Estimate of ASIC pre-orders: 13 to 15 PH/s (diff 1.8B to 2.1B) by end of 2013

[World]

Avalon expect to sale second generation chips (55nm) late October 2013 for immediate shipping.
http://thegenesisblock.com/avalon-refunds-22000-btc-2-9m-to-asic-miner-customers/

[1713915031]

1713915031

[1713915031]

1713915031

[1713915031]

1713915031

[1713915031]

1713915031

[DeathAndTaxes]

Avalon expect to sale second generation chips (55nm) late October 2013 for immediate shipping.
http://thegenesisblock.com/avalon-refunds-22000-btc-2-9m-to-asic-miner-customers/

So any ideas on size of their batch?

We don't have a lot of facts to go on for Avalon v2 but:
a) We know it is 55nm compared to 110nm.  Die efficiency should be 4x that of v1 Avalon.
b) It is "backwards compatible" which I interpret to be same package/pinout.  
c) Avalon says late Oct which is code word for Dec and given their delays in shipping, "customs issues" I probably will put them as complete miners hashing in January 2014 or later.
d) Avalon says they won't be doing presales.  That means they accept the full risk of the NRE + batch run costs (as fab will want that up front).  I think this will act as a limit on # of wafers ordered.

Guestimate:
Avalon current design is a single hashing engine (1 hash per clock) running at up to 330 Mhz for 330 MH/s.  The current chip is on a 110nm process and has a die size of 16.12 mm^2 which means a die efficiency of 20.5 MH/s per mm^2.  A perfect die shrink would make that ~80 MH/s per mm^2.  For comparison BFL is ~70 MH/s per mm^2 and Bitfury is 138 MH/s per mm^2 in efficiency.  To give Avalon some margin lets say they improve die efficiency to 100 MH/s per mm^2 (improved clock speed, tweaked hashing engine).   Now Avalon likely isn't going to keep a single hashing engine design as the die would be insanely small (~4mm^2 as 55nm).  So it seems plausible they will take the v1 design shrink it and use the additional space to put more hashing engines running in parallel, similar to how Intel can make a 6 core chip today which has the same die size as a 130nm chip from the past.  Exactly how many parallel hashing engines Avalon uses doesn't really matter because it doesn't change the die efficiency (MH/s per mm^2).  A 6 HE chip is going to have 6x the throughput but take up 6x the die space.  There is some overhead for controller logic and I/O but it is minimal.  Bitcoin mining is an "embaressingly parallel problem".  My hypothesis is that since time is short, Avalon isn't looking for a redesign from scratch but rather they will take the v1 hashing engine (possibly with some minor tweaks/improvements and higher clock speed), shrink it to 55nm and then use as many in parallel which will keep the die size small enough to fit their existing package.  My guess is that would be 4 to 6 making the nominal output per chip 2GH/s to 3 GH/s.

A 300mm wafer has an area of 70,685 mm^2, lets assume 80 MH/s per mm^2 with a 10% drop due to wasted space (squares inside a circle).  That comes out to ~6.5 TH/s per wafer.  There are more accurate methods of calculating number of full dies but it requires knowing the exact dimensions of the chip but 90% effective is a good enough guesstimate.  If they get 95% yield that would be ~6TH/s usable chips per wafer.  Standard batch is 50 wafers so we are looking at 300 TH/s per batch.   Now they can order as large of a run as they want but given they have to front this cost/risk and with uncertainty on demand/pricing I could see them going with 2 batches (100 wafers) or 600 TH/s.  Avalon should have the resources to do that ($1M @ $10K nominal per wafer plus NRE).  While they could prepay for more that would be taking a big risk (no pre-orders to dump the risk on the customer).  They can always purchase more wafers at a later start is their is enough demand for follow on months.

So 600 TH/s in January for Avalon 55nm?  What does anyone else think?

[console_cowboy]

That sounds like a very good estimate of the Avalon v.2 chips. Considering their short time frame and backwards compatibility, I would say your educated guesstimate is very close to on the money.

• The comments about using the same boards as the v.1 chips causes me to agree that the chips will be the same die size.
  
• The short time frame and backwards compatibility also support scaling out the single hash engine to accommodate the smaller device size.
  
• Your estimates on cash flow of avalon support the batch sizes you've suggested as well. The lack of preorders does place a much higher risk on BitSynCom, and I doubt they will sink their entire profit of v.1 into v.2. It honestly seems like the demand for asics has cooled off some.
  

Seems like another great estimate as always.

