Bitcoin Forum

With many reports of people getting the november issue of Bitcoin Magazine, and indeed even some scans surfacing, I was wondering if there was any relevant new information to be found in that magazine.

I was under the impression, that the november issue of Bitcoin Magazine would contain more detailed information regarding BFL's ASICs, and stated that issue #4 will answer many questions that people were asking. I've seen scans of this magazine, and it all looks like 6 month old information to me.

Could anyone with access to the full magazine confirm if there indeed is any previously undisclosed detailed technical information on BFL's products?

No, none. Already read all the pages.

I think there was a SUSFU/FUBar (or perhaps even a TARFU) moment when people realized it.

http://en.wikipedia.org/wiki/List_of_military_slang_terms#FUBAR

A BOHICA may soon follow.

That's cool, another potential candidate for them to engage in revisionist editing if it blows up, be hit on the fingers by the community and back-peddle.

Didn't work out too well with issue #1 and Bitcoinica did it?

Did you read the actual magazine, or just the scanned pages?

Wait, so the new information in the mag is not new or just wasn't included when it was hinted it would be? I was under the impression the BFL article contained spoilers.

Did the article mention Nasser's background? I'm not a hardware/software geek, therefore I don't know if his stated background qualifies to engineer ASICs: http://www.freelancer.com/u/Nasser750gx.html

This is the information that was supposed to be contained.

Bogart - that information is indeed in there.  Punin must've missed it.

No, indeed it isn't. The interview refers to the future direction and processes but says nothing about the current chip.

I didn't see it. Could you please quote it?

I was also looking forward in reading about current chip. Grrrr....

I won't quote it - I haven't received the magazine myself yet, and don't want to be responsible for "leaking" the information if you all are lying about having received it.  ;)  But look at the last sentence on page 63.

The copy posted on the BFL forums (with Bitcoin Magazine's permission) doesn't have the page numbers, but would the sentence you're talking about be paraphrased "BFL skipped structured ASICs entirely and went to the full custom approach"?

Oh my... it appears as though I do not have the final revision (though I thought it was the final revision until now), and somehow, between the revision I do have, and the printed copy, the important part was removed.  What the actual f...

I have the magazine. The general article "Fluttering Expectations" is on pages 66 to 69. Than there are two pages of something like Q&A. I found nothing about die size nor exact process size (apart of "the company already plans on reducing the size of their next generation offerings to 45nm or even dipping into the sub 40nm domain").
The new thing is something important for P2Pool - the machine will "send back shares as they are found".

I'm going to examine other pages.

I've ordered issue #2, 3, and 4 about a month ago and have yet to receive any of them yet. It does not surprise me one bit though because it took around 2 months for me to receive issue #1 and by then they were already advertising issue #3. I don't know why it takes so long, but I'm going to continue to buy them either way.

Maybe they removed the important bits so that there was room in the interview with Josh to include two full sets of stats; one for the specs with no power figures, and the newly revised specs with the little single and power information. :P

We'll fix it soon.

This is more like it: http://bitcoinmagazine.net/bfl-confirms-65nm-process-for-sc-lineup/

:)

I guessed right, 65nm, big surprise?

The disturbing thing is they are using Full Custom ASIC, at 65nm process node, at 60watts. (@60GH/s)

The competitor (bASIC) is using structured ASIC I believe and is below the 100watt mark on a 90nm process node. (@54GH/s)
Some leaked pictures show that it might be at around 58 to 60watts. But the caption was removed so it is hard to figure out if true or not.

Either way, for a smaller process node...not as efficient as it should be. Perhaps I am missing something....?

I know that 60 watts is an estimated and conservative figure.

When did Tom say that they were using a structured ASIC?

Altera Hardcopy is "65nm" too.

Of course they won't admit that they are using it.

You're saying they are lying about creating a full custom ASIC?

Yes

I don't see 65nm on the product page: http://goo.gl/LhjH7

Ok they probably are not using Altera Hardcopy.
Perhaps Xilinx Easypath? You see there are several companies who off fpga conversion "ASICs", I don't know them all.

Could you point us to which Hardcopy family members use a 68 pin 11mmx11mm QFN?
http://www.altera.com/devices/asic/hardcopy-asics/about/hrd-index.html

So is the lower the nm, the better?

so if someone came out with 20nm, that would be faster or use electricity than a 45nm chip?

Probably both.

You are wrong again, ElectricMucus. The few times I have made predictions(*) against "mainstream thoughts" in the Bitcoin world, I have always been right, and I will tell you: I predict SC is not based on a FPGA-to-ASIC chip.

(*) Just a few examples: I correctly estimated HD 7000's MHash per Joule before release, from theoretical computations based on its known process node. I correctly predicted pirateat40 was running a Ponzi when many refused to believe. I correctly predicted 65nm (https://bitcointalk.org/index.php?topic=95762.0) for the SC lineup. And I seem to be on a path to win my SC mining efficiency bet (http://betsofbitco.in/item?id=665) when people where ignorant enough to bet 50 BTC against me.

^ Prepare yourself to be appended to this list in the near future. I have the track record to justify my confidence :)

The toggle-rate is so high that they are running the chips with much lower frequency that you would normally expect to run 65nm designs on to protect them from overheating. Had they gone with other package (eg. flip-chip), they could've run them at much higher clocks.

What leaked pictures are you referring to? URL?

Regarding the chip being hardcopy or easypath or any other structured ASIC, please be advised that none of those chips come in small packages. They are all 480pin or above (BGA chips), not QFN. The reason behind it is that most industries going after hardcopy or easypth are the ones using the high IO count FPGA version of those chips (which are very expensive for a mass-produced product).

I hope this clarifies the situation.


Regards,
BFL-Engineer

Better later than never ;)

WE just have to wait and see.

My issue came yesterday :)

Read the article and now i'm thrilled to see an actual Product out and hashing!
If that happens i might consider buying one or two singles SC; or even more! (i won't preorder ..... that is not about to change since i don't like that "tactics" of bfl. Maybe if you had called it community funding with preorder guarantee back then, i would have.....  ;))

Can the chip package be changed to a more thermally efficient design in a subsequent production run of chips without having to make a new mask, and incur a whole new set or NRE costs?

This is an excellent question.  Here's my educated guess for an answer:  It is likely that a new mask would need to be made for the top layer at a minimum.

Edit: Yes flip-chip would be a much better thermal design.  We do flip-chip for our custom chips.

In the article is mentioned that these chips can in theory, run up to 16Gh/s max., but will not live long if pumped up to this. What is needed to overclock these single SC's chips to maybe 10-12Gh/s a piece, so a single would do 80-96Gh/s? Just the right mining software, or is a modded firmware needed?

Well, they have the waterblock heatsink that could be used.  I bet that'd get some nice overclockz going.

The article definately alludes to the chips running 50% theoretical I would guess another 50% (80-90Ghash for SC) would be safe with the right cooling set up. It doesn't really excite me as much as it used to because, hey everyone else is going to do it too.