BFL - Update 3/4

[witherworth]

For anyone who doesn't want to browse their forums:

https://forums.butterflylabs.com/announcements/692-bfl-asic-status-2.html#post16987


From Josh:

4 March 2013 Update

Today the bumped wafers were confirmed at the packaging facility and they have been "bumped nicely," so all is well as far as bumping goes. Packaging is currently scheduled to be done on Wednesday, although that's subject to change, I don't see any reason that is going to change at the moment.

The chips achieved 350 MHz hashing on 4 engines as well. We are going to have to wait until the chips are packaged to take the chips higher than that with more cores, as the test rig is not able to supply enough power and there are some wire bond issues that make it unstable beyond that. Once the chips are packaged, they will be tested more fully. Regardless, the chips are hashing properly and all IO appears to be correct, which is the important part. I will update when further testing is complete.

Timeline for the second set of 6 wafers is currently scheduled several days out after we have a successful test of the packaging (they'll have to make their way through bumping and packaging as well) and then the remainder is scheduled for ~7 days after that. I can't be more specific with regards to a date on the 6 wafers yet, as I don't have specific dates. I will update once I do.

I wonder why they sacrificed the 1k chips for testing, if they can't even fully test them? It was only a few days ago that they entered testing phase. It would have made more sense to save the 1k chips, wait the 5 days, and be able to deliver that many more chips to customers instead of making them wait even longer. Me thinks something else may have happened, but that's pure speculation.

[1715481854]

1715481854

[1715481854]

1715481854

[miter_myles]

Can't wait for tomorrow's thread 3/5 .... then maybe 3/6??? yes?? PLEASE?!?!?!

[Clearfly]

28th of Feb they were going to be testing to 500Mhz the following day, ie the spec they were designed to work at. 3-4 days later still only tested at 350Mhz and some mumblings about certain non-specific issues.

Sounds like the same old delaying tactic deployed for months on end now. Can't see BFL actually surviving much longer.

If they don't deliver something soon they WILL go bust. Avalon aren't far behind following suit with their BS stunt using #2 to fund #1 orders.

Think i'm gonna place a bet on BFL not delivering this month, and take the winnings on an up.

[PuertoLibre]

You know what I find interesting?

The fact that in this ASIC project they are using valuable time to test the chips and at the same time convince customer(s) they actually work.

This is what should normally take a few weeks as pointed out by a few hardware folks on this forum. This necessary testing process is being compressed into a few days in an "Ad-Hoc" manner.

What is even more interesting, is that some false impressions are being beamed out.

For example, BFL customers are being comforted with the illusion that their chips will ship soon "for certain". But...they are still running the chips through "functional tests". NOT full throttle performance evaluation(s) mind you, no, just basic low speed functionality tests.)

So in essence BFL is taking a calculated risk for the sake of speed. What normally would have been done in a back room somewhere (away from prying eyes) for a few weeks is actually being done in open view.

That doesn't bother me much as a spectator.

What really does bother me, is that two mixed impressions are being given out. The first false implied impression that the chips work and will ship soon. (they don't know this yet, but thats why they are testing it.)

The second false impression is that the chips are just undergoing a "little testing" to confirm things are in working order. That isn't the case. The testing going on is whether the design actually functionally works, whether it meets (performance/electrical) specs and finally whether the chips ultimately are worth while.

Anyone with a semblance of understanding would read these updates and realize that alot of praying is going on as they rush through each step.

--------------------

Keep this in the back of your mind when you read what is being disseminated or at the very least when you ask the next question. Because someone who has a false impression would be disappointed if the chips break at stock speed.

While someone who is an engineer or a semi-conductor designer knows this is simply a reality of the "testing phase" of a design.

-------------------

[PuertoLibre]

BFL is making it look like the chips working are a done deal. To give the aura that things will be done soon. All the while updating people with details on what....well what they actually don't know yet "for sure"...whether the chips actually work at the speed they originally intended.

Josh mentioned at the beginning of the weekend that the chips were about to go through full speed tests a few days ago after they ran a low speed functionality test. Then he turns around at the end of the weekend and reveals that this can't be done without the full packaging.

You can tell they are cowboys. Cowboys that are clearly disconnected from the process or the innate knowledge of what is necessarily possible by the professionals making the ASIC design.