[btc_uzr]

KNC Sept/Oct 500TH/s is way too low, because September are day1+2 [1,7] only(?) and ~800 orders [2].
Most of the orders contain at least one Jupiter, few orders contain one Saturn [6] which has not upgraded meanwhile [speculation].
October orders are shipped on 11 business days, so lets assume another 11*(800/2)= 4400 orders (containing Mercuries as well).
According to knc-staff Jupiter will have at least 500GH/s [3].

So let's calculate with 0,5TH/s as an average per order [6]: (800 + 4400) * 0,5TH/s = 2.600TH/s
Maybe even more since there are orders with 6,8,10+ Jupiters [6].
Cloudhashing wanted to buy from knc as well [4], don't know whether they actually pre-ordered. If so I guess it's one huge order. Looks like they did [7].


Those days knc is shipping ppl will think the network-graph must be broken as it is showing hashrate increase at an angle of 88° or so until October 15th when last ones are shipped [5]. At least it will not be comparable to anything we saw before during this time x_X



[1] https://www.kncminer.com/news/news-32
  "Finally we wish to confirm that yes we are still on track for our delivery towards the end of September."

[2] https://www.kncminer.com/news/news-20
  "The number was 890 but has dropped a little as not all bank transfers arrived in time."

[3] http://forum.kncminer.com/forum/main-category/hardware/107-psu-recommendation-jupiter-the-largest-planet?p=1050#post1050
  "According to our engineers an 850W, 80+ Gold certified PSU should be enough for a Jupiter running at the expected speed of 500 GH/s."

[4] https://cloudhashing.com/blog/13-blog/press/65-bitcoin-mining-as-a-service-maas
  "He (Cloudhashing’s founder, Emmanuel Abiodun) is also in negotiation with KNC, which he says is pledging September delivery for its own ASIC products."

[5] https://www.kncminer.com/news/news-33
  " All current paid orders will be shipped no later than October 15th.
    All new sales will reflect our November price point of;"

[6] https://bitcointalk.org/index.php?topic=249065.0
   "In Total 138 Jupiter 65 Saturn 17 Mercury"

[7] https://www.cloudhashing.com/frequently-asked-questions
  "A1: (...) We are currently mining and by September 2013, will be operating at over 115 terahashes. This number will climb steeply to 300 terahashes by November 2013. "
  "A2: We expect mining for September contracts to start by 26th September 2013 (...)"


*edits:
-added [speculation]
-removed [unverified] - thanks for contribution!
-added new links
-made it more readable

[geofflosophy]

KNC Sept/Oct 500TH/s is way too low, because September are day1+2 [1,speculation] only and ~800 orders [2].
Most of the orders contain at least one Jupiter, few orders contain one Saturn which has not upgraded meanwhile [speculation].
October orders are shipped on 11 business days, so lets assume another 11*(800/2)= 4400 orders (containing Mercuries as well).
According to knc-staff Jupiter will have at least 500GH/s [3].

So let's calculate with 0,5TH/s as an average per order: (800 + 4400) * 0,5TH/s = 2.600TH/s
Maybe even more since there are orders with 6,8,10 Jupiters [unverified]. Cloudhashing wanted to buy from knc as well [4], don't know whether they actually pre-ordered. If so I guess it's one huge order.

Those days knc is shipping ppl will think the network-graph must be broken as it is showing hashrate increase at an angle of 88° or so until October 15th when last ones are shipped [5]. At least it will not be comparable to anything we saw before during this time x_X



[1] https://www.kncminer.com/news/news-32
  "Finally we wish to confirm that yes we are still on track for our delivery towards the end of September."

[2] https://www.kncminer.com/news/news-20
  "The number was 890 but has dropped a little as not all bank transfers arrived in time."

[3] http://forum.kncminer.com/forum/main-category/hardware/107-psu-recommendation-jupiter-the-largest-planet?p=1050#post1050
  "According to our engineers an 850W, 80+ Gold certified PSU should be enough for a Jupiter running at the expected speed of 500 GH/s."

[4] https://cloudhashing.com/blog/13-blog/press/65-bitcoin-mining-as-a-service-maas
  "He (Cloudhashing’s founder, Emmanuel Abiodun) is also in negotiation with KNC, which he says is pledging September delivery for its own ASIC products."

[5] https://www.kncminer.com/news/news-33
  " All current paid orders will be shipped no later than October 15th.
    All new sales will reflect our November price point of;"


*edits:
-added [speculation] and [unverified]
-added an additional link
-made it more readable

Cloudhashing has 110 TH from KnC promised to be online by 9/26, and at least another 50 TH coming online from KnC for October.

