Author

Topic: HashFast announces specs for new ASIC: 400GH/s - page 546. (Read 880461 times)

donator
Activity: 1218
Merit: 1079
Gerald Davis
It is a good thing they are still way above cost, cause the worth of such machines will drop like a stone.

Well Avalon charges $9 per chip (AFAIK someone correct me if I am wrong) and that is obviously inflated 1000% or more over silicon cost and that cost will sink like a stone.  So you could remove $2400 from the system by assumming the price of chips in bulk drops from $9 to $1 to remain competitive.  However my point was that even doing that looking at an Avalon, the balance of the system cost (PCB, heatsinks, assembly, DC to DC power supply, power distribution board, controller board, even mundane things like fans, power supply, and cases) isn't being marked up 1000%.  Excluding the cost of the chips you couldn't drop the cost on the rest of the system 90%.   So while the value (and price) of older chips will indeed fall like a stone the BOM (balance of materials) cost will subdue the price drops.

On a related topic this is why I believe those that think $1/GH complete miners are reasonable are just silly.  Still in full disclosure I own miners, shares, and pre-orders with multiple companies but not Avalon so I may just be biased.
donator
Activity: 1218
Merit: 1079
Gerald Davis
Product who has for order? 

I updated the post.  The product in question is the one available from the person I was responding to.
legendary
Activity: 1890
Merit: 1003
But I agree, it's nice to see different designs popping out, we followed Bitfury's approach too for our ASIC and we're confident that the road to success is huge grids of small and cool chips.

$6250   700W       85GH/s      8.23W per GH   $73 per GH
$5600   350W      400GH/s      0.88W per GH   $14 per GH

Product who has for order?  I only paid $1500 for my B2 Avalon - $18/Gh/s.  Delivered In June.

Only a few dollars more expensive per Gh/s then HashFast. Clearly they're not charging anywhere near their actual mfg cost.
B1 Delivered in March.

It is a good thing they are still way above cost, cause the worth of such machines will drop like a stone.
full member
Activity: 238
Merit: 100
But I agree, it's nice to see different designs popping out, we followed Bitfury's approach too for our ASIC and we're confident that the road to success is huge grids of small and cool chips.

$6250   700W       85GH/s      8.23W per GH   $73 per GH
$5600   350W      400GH/s      0.88W per GH   $14 per GH

Product who has for order?  I only paid $1500 for my B2 Avalon - $18/Gh/s.  Delivered In June.

Only a few dollars more expensive per Gh/s then HashFast. Clearly they're not charging anywhere near their actual mfg cost.
donator
Activity: 1218
Merit: 1079
Gerald Davis
D&T - are you comparing a 110nm asic to a 28nm asic?  Would probably be more realistic to at least compare against bitfurys 55-ish nm chip.

I am comparing the product they have available for order.  If they have a bitfury based product I would have used that.

Still maybe I wasn't clear.  A 4 board avalon uses 300 chips correct?  They are currently excessively marked up ~$9 ea so lets pretend the price was dropped to ~$1 per chip.  That would remove $2400 from the cost of the system.  It is easy to see that either the seller has massive profit margin (I doubt it since given the Avalon uncertainty they would prefer to see out of pre-orders) or the BOM (balance of material) on that "huge number of low powered chips" contributes significantly to the overall costs.

Lets look at GPUs.  With crossfire (some software magic) 2 GPU can split workload.  A 7990 is essentially two downclocked 7970s internally crossfired.  AMD doesn't try to make a monster single chip 7990 because rightly so there is point where you get diminishing returns on larger and hotter chips.  However cross fire isn't limited to two chips.  Why not make a single chip of reasonable middle range computing power.  Then you could put one chip on a board and call it a 7950, two chips on a board and call it a 7970 and 4 chips on board and call it a 7990.  One inventory part, 3 products.  There was some research into that (and I am sure AMD/NVidia still are) but having a single unified chip reduces the post-fab assembly production/yeild issues.

All I am saying (now in an excessively verbose way) is there are pros and cons to each approach.

Take KNC for an example they use 4 chips on their top of the line model.  Aha! See even KNC is taking the "more less powerful chips aproach" leaving only HF taking the path less traveled.  Except not quite.  KNC chip uses roughly the same power as HF.  The need to use 4 chips is simply due to a less efficient architecture*. KNC isn't using 3, or 2, or 1 chips because at 250W they (and HF) are reaching the limit of what can be cooled realistically.


*Before the KNC warriors attack me this isn't an insult.  Just comparing reported specs.  It may be that KNC is overly cautious on power projections and HF is overly optimistic.  It may be that KNC decided they could make a more efficient chip but time to market was more important and 2.5W per GH was "sufficient".  