So far, they sacrificed 1,000 chips (a full wafer) for functionality tests. Seemingly without a reason (they are cowboys remember?). Had they actually known that the full speed tests could not be done without proper packaging, they would probably have not scarified a full wafer. When they did sacrifice a full wafer, they seemed to be unaware that full speed tests would not be "do-able".

Next, they have begun finalizing the next batch of 6 wafers before they actually know what will actually happen when all 16 cores hash at full speed. (magic smoke?)

So if there is anything wrong, the second batch will likely suffer from it as well.

[johnyj]

Just hope that they do all the function test thoroughly, returning a buggy product after several weeks of running will be even more terrible from an end user's point of view

[pc163]

any update of shipping date? the bumping took so  much time, i have a bad feeling about the shipping

[cdog]

Yeah its definitely concerning. Why test the chips at all if they cant do it at the full speed? What happens if they build the prototype and only 50% of the chips are stable above 400mhz?

Until they have working protoypes in the casing with a stock powersupply and cooling, everything is still a gamble.

Im a customer so Im trying to stay optimistic but I cant even imagine how people who ordered summer of 2012 are feeling with thousands of $ locked up for 8 months and endless excuses and delays. 

[Bogart]

Next, they have begun finalizing the next batch of 6 wafers before they actually know what will actually happen when all 16 cores hash at full speed. (magic smoke?)

I don't believe they have actually given the go-ahead to finalize the next 6 wafers.  Read the announcement again.

Just hope that they do all the function test thoroughly, returning a buggy product after several weeks of running will be even more terrible from an end user's point of view

In this case I disagree.  For as normal product this would be true, but in this case I think that customers would rather have it for a few weeks and have to return it than not.

It would suck for the vendor though.

Im a customer so Im trying to stay optimistic but I cant even imagine how people who ordered summer of 2012 are feeling with thousands of $ locked up for 8 months and endless excuses and delays.  

Imagine ppl who ordered then and paid in BTC.  A Single would cost about 216 BTC.

[ab8989]

Actually I do not find this sacrificing one wafer odd at all. I am not sure if they are sacrificing really anything. My bet is going to be they are not.

The thing you all are overlooking is that is really really difficult to ramp up production so fast as BFL claims and as everyone expects. Avalon did not ramp up that fast either as they claimed to have done.

My bet is that the first batch size was selected to be high enough so that they do not run any risk of having shortage of chips while there are other factors that are going to affect the number of boxes they can possibly build on the first couple of weeks. So now they probably have realized that they were really optimistic when choosing the size of first batch of ASICs and they can safely "sacrifice" one wafer without actually loosing anything. They could probably use a second and third wafer from this batch for target shooting practice and still not limit the box output in any way and still have unused chips in corner when the second batch of ASICs arrive. All they "lost" was some manpower from their subcontractor which is going to send them a bill for.

Someone made a list of tests a reputable ASIC design house runs to declare whether a ASIC is really good or not. My bet is none of the bitcoin ASIC design teams are going to run even 10% of those tests. They cost money to run and who cares if the ASIC fails in some temperature. Actually if the ASIC fails some of the more obscure tests it costs the bitcoin ASIC design house a huge amount of money as they have to throw the whole lot to the bin and make a respin. Better make sure no tests are failed and that is made sure by not running the tests. It is a basic seems to work OK or NOT only test and away you go with manufacturing. Good luck with warranty claims covering the whole lot, not just some individual machines. This basic OK/NOT test has now been run on the test chip and it will be repeated on packaged chip and that is pretty much all testing there is going to be. Or if there any any other tests run it is just for giggles only and any fails observed while doing so will be promptly un-seen.

[DrG]

Im a customer so Im trying to stay optimistic but I cant even imagine how people who ordered summer of 2012 are feeling with thousands of $ locked up for 8 months and endless excuses and delays. 

Imagine ppl who ordered then and paid in BTC.  A Single would cost about 216 BTC.

That's about what I paid for my Singles on Day 1.  I thought I would get my ASIC before most everyone and be able to make the 200+ BTC back.  I'll make my $1350 back but I did not intend to "sell" the BTC - I wanted to leverage it into more BTC.  In that aspect I have failed (or rather somebody failed me).