[DeathAndTaxes]

KNC Sept/Oct 500TH/s is way too low, because September are day1+2 [1,speculation] only and ~800 orders [2].
Most of the orders contain at least one Jupiter, few orders contain one Saturn which has not upgraded meanwhile [speculation].
October orders are shipped on 11 business days, so lets assume another 11*(800/2)= 4400 orders (containing Mercuries as well).
According to knc-staff Jupiter will have at least 500GH/s [3].

So let's calculate with 0,5TH/s as an average per order: (800 + 4400) * 0,5TH/s = 2.600TH/s
Maybe even more since there are orders with 6,8,10 Jupiters [unverified]. Cloudhashing wanted to buy from knc as well [4], don't know whether they actually pre-ordered. If so I guess it's one huge order.

Those days knc is shipping ppl will think the network-graph must be broken as it is showing hashrate increase at an angle of 88° or so until October 15th when last ones are shipped [5]. At least it will not be comparable to anything we saw before during this time x_X



[1] https://www.kncminer.com/news/news-32
  "Finally we wish to confirm that yes we are still on track for our delivery towards the end of September."

[2] https://www.kncminer.com/news/news-20
  "The number was 890 but has dropped a little as not all bank transfers arrived in time."

[3] http://forum.kncminer.com/forum/main-category/hardware/107-psu-recommendation-jupiter-the-largest-planet?p=1050#post1050
  "According to our engineers an 850W, 80+ Gold certified PSU should be enough for a Jupiter running at the expected speed of 500 GH/s."

[4] https://cloudhashing.com/blog/13-blog/press/65-bitcoin-mining-as-a-service-maas
  "He (Cloudhashing’s founder, Emmanuel Abiodun) is also in negotiation with KNC, which he says is pledging September delivery for its own ASIC products."

[5] https://www.kncminer.com/news/news-33
  " All current paid orders will be shipped no later than October 15th.
    All new sales will reflect our November price point of;"


*edits:
-added [speculation] and [unverified]
-added an additional link
-made it more readable

Very good.  Here is another link (distribution of 141 orders)  https://bitcointalk.org/index.php?topic=249065.0. if you want to add it to your cites.  It works out to 495 GH/s per order which I will round up to 500 GH/s.  I don't think 11 business days in Oct means they need to ship 400 units * 11 days = 4,400 unit backlog.  I think KNC is smart enough to give them some slack and then never "ran out" of Oct orders they simply closed it and pushed the delivery of new units back to Nov when the price dropped.  I think this was for two reasons a) to not have their customers ask for refund and reorder (if you are late order paying full price why not cancel and maybe get a unit a day or two later at reduced price) and b) to avoid "burning" their customer loyalty.  So I think assuming full queue 11 days * 400 is probably an overshoot still I do like the math and analysis.  I will discount it slightly to 4,000 orders today (sept/oct) with average of 0.5 TH/s which gives us 2,000 TH/s.  I have a high confidence it will be at least that much.  It may be more but I will update it to that for now.

[geofflosophy]

Very good.  Here is another link (distribution of 141 orders)  https://bitcointalk.org/index.php?topic=249065.0.  It works out to 495 GH/s per order which I will round up to 500 GH/s.  I don't think 11 business days in Oct means they will ship 800 units per day everyday.  They simply closed the Oct window when they announced lowered Nov prices so I think we can discount that 10% or so.  If we figure a conservative 4,000 orders @ 0.5 TH/s per order = 2,000 TH/s.  I have a high confidence in 2 TH/s it may be more but I will update it to that for now.

I think this is a solid estimate. Many orders were also Mercury and Saturn, keep in mind, but many orders were also multiple Jupiters, I think it's fair to assume that these balance out.

[Ytterbium]

a) We know it is 55nm compared to 110nm.  Die efficiency should be 4x that of v1 Avalon.
...
Avalon current design is a single hashing engine (1 hash per clock) running at up to 330 Mhz for 330 MH/s.  The current chip is on a 110nm process and has a die size of 16.12 mm^2 which means a die efficiency of 20.5 MH/s per mm^2.  A perfect die shrink would make that ~80 MH/s per mm^2.  For comparison BFL is ~70 MH/s per mm^2 and Bitfury is 138 MH/s per mm^2 in efficiency.  To give Avalon some margin lets say they improve die efficiency to 100 MH/s per mm^2 (improved clock speed, tweaked hashing engine).

