivered
- Inaba, to help him sort out what he has said already
- Everyone else who wonders what the heck is going on with that ButterflyLabs testing nonsense.
Thank you ! Very useful resource for people. Is that demo going to come this weekend or not ( yet again ) ?
I think we should keep the off topic to a minimum on this thread and concentrate on the hard facts. Let's keep the "Inaba = BFL" conspiracy in another thread if we can.
Agreed.
I would prefere to let BFL do their job. They will present their product and specifications when it is ready for release.
Speculation about what might be or making them hurry won't improve neither the result, nor speed up the development.
Even if it would not meet the initial specs by a considerable margin it will be superior to most current alternative or at least competetive.
I also understand the current situation as they are tuning on an optimised release solution and will present it to us as soon as it is reasonable.
So maybe lets stay put for the moment and wait for some hard facts
December 10, 2011, 09:31:41 AM
I noticed that many people here are totally lost about what has be told and done, told but not done, done but not told and neither told nor done regarding the third party testing of ButterflyLab BitForce Single product.
With all the stuff scattered over two lengthy threads and weeks of endless debate, it is geting so confusing that even Inaba seems to be lost about what he has said and done.
This is why I have written for you the little pocket guide of "ButterflyLabs 3rd party testing for Dummies".
Thread : https://bitcointalksearch.org/topic/1ghs-20w-700-was-500-butterflylabs-3rd-party-testing-for-dummies-54169
All the quotations are taken from the "is it a scam" and "is it for real" threads (courtesy of BFL and Inaba).
The emphasis in bold and comments are from me.
I have tried to remain objective, but I am of course totally biaised, so do not base your opinion solely on my comments.
Please base your opinion on the facts, quotations, and linked messages.
This is a recommanded reading for :
- People waiting for their pre-order to be delivered
- Inaba, to help him sort out what he has said already
- Everyone else who wonders what the heck is going on with that ButterflyLabs testing nonsense.
With all the stuff scattered over two lengthy threads and weeks of endless debate, it is geting so confusing that even Inaba seems to be lost about what he has said and done.
This is why I have written for you the little pocket guide of "ButterflyLabs 3rd party testing for Dummies".
Thread : https://bitcointalksearch.org/topic/1ghs-20w-700-was-500-butterflylabs-3rd-party-testing-for-dummies-54169
All the quotations are taken from the "is it a scam" and "is it for real" threads (courtesy of BFL and Inaba).
The emphasis in bold and comments are from me.
I have tried to remain objective, but I am of course totally biaised
, so do not base your opinion solely on my comments.Please base your opinion on the facts, quotations, and linked messages.
This is a recommanded reading for :
- People waiting for their pre-order to be delivered
- Inaba, to help him sort out what he has said already
- Everyone else who wonders what the heck is going on with that ButterflyLabs testing nonsense.
Thank you ! Very useful resource for people. Is that demo going to come this weekend or not ( yet again ) ?
I think we should keep the off topic to a minimum on this thread and concentrate on the hard facts. Let's keep the "Inaba = BFL" conspiracy in another thread if we can.
December 10, 2011, 09:16:48 AM
I noticed that many people here are totally lost about what has be told and done, told but not done, done but not told and neither told nor done regarding the third party testing of ButterflyLab BitForce Single product.
With all the stuff scattered over two lengthy threads and weeks of endless debate, it is geting so confusing that even Inaba seems to be lost about what he has said and done.
This is why I have written for you the little pocket guide of "ButterflyLabs 3rd party testing for Dummies".
Thread : https://bitcointalksearch.org/topic/1ghs-20w-700-was-500-butterflylabs-3rd-party-testing-for-dummies-54169
All the quotations are taken from the "is it a scam" and "is it for real" threads (courtesy of BFL and Inaba).
The emphasis in bold and comments are from me.
I have tried to remain objective, but I am of course totally biaised, so do not base your opinion solely on my comments.
Please base your opinion on the facts, quotations, and linked messages.
This is a recommanded reading for :
- People waiting for their pre-order to be delivered
- Inaba, to help him sort out what he has said already
- Everyone else who wonders what the heck is going on with that ButterflyLabs testing nonsense.
With all the stuff scattered over two lengthy threads and weeks of endless debate, it is geting so confusing that even Inaba seems to be lost about what he has said and done.
This is why I have written for you the little pocket guide of "ButterflyLabs 3rd party testing for Dummies".
Thread : https://bitcointalksearch.org/topic/1ghs-20w-700-was-500-butterflylabs-3rd-party-testing-for-dummies-54169
All the quotations are taken from the "is it a scam" and "is it for real" threads (courtesy of BFL and Inaba).
The emphasis in bold and comments are from me.
I have tried to remain objective, but I am of course totally biaised
, so do not base your opinion solely on my comments.Please base your opinion on the facts, quotations, and linked messages.
This is a recommanded reading for :
- People waiting for their pre-order to be delivered
- Inaba, to help him sort out what he has said already
- Everyone else who wonders what the heck is going on with that ButterflyLabs testing nonsense.
December 10, 2011, 07:07:49 AM
...
Could you please stop posting? I would like to read posts with concrete data and not posts saying that there isn't concrete data. We already have dozens of pages and one locked thread with that.
Exactly! There are far more talks than actions. Am getting bored reading the thread.
December 10, 2011, 07:01:21 AM
you mean 1.5MH/$? yeah, it's extremely possible in the next 6-8week. 
Meaning $120 for a fully assembled/tested ~180MH/s 6s150 FPGA miner?
-rph
or a $1200 for a 1.8Gh/s miner
December 10, 2011, 04:39:51 AM
...
Could you please stop posting? I would like to read posts with concrete data and not posts saying that there isn't concrete data. We already have dozens of pages and one locked thread with that.
December 10, 2011, 03:32:38 AM
Holy Hell these forums are the most cynical I have ever read..
BFL has said that his product was discovered before they were truly ready to announce it. I take that as they were gearing up the marketing ready for their "go-live" just as soon as they had the hardware running like they wanted. This is normal business practice, so there is no lag between the hardware readniess to ship and the marketing team promoting (the website in this case).
It appears some of us (like Chugalug, a Linux Users group in Chattanooga that I am part of) heard about it early and started the discussions with good intentions.
I do think BFL has a product, and that it might meet expectations. I have not pre-ordered any, simply for the fact of being short on cash (tuition due next week). Theres no reason to bash Inaba since he has volunteered his time to test this thing out, and post reviews. You as a consumer vote on BFL's success with your dollar votes. I can't wait to see what they come up with, and I am excited to see (hopefully) a new era in mining.
BFL has said that his product was discovered before they were truly ready to announce it. I take that as they were gearing up the marketing ready for their "go-live" just as soon as they had the hardware running like they wanted. This is normal business practice, so there is no lag between the hardware readniess to ship and the marketing team promoting (the website in this case).
It appears some of us (like Chugalug, a Linux Users group in Chattanooga that I am part of) heard about it early and started the discussions with good intentions.
I do think BFL has a product, and that it might meet expectations. I have not pre-ordered any, simply for the fact of being short on cash (tuition due next week). Theres no reason to bash Inaba since he has volunteered his time to test this thing out, and post reviews. You as a consumer vote on BFL's success with your dollar votes. I can't wait to see what they come up with, and I am excited to see (hopefully) a new era in mining.
Jump to: