Topic: ASICMINER: Entering the Future of ASIC Mining by Inventing It - page 497. (Read 3917531 times)

You are so off I don't know where to start. Let's just say that AM's 40 nm chips are spec'd to be more efficient than any shipping or announced chips. Next as it has been discussed on here and several other threads, AM is focusing on selling chips not on mining.

Standard ASIC miners in the market (28nm) consume an average of no less than 1000 Watts per Terahash. (it is in reality a lot worse, about 1.3kW/TH).

1 Kilo Watt    --> 1 TeraHash
1 Mega Watt  --> 1 PetaHash
1000 MegaWatts --> 1000 PetaHash

Where is ASICMINER building the nuclear power plant they need to power their miners? The biggest nuclear power in the world generates 6200 MegaWatts, and employs 4000 people. Then ASICMINER needs, not the biggest, but an average sized nuclear plant.

They also need to start buying some land, to start building supermegadatacenters, thousands of times bigger than Google ones. But maybe they have that problem figured out already, as they said, they will outsource those minor issues to local chinese shops.

Or is it that ASICMINER will only build the waffers? (with only <10 PH per year becoming working bitcoin miners).

And they are not even talking about bleeding edge technology. They will build waffers with 40nm chips.
2013 was the 28nm ASIC miners, and this 2014, the 20nm ASICs will start mass delivery.
There is a reason for building ASICs with newer generation chips. Electricity consumption is tightly related to it.
Ask Cointerra, which promised that their new 28nm ASICs were going to be 2x more efficient than KNCs 28nm ASICs because of some magic technology they had. Once Cointerra chips became a reality (TerraMiner IV), their 28nm ASICs were exactly as efficient as KNC ones. How efficient will be ASICMINER chips? At 40nm? They will consume about 50% more electricity than 28nm ASICs, and double what 20nm ASICs (2014 miners) will consume, regardless of any storytelling they might tell now about some magic technologies that will make their ASICs more efficient (exactly as Cointerra did).

Libitum

You are very likely correct -- I just assumed 1 die per chip. My 700 dies per wafer is just a guess as well -- I just counted an approximate radius from the photo and did pi*r*r. I did hear as well they hadn't met their speed/energy marks.

Sorry if I'm being ignorant here but I believe Cointerra like KnC use 4 DIE per package so the 700 chips on the wafer are 125GHs each and they missed their speed and power marks first go around.  If I'm missing something please correct me.

Avalon will have their 3rd gen around the same time as AM:

Speaking of Avalon, based on my calculations AM pricing their gen3 chips at $0.99 GH/s will pretty much overnight make Avalon's 2nd gen 55nm chips unprofitable to produce at the wafer/chip level (not to mention the miner level). This is assuming a generous 4K chips per Avalon wafer, priced at $.99 GH/s (way overpriced considering their energy usage is much worse). That gives you a total selling price of around $5K (again some very generous assumptions on my part) per wafer of chips. That just isn't going to be profitable to make.

I wonder at what price point KNC / Cointerra / etc. chips are no longer profitable to produce at the wafer/chip level? That would be an interesting number to calculate.

That could be an interesting strategy for AM: price the chips aggressively enough that few other chips are profitable to produce, thus giving them a corner on the market until a more competitive chip is available.

(update: from a photo of a Cointerra wafer I found, it looks like they can probably make around 700 chips per wafer, which at 500 GH/s a chip is around 350 TH/s per wafer; much higher than AM's 40 TH/s per wafer. thus at the wafer/chip level Cointerra might be cheaper than gen3 AM, even though their 28 nm per wafer costs are likely much higher)

IMO, it totally validates that the numbers are in the right ballpark and aren't just some BS that AM has made up.

The issue isn't going to be getting that quantity of chips fab'd/diced/packaged. The issue is going to be pricing the chips aggressively enough that they can move that quantity of chips (and still make a good profit). I believe $.49 to $.99 / GH/s pricing is a strong enough catalyst to make that happen, presuming they can start cranking within 2 months.

Those figures show that around 50,000 chips per day can be produced. That doesn't mean that 85k can be produced but it doesn't mean that it can't either.

Basically confirming that its not impossible/unrealistic to produce 85k chips per day.

I forgot about Avalon. Let's look at them.

From the Avalon news post dated 14 October 2013:

"Internally, Avalon has been busy running tests on our A3255 prototype (with 96 chips running at 1.6Ghs) for several days and will continue testing the hardware as well as overclocking and system performance. We will release additional information as it becomes available on the unofficial Avalon wiki page."

From the Avalon news post dated 15 January 2014:

"Until today, we have shipped 6PHash Avalon chips all over the world. Almost 85% was from the Avalon2 A3255 chip. We will keep trying to make the best, creative, open bitcoin auditing solution for you."

From the datasheet:

"1G Hash Rate guaranteed when core voltage is 0.9V overclocking available."

This gives us the following data for the production of Avalon Gen 2:

Time frame = 1 October 2013 to 15 January 2014
Hash Rate: 5.1 Ph/s

5.1 million chips over 107 days
47,663 chips per day

I think it is. Until recently mining companies have been focused on selling high profit mining hardware, but we are seeing (and started to see when Avalon dumped 1 million chips on the market) a shift of ASIC companies just cranking out chips as fast as they can while they leave the end hardware production to 3rd parties that can fight to reduce costs and get miners produces as fast as possible to squeak out a small ROI. This is a fundamental shift for an ASIC company, but a smart move given the quantity of end mining hardware that is going to be produced.