I don't think we should just assume a straight die shrink efficiency gain.  Labcoin is 110nm and is claiming around ~2gh/s. on a 6.5x6.5mm 130nm die.  That's 47.6 Mh/mm² with 1.39 times the features per area. So scaled to 110 it would be 66Mh/s/²  Doing a 2x die shrink would mean 4x the performance, or all the way up to 266mh/s/².

And, on top of that, the smaller the transistor the less time it takes to, well transition. That means both power savings (because you lose less energy waiting for the cmos gates to close and, of course improved performance. So you could theoretically  boost performance even more.

So, if Avalon used it's money to hire some new, higher-skilled chip designers, they could potentially get a lot more performance out of their next-gen chips.

[DeathAndTaxes]

Well I gave them a 25% bump over straight die shrink.  As for Labcoin well I don't believe their claims but am willing to be surprised if/when they show hardware operating at that efficiency.  Still even if Labcoin can achieve that level of die efficiency it doesn't mean competitors will.  For example if you take Labcoin's claim 130nm to 55nm is actually (130/55)^2 = 5.5x reduction in die space.  Throw in 20% more for higher clock (faster switching time due to smaller process) and you are looking at closer to  314 MH/s per mm^2.  When you see that Bitfury is "only" 132 is starts to show how optimistic Labcoin's claims are.

Granted I don't know they can't do it but great claims requires great proof.   Labcoin's die efficiency claims are 30% to 70% better than any other announced specs when normalized for transistor size.

[Delmonger]

I think you may be underestimating KnC as they pan to mine on their own devices and we have no idea how much that will add.

True however we need to start somewhere.  Who knows what the future hashrate will be BUT if we know 6 PH/s has been pre-ordered then we know it will at least rise by that much.  IF we believe it is realistic that will be deployed by end of Dec then that lets of start to see growth of the network.

We kinda do know actually and is a good consideration:

https://www.kncminer.com/pages/faq

"We will not mine with more than 5%"

[sobani]

KNC Sept/Oct 500TH/s is way too low, because September are day1+2 only(?) and ~800 orders [2].

[2] https://www.kncminer.com/news/news-20
  "The number was 890 but has dropped a little as not all bank transfers arrived in time."

I don't read that news item as "890 orders will be day 1+2 deliveries". I read that as "The priority of the first 890 orders is based on order date and not pay date, like the rest of the orders will", so it might not be that bad.

Although looking at reference 6 (KnCMiner list of orders) though, I see day 2 deliveries all the way up to order 25xx, which could make the estimate of total KNC hashing power way more scary.

[DeathAndTaxes]

The 25xx includes unpaid orders.   There were ~800 orders shipped out on day 1 and day 2 based on the cutoff (purchase) date for day 2 shipping.

[eiliant]

I will discount it slightly to 4,000 orders today (sept/oct) with average of 0.5 TH/s which gives us 2 TH/s.  I have a high confidence in 2 TH/s it may be more but I will update it to that for now.

2 PH/s?

[World]

Update from Bitmine:
Shipping of the miners is starting from the 10th of December 2013
http://www.prweb.com/releases/2013/9/prweb11114003.htm

[btc_uzr]

regarding cointerra, you could link this source, too.

Last updated September 12, 2013 00:32

In December this year, we plan to deploy approximately 2,000 TH/s. "

**if you consider pre-ordering from them, this might be a good starting point for research:
https://bitcointalk.org/index.php?topic=269093.msg3138029#msg3138029
 

[Puppet]

these seem to be missing:

http://virtualminingcorp.com/
No idea of volume, but they sell seriously large machines (up to 25TH) and  they claim "Shipping Starts October/November 2013"
If they can get even close to that, its probably reasonable to assume they will have ordered a significant number of wafers (they would be crazy not to) and they may even challenge Cointerra for volume.

Note fast-hash=!hashfast

Also missing afaict:
https://www.xcrowd.co.uk/
again, no idea of volume, Im guessing no where near the other big players, as it seems too little too late, but shipping supposedly starts this year
"Bulk shipping start date - 20-12-2013 200 or 20% of total orders to be shipped per week"

One last remark; the exact timing of the global hashrate increase is of little importance for those not mining yet or not mining very soon. If all those 28nm asics end up being a few months later than projected, it means customers (on average) will receive them that much later, but also that the associated difficulty increase will be delayed. In the end its a wash on average. Obviously what does matter greatly is difference in shipping dates and execution between companies, but its pretty much impossible to guess who will meet their goals and who wont (and by how much they miss it). May as well assume promised delivery date, or apply the same delay to everyone.

