HashFast announces specs for new ASIC: 400GH/s

[DPoS]

be sure to clear your calendar for the next three weeks and forgo sleeping as you nursemaid these babies

[1714040586]

1714040586

[1714040586]

1714040586

[1714040586]

1714040586

[cedivad]

So a sierra does 1200GH and a BJ 7/800? We are only losing time with all of this speculation, but why should we accept the delivery if they don't even tell us real numbers? (i'm supposing that the EXWORK thing isn't something they can appeal at)

[minternj]

be sure to clear your calendar for the next three weeks and forgo sleeping as you nursemaid these babies

I have my teat ready.

[perezoso]

The cold fusion reactor produces more than enough power for 1.2th.  The Seasonic thing is a new, uber-secret unicorn generation device.

I still think they popped another board into the BJ for testing.  If, in the real world,  the additional hardware farked up their airflow or overtaxed some component, and there's a fix, they want to know now. 

Who else here is planning to ditch the Raspberry Pi and use another controller?

[jspielberg]

Because they told us so.

-Instigate your customers to suppose what you want to tell but can't
-Don't deny it
-Win

It looks like that everyone in this thread is sure that a single board does 800GH+. They told us that the 1200GH is a single BJ, so why we shouldn't believe them? I'm so tired of this...

I must have missed something. Where did they state that the 1200GH/s was a single BJ?

They implied 1200 GH/s for a single machine because they posted this tweet:
https://twitter.com/HashFast/status/416826349337460736

The implication was that this machine was a single GN chip.

[fractal02]

I must have missed something. Where did they state that the 1200GH/s was a single BJ?

They said that the pool address of their twitter feed is connected to a single BJ. If they connected a sierra instead, they should have told us, they didn't.

Maybe you need chill a little...you startin' to write pretty big BS...

[paranoidx]

A BJ does ~600 gh/s from my reading of the eligius charts.  On the date of December 27th before the eligius was having issues it was reporting highs of 800 gh/s and lows of 400 gh/s with an average of about 550gh/s-600gh/s if you look at the chart.  This would mean that the 1200 gh/s is actually two bj's.

[jjiimm_64]

Thanks, now we know their yield per wafer is around 110Th/s (@ 700Gh/s per GNonce board). Now we just need to know roughly how many wafers per batch and how often the batches will be and HF's contribution for the next 12 months will be established. Icedrill purchased 500Th/s worth of sierras and there were 550 Batch 1 Baby jets. It looks like (with a guess in the dark) each batch HF will bring online around 20-25 Wafers per month worth of chips, or roughly 1.75-2.75Ph/s depending on good chip yield (number of good chips per wafer).

So you can expect the hash rate for hashfast to increase the network hash rate by roughly 2.25Ph/s every month for the next 12 months in your calculations.
Bitfury is bringing online about 1Ph/s per month (2Ph/s every 2 months). So we're looking at 4.25~Ph/s per month with just bitfury and hashfast.
It looks like KNC is going to be bringing about 3Ph/s per "wave", but how often they will do these waves is not clear. Write them down for 2Ph/s a month and we're up to 6.25~Ph/s a month growth per month.

My current growth figures are on track (at least until cointerra, bitmine.ch, etc start shipping). But, I'm betting that hashing power won't hit the network until feb/march.

the one thing inoticed about this post.... not one word of bfl

[cedivad]

Maybe you need chill a little...you startin' to write pretty big BS...

I had some great fpv flights today. Where is my BS?

[paranoidx]

You guys are being ridiculous if you think a single BJ can put out 1200 gh/s.
You went from a mob wielding pitchforks to ecstatic optimistic fools with dollar signs in your eyes.

[minternj]

Im expecting 400-500. At this point i want to be done with this company. Now where are the fucking MPP chips?

[cedivad]

@paranoidx,
I don't, actually, it's what i'm trying to point out. The problem is that many in this thread will believe that.

[Puppet]

Thanks, now we know their yield per wafer is around 110Th/s

I'm terribly sorry that i made you lose the time needed to write all of that, but that is Cointerra chip, not Hashfast's one.

Doesnt make too much difference. Hashfast published their die size. Each chip has 4 dies of 81mm² IIRC. A 300mm wafer would hold ~800 die candidates for up to 200 chips x 500GH each = 100TH.

[jspielberg]

Well... I think at this point... the only thing we can be sure of is...
... the person tweeting for HF has very little or no knowledge of the actual testing.