As to how quickly wafers can be produced, diced, and packaged, it is clear that the fab industry is set to handle these types of volumes. Let's break down the numbers assuming 10K wafers over a 12 month period into end mining hardware that will be produced (not by AM, but by 3rd party chip buyers). Assume a hypothetical gen3 miner based on the chip count of AM's Cube (96 chips). 96 gen3 chips will give us a 1.23 TH/s miner that consumes about 500 watts (note those specs are better than any shipping or announced miner despite gen3 using a much larger node size!). Some calcs:

Chips per Wafer: 3,125
Gen3 Miners (96 chips) per Wafer: 32.54

Assuming 10k wafer over 1 year, we get 27.4 wafers / day:

85,616 Chips per Day
891 Gen3 Miners per Day

For a final total of 325K Gen3 Miners over 12 month period. Now that is a TON of miners, and I think it illustrates why AM is getting out of the selling miners business and into the selling chips business. The chips are where the profit is. Scaling up AM to handle producing/selling/shipping/supporting that type of quantity of end hardware (not just chips) would be a huge undertaking and would distract from AM's core value: its ability to create the most efficient ASICs as quickly as possible. So they are focusing on their core competency, which thankfully happens to be where the profit is. AM (smartly) wants to be Intel of the BTC space, not DELL.

Anyhow, if one presumes they can get 10 major buyers for the chips and they each end up with the capability to produce just 100 miners a day, then the math works out. The numbers might sound aggressive, but consider the price point AM is planning on selling the chips at. $.49 to $.99 per GH/s pricing 2 months from now is very cheap compared to competition. I think at those prices they are going to have more orders than they know what to do with (at least for the first few months, then they will have to lower the price to further entice buyers) and they will indeed be heavily supply constrained.

AM is aggressively getting into the chip business and doing so with a product at a low enough price I predict we will have a significant hash rate / difficulty spike in about 6 months. There will also be several new mining hardware businesses that are going to spring up to race product to market to take advantage of the gen3 chip pricing.

Of course I will redirect any questions you give me. The only thing I can't guarantee is whether and how they will be answered.

On the level of details it is my understanding that there is little to fear if they are mainly investor specific. All mining equipment manufacturers are intelligent people and probably the best at understanding the "mining game". As such putting projections out there doesn't really disclose anything, but gives the general audience a clue of what the companies are capable of. Everybody knows it's an arms race to extinguish the profit margin of the competitor.

You're currently in the Securities section.  I recommend you try posting here or here.

Hi guys,

Is there any info or support available for this unit? Mine worked for like 15 hours, but now i'm unable to ping the device. Also the unit seems to turn on (fan is spinning), but the lights on the back stopped working.

Any help would be greatly appreciated !

Thank you

Packaging is done by specialist houses. Neither the fab, nor AM (or any other asic vendor) does the packaging themselves. Clearly if the industry produces 100's of 1000's of wafers per month, those packaging houses have the capacity to slice and package as many chips as are on those wafers. Its not like we are piling up unpackaged wafers.

That said, I also find AM numbers (far) too high to be believable, but its not like the semicon industry (fabs+packaging houses) would have any trouble with that kind of volume.

I've yet to chime in to thank Jutarul for submitting our questions to friedcat. Thanks a lot! I think the answers served well to sort out some burning questions remaining on our side.
I'd like to do this more often if possible. Collect questions for friedcat and the board and submit them. I think bi-weekly is something we can't expect, but maybe something every month or so. If friedcat has specific questions I guess it's easier for him to answer them than to guess what we'd like to know. What do you think? Would that work and be OK with you, Jutarul?

The question is, whether this applies to all process sizes. We need capacities for 40nm, which is not super small but also not something incredibly old and outdated. We'll see. I think friedcat knows better than us.

http://www.internetnews.com/hardware/article.php/3817066/Idle+Hands+in+Chinas+Chip+Business.htm

The Chinese government heavily subsidized the production of integrated circuit manufacturing capability all through the 90's and 00's. Today the amount that is just sitting idle is just staggering.

There exist facility after facility available to be brought online if only someone had something that needed to be produced. Announcing a chip order in Shenzen is like throwing a porkchop into a den of starving wolves. The misallocation of resources we normally associate with centrally-planned economies appears to have worked in AM's favor in this particular instance.

If AM develop a chip to be produced by GF, then it could only be produced by GF as far as I know. Each fab requires their own mask because their technology is different. So AM can use a single fab and they'll have top share that capacity with the fabs other customers. It takes a good few week to produce wafers. If you look at how long it took the other ASIC manufacturers to package their wafers into chips, you'll see that it took them a good few days (and in some cases a couple of weeks) to get just their samples packaged.

This is where an expert needs to step in. Let's say it takes 2 months to produce wafers and AM continuously produces them. Each batch of wafers would need to be packaged before the next batch arrives for packaging. So how many wafers can AM have packaged in those 2 months? That's the bottleneck I'm talking about, not total fab capacity.

Based on what we've seen from all previous bitcoin ASIC manufacturers, it just doesn't seem possible for AM to produce the stated amount of chips in a reasonable time frame.