[Ozymandias]

http://virtualminingcorp.com/
No idea of volume, but they sell seriously large machines (up to 25TH) and  they claim "Shipping Starts October/November 2013"
If they can get even close to that, its probably reasonable to assume they will have ordered a significant number of wafers (they would be crazy not to) and they may even challenge Cointerra for volume.

Note fast-hash=!hashfast

D&T mentioned Active Mining (AMC/VMC I don't care enough at the moment to remember which does what now a days) in the OP but used the hashing power of a single maximally upgraded unit. If I had to guess, and this is a complete out-of-my-ass guess, since Active Mining will be getting its first round of 28nm chips in November (with high volume coming Januaryish), I would say that by the end of Q1 2014 it will have brought at least a couple (2-3) of ph/s to the network (assuming all goes as planned and on time - ha ha).

Disclosure, I own ActM shares so my guess is likely overflowing with bias - unintentionally of course.

[Puppet]

D&T mentioned Active Mining (AMC/VMC I don't care enough at the moment to remember which does what now a days) in the OP but used the hashing power of a single maximally upgraded unit.

Indeed, its the same company now, I totally missed that.

It makes no sense to estimate their total output at 24TH though, considering thats the output of only a single of their top end machines. A very conservative guess would be at least 10x higher, and more likely 100x higher, though how much of that will happen in 2013 is probably anyone's guess.

[DeathAndTaxes]

D&T mentioned Active Mining (AMC/VMC I don't care enough at the moment to remember which does what now a days) in the OP but used the hashing power of a single maximally upgraded unit.

Indeed, its the same company now, I totally missed that.

It makes no sense to estimate their total output at 24TH though, considering thats the output of only a single of their top end machines. A very conservative guess would be at least 10x higher, and more likely 100x higher, though how much of that will happen in 2013 is probably anyone's guess.

Yeah not sure where that came from likely me misinterpreting some earlier cite.  I decided to move ActM, XCrowd, BTCGarden, and Labcoin to the "announced but unsure" category.  I just have found it very difficult to get any real information on the companies.  Some of them in my opinon, including ActMam, have circumstances which make me question if they will ever get to market.  Still if someone can find more details (tape out, contracts, designs, chip volume) I would be happy to be corrected.  To be perfectly honestly I haven't looked very close at AMC/VMC?, XCrowd, BTCGraden, or LabCoin.  They all appear to be very opaque and I would be willing to bet at least one if not all four never make it to market in any reasonable volume.

[Puppet]

D&T mentioned Active Mining (AMC/VMC I don't care enough at the moment to remember which does what now a days) in the OP but used the hashing power of a single maximally upgraded unit.

Indeed, its the same company now, I totally missed that.

It makes no sense to estimate their total output at 24TH though, considering thats the output of only a single of their top end machines. A very conservative guess would be at least 10x higher, and more likely 100x higher, though how much of that will happen in 2013 is probably anyone's guess.

Yeah not sure where that came from likely me misinterpreting some earlier cite.  I am going to move it to the "planned" section because honestly I don't have a high confidence they will release anything.  Their pricing seems incredibly dubious given they are targeting miners looking to deploy 24 TH/s not a handful of USB miners.  Also looking at their website it seems they are going the insanely complex route with multiple base models, and then different expansions levels with multiple boards and multiple expansion chassis. 

Given Bitcoin mining is an "embarrassingly parallel" problem all that complexity and options makes me think it is very poorly thought out and that might mean other problems.  Make a single unit, make it efficient, make it cost effective.  If you sell a 2TH/s unit cheaper than anyone else and someone wants 20 TH/s they can just buy 10 rigs.

Still if someone can find more details (tape out, contracts, designs, chip volume) I would be happy to be corrected.  Honestly I haven't looked very close at AMC/VMC?, XCrowd, or LabCoin.  They all appear to be very opaque and I would be willing to bet at least one if not all three never make it to market in any reasonable volume.

VMC/AMC seems legit. Have a look here:
http://www.easic.com/vmc-uses-easic-to-achieve-24-756-ths-bitcoin-miner/

easic is a real company, and they have designed the chip for them.

My guess is xcrowd is legit too, although I doubt they will have a compelling product in time.

Havent looked in to labcoin, but there seem to be plenty of posters who have, and have been in contact with them and are convinced its no con. That doesnt guarantee a timely product within spec, but Id expect them to ship something someday