Topic: 1GH/s, 20w, $700 (was $500) — Butterflylabs, is it for real? (Part 2) - page 62. (Read 146936 times)

I think the important things are price, how much it consume, mhash/s and reliability.

I facepalm about the libya/egypt thing because it's ridicolous to troll about ETHICAL ISSUES when ALL electronic is made in china or nearby places where we KNOW laborers are exploited all the way.
Also if it's from libya, it's a + for me, helping their economy now that they are free

So, does that mean that BFL needs to do the exact same thing as others? If they jump off a cliff should BFL follow them also?

Facepalm

Yeah sure, i bet that is the problem that generated 22 pages of troll...erh... critics

Meanwhile you all buy almost everything from china and related places where people are exploited for real but there ethical issues seems to not exists, maybe chinese ppl are robots 

Exactly. +1

The 20W -> 80W fiasco is a big dent on BFL credibility. I can understand when you go maybe 1.5 times wrong but 4 times wrong is just laughable for respected engineers with 10 years of industry experience like BFL claim to have. Maybe they simulated that power draw too and that 19.8W is what the simulator showed ?

Spartan6 has been available in qty 100+ for over 1 year now.

28nm FPGAs are already sampling and will reach production in 1H2012.

Spartan6 today is around 2X the MH/W of BFL, and the 28nm FPGAs should be 3X+.
I know it can be argued both ways but, in the long run, MH/W determines how much
profit a large miner will make, much more so than MH/$. The only way to remain profitable is
to have a MH/W that is significantly higher than the network average, and the only way to
do that is to adopt the most power-efficient technology as early as possible
(hosted in an area with the lowest possible electricity costs).

-rph

AFAIR Easypath is a way to make cheaper chips by using less strict tests (only for two bitstreams instead of full FPGA test). You get same FPGA that needs flash to work, only 35% cheaper. No speed or power benefits.

Easy path isn't going to be cost effective is small batches.  You are looking at a lower per unit cost but a ~$100K to $250K NRE (depends on how much they like you ).    On 1K chips that is a >$100 premium per chip (plus production cost of chip).  Easy path really only makes sense at 10K units+ where than $100K can be ammortized to a $10 per unit cost.

No idea, they're not actually shipping yet. Obviously XC7K325T costs about $1,500 from Digikey but it's always possible the smaller chips might offer better value for money.

There should probably be a T on there actually.

 And what is the cost of one of those chips?
Edit; I also notice you guys are not including a T when speaking about the different kintex chips. Do they make one without the gtx transceiver in it? And does anyone know of the Easy-Path offering is going to be cost effective for smaller batches? I.E., <1k chips. Or even what the costs fo Easy-Path are and if it would even be useful in our situations?

Well, I did a synthesis run ages ago using the preliminary support for the smaller XC7K70 and ISE reckoned that could hit 250 MH/s - and this was just by throwing some Verilog I already had lying around at it and using the default options. Only took about 20 minutes or so too. (This is based on post-place-and-route timing and so should hopefully be fairly realistic.) Obviously the XC7K70 isn't actually available yet, and ISE apparently tends to have problems with multiple-miner designs so this may not scale up to larger chips, but it seems reasonably promising.

Inaba did this at the request of the forum members.  Who should be compensating him?  Has his tip jar been filled?

Inaba's Tips gladly accepted: 168Bgg6HL6bLiUxQDBbdmHRbXtSsMixYCT

+ 2 to this

Good question although I would imagine they are using an off the shelf switching powersupply and they almost universally support 120v-240v.

Would you allow Inaba to keep a unit overnight (or weekend) so he can run a 24/72 hour continual usage test.  I trust Inaba testing more than any internal testing you have done.  Your internal testing lead to a claim of 1.05GH @ 19.8W.  Inaba testing revealed the real performance specs

Even better would you consider giving Inaba a unit in compensation for all the work & time he has put into providing third party testing/assurances/validation.

Hi BFL, what input voltage can the PSUs shipping with the units handle?

My cheapest electricity is at 208v currently and I am hoping your PSUs can handle 120v-240v like a computer PSU.

Thanks.

We need to know what the chip under the hood is.

Couple of reasons :

-maybe these were sourced from Libya / Egypt and there may be some ethical issues there
-this is not something insignificant but the main component of the Bitforce product
-how reliable is this going to be and how long will it last
-can it be repurposed for some other computation ?