[smoothie]

...
Anyone with a semblance of understanding would read these updates and realize that alot of praying is going on as they rush through each step.

...

This ^

Reality is going to set in fast with BFL once their customers get their units that dont run at the speeds advertised or they do and then burn out or BFL doesn't ship at all and continues delays.

Pick your poison, at the end of the process, BFL is essentially screwed, but BFL customers are screwed^2.

[Puppet]

I wonder why they sacrificed the 1k chips for testing, if they can't even fully test them? It was only a few days ago that they entered testing phase. It would have made more sense to save the 1k chips, wait the 5 days, and be able to deliver that many more chips to customers instead of making them wait even longer. Me thinks something else may have happened, but that's pure speculation.

Its Josh or whoever is calling the shots for him, not having a clue. I pretty much predicted this spot on:

Its just Josh learning. Over the past few weeks he learned that "2 days" means "the bumping process takes 2 days, but can only be started in 16 days" and that packaging facilities have their own internal planning and arent going to be bribed for a tiny 5 or 6 wafer order.  Shocking.

So now he learned that testing chips takes longer than 24 hours so they are tossing away ~6000 chip candidates to get a handful of chips in testing a few days earlier than otherwise possible. That is 8 months after they started taking your money.  Does that make sense? No. Unless you planned this completely wrong, started production before testing and are still trying to stick to a ridiculous timetable you shouldnt have ever promised in the first place. And even then it doesnt make a lot of sense, because you can only test so much on wire bonded chips. Among other things, it will tell you nothing about the physical properties of the BGA chip  and its going to be hard to do any board level testing.

But he will learn that too, eventually.

So as predicted,  he now learned that testing a manually wirebonded chip doesnt really tell you all that much and they still cant green light the other wafers (which was the stated purpose) until they have a few BGA packaged chips to test with. Making that 6000 candidate sacrifice an utter waste.

Only in bitcoin world can a semi company that constantly screws things up this badly still be considered a market leader thats on track to making a shitload of money. In any other market they would have been bankrupt by now. Lets hope helveticoin is for real and shows them how its done.

[nathanrees19]

Anyone with a semblance of understanding would read these updates and realize that alot of praying is going on as they rush through each step.

More proof that prayer doesn't work.

[smoothie]

Anyone with a semblance of understanding would read these updates and realize that alot of praying is going on as they rush through each step.

More proof that prayer doesn't work.

More proof that BFL is full of BS.

Prayer works, just how you use it matters. But we can discuss that elsewhere.


"Please oh please God, just let these chips perform super duperly so we can satisfy all of our empty promises to our customers up until now. And so we can make lots of MOH-KNEEz! Amen"

Just a thought of what they are praying...sounds ridiculous lol. 

[ralree]

Even if only 30GH is obtained, the product is so much better than Avalon's and there are so many more TH sold that it won't actually change profits that much.  Sure, 66GH would be better, but we're talking about a 10x increase in difficulty here.  Go ahead and do the math - for 33GH, say the diff increases 13x (because of avalon+asicminer), and for 66GH say it increases 21x, and see what comes out.  Running that shows $11/day vs $13/day per single SC - not much to cry over.

[Cablez]

This whole testing fiasco is really a wonderment.  To lose that many precious (and expensive) chips to only gain a marginal amount of useful information is insane.  At any company I have worked for that kind of decision would have landed someone a pink slip.

[Puppet]

This whole testing fiasco is really a wonderment.  To lose that many precious (and expensive) chips to only gain a marginal amount of useful information is insane.  At any company I have worked for that kind of decision would have landed someone a pink slip.

To be fair, the cost of a processed 65nm wafer is likely somewhere between $1500-$2000. Not exactly a fortune.

[PuertoLibre]

This whole testing fiasco is really a wonderment.  To lose that many precious (and expensive) chips to only gain a marginal amount of useful information is insane.  At any company I have worked for that kind of decision would have landed someone a pink slip.

To be fair, the cost of a processed 65nm wafer is likely somewhere between $1500-$2000. Not exactly a fortune.

Avalon pays (it is estimated) about 4k per wafer at 110nm.

How is BFL paying half that at 65nm? Somehow your estimates sounds kinda off.

[Cablez]

Actually I meant expensive in the sense of recouping the NRE with shipped product and not the actual cost of the wafer.