The tweets... "ooohhh... thats fast... see how fast it is..."
and then a separate tweet:
"... wait a minute... that can't be right?! we don't know what the speed is "

Where is Simon?  And why isn't he tweeting for HF?

[perezoso]

Where is Simon?  And why isn't he tweeting for HF?

The CEO and fisting porn purveyor marketing department have banned him, since his early chip testing tweets were too frank.

You guys are being ridiculous if you think a single BJ can put out 1200 gh/s.
You went from a mob wielding pitchforks to ecstatic optimistic fools with dollar signs in your eyes.

I certainly don't think a 1 board BJ could put out 1200gh, but I can see a recent average around 1100-1150 that I think could be a BJ with two boards or, as others have pointed out more pessimistically, a Sierra.

But none of that means I've remotely forgotten about the October lies, nor that I plan to let HF off the hook for them. Pitchfork is simply resting while we watch the drama unfold.

[aerobatic]

Thanks, now we know their yield per wafer is around 110Th/s

I'm terribly sorry that i made you lose the time needed to write all of that, but that is Cointerra chip, not Hashfast's one.

Doesnt make too much difference. Hashfast published their die size. Each chip has 4 dies of 81mm² IIRC. A 300mm wafer would hold ~800 die candidates for up to 200 chips x 500GH each = 100TH.

would like to know for sure that the die is 81mm²... dies are rarely exactly 9x9 mm... so it'd be nice to know exactly how big it is to at least one decimal place rather than the very rough and unlikely sounding 9x9mm number.  e.g.,if it was 9.5x9.5mm that'd be a 90mm² die.

makes a bigger difference when we're trying to calculate the gh per mm²

i also heard for instance and by comparison that knc has a 6.6mm x 6.6mm die = 43.5mm², which hashes at 650 GH for 4 chips, and each chip has 4 dies.. thus each die hashes at 40.625 GH...  thus the gh/mm is 0.94 gh/mm²

[Puppet]

would like to know for sure that the die is 81mm²... dies are rarely exactly 9x9 mm... so it'd be nice to know exactly how big it is to at least one decimal place rather than the very rough and unlikely sounding 9x9mm number.  e.g.,if it was 9.5x9.5mm that'd be a 90mm² die.

I got that number by combining two different posts.
The first one is a bit scary:
https://bitcointalk.org/index.php?topic=270384.msg2975145#msg2975145

If you know anything about asics, that post may make your heart stop once you notice its written by Hashfast's VP of engineering. Estimating a competitors die size just by looking at the heatspreader...   that was my first major red flag.

Anyway, the number Amy wrote there is 324mm^2 per "die". Later it was clarified that each chip consists of 4 smaller dies in a single package.  Sorry, dont have the link at hand. I just divided  324 by 4.  But again this makes you wonder if VP Amy had absolutely no clue what she was talking about, not knowing what a die is, or if they changed this since the above post and initially they were crazy enough to actually shoot for a  300+ mm³ mammoth die. Either should have made potential customers worry.

[jspielberg]

would like to know for sure that the die is 81mm²... dies are rarely exactly 9x9 mm... so it'd be nice to know exactly how big it is to at least one decimal place rather than the very rough and unlikely sounding 9x9mm number.  e.g.,if it was 9.5x9.5mm that'd be a 90mm² die.

makes a bigger difference when we're trying to calculate the gh per mm²

i also heard for instance and by comparison that knc has a 6.6mm x 6.6mm die = 43.5mm², which hashes at 650 GH for 4 chips, and each chip has 4 dies.. thus each die hashes at 40.625 GH...  thus the gh/mm is 0.94 gh/mm²

Amy from HF wrote about this a couple of months back.

HashFast's Golden Nonce chip: I don't have to estimate the size because I work at HashFast.
   One 18x18mm die is able to do 400 GHash (nominal - more overclocked**)
   Hashing per square mm:
      18x18mm = 324mm^2
      400 GHash / 324mm^2 = 1.23 GHash/mm^2

So I guess the 18x18 is really 4x9x9

[Puppet]

Ah found the other post I was looking for:

There are 4 dies, each 9mm x 9mm, spaced out by 5mm

https://bitcointalk.org/index.php?topic=270384.msg3195006#msg3195006

[Anddos]

https://twitter.com/HashFast/status/417022058908770304
"Tests came out very well tonight. Production has been given the go-ahead!"

What night are they talking about since that they are in the pacific side of the USA? (this bothers me since the beginning of their chip updates)

dosent make sense to me to,it would of been daytime,so they must ment last night