Basically, compared to GPUs, with these you are screwed in terms of warranty and reselling. If you don't get your money off them in time or they break you are royally screwed. Will BFL Labs really afford to honor them warranties when they are still sucking off pre-order money ( around $50 000 ) ? Will the boxes last enough to reach break even ? Only time will tell.

I gotta give them this though : good idea for the swappable 80mm fan. Will opening the unit and changing that dead fan affect the warranty ? Thank you !

BitForce unit reliability insights:
 
Regarding our units reliability, we must first consider the environment and conditions under which these units have been operating so far. The system actually consumes about 80 watts of power. The 1.1V core voltage regulator on the prototype system was designed with only 50A (around 55Watts) in mind. The power brick is rated at only 48 watts. As is clear, the system has been running over-currented (by some 20A) both on regulator side and PSU side (32 watts on PSU).
 
In one test, under these harsh conditions, our units have been running without the smallest issue for more than 72 hours without interruption. The temperature on chips barely reach 62 Degrees Celsius, and on power-mosfets they reach near 100 degrees (Max Allowable on MOSFETs is 150). The chips are not reaching this temperature by themselves, but it is the heat from the MOSFETs that are transfered to the chips through the power plane.
 
Having passed all our tests without any problem, it is possible to say that the life expectancy of these units surpass an ordinary computer due to all of the safety margins that are taken into consideration. Furthermore, since both power-regulator and PSU are overcurrented in our prototype unit, their efficiency is reduced to under 70% instead of 85%+. This does mean that the chips themselves are not really consuming 80 watts, but more like 60 watts, and the rest of the power is converted into heat (both in power-brick and regulator on board) due to overcurrenting in our actual prototypes.
 
Regarding BitForce vs. GPU reliability, we must say that the GPU has a much greater risk of failure. It is not uncommon to have graphic cards failing without any explanation. Since graphic cards have to regulate huge amounts of current to support their power-hungry processors (exceeding 400 amperes in some cases), their power regulation system is constantly under threat of capacitor failure due to power-surges in switching-cycles or GPU pipeline/shader on-and-off step transitions.
 
Our production system that is currently in manufacture for delivery, has two regulators on board, supporting max 120A of current implemented as a safety margin. The PSU shipped with the units will be equally supporting up to 135 Watts. This large gap between unit consumption and maximum supported power, ensures high-efficiency, less heat generation and extremely low risk of failure.  The production system will be accompanied by a nearly-silent 28dB fan and an on-board temperature giving miners the ability to monitor on-board temperature in real-time as well as providing auto shut down capability in the event of fan failure.

28nm FPGA's are already avaible. Digikey yesterday have few Kintexes XC7K325 (today they gone). Board design, assembly and synthesis of current code shouldn't take more then month. But it not be economical without new code (maybe 500MH/s).

for the 5870 its a 90 day warranty by the website it still holds a 2+ year warranty yes its a little harder to get it bc there usally oem vizontek cards wich you need your receipt to get the warranty but ive have had no problem getting my cards warranted.

Hi guys,

Regarding the warranty period.  These are solid state devices which aren't 'overclocked' to achieve their performance as some have suggested.  After an initial use burn in test, reliability expectations should be extremely high and no different than a new computer purchase.  

So why are we offering a six month warranty?  Well, let's look at that...  Firstly, I believe 180 days is longer than any other offering in this new class of mining hardware.  (please correct me if I'm wrong, ztex & ngzhang).  Also, right here in this forum, miners are happily snapping up GPU's with less coverage.  See the brand new 5870's advertised in another thread with only 90 days warranty.  (granted, it's a good price but it's still only a 90 day warranty).

I think our six month warranty is pretty good and it goes a long way to establish a high level of confidence in the product. Why not longer then?  Because this is a specialty device in a particularly volatile market who's future is unknown.  It's not inconceivable to imagine a scenario where bogus warranty claims are filed as newer generation units are introduced or market conditions change general interest levels.  As a business, it makes sense to anticipate these factors.  The bottom line is that if your BitForce unit works well for six months under continuous use, there's no reason not to expect it to keep working reliably for the next decade and beyond.

Regards,
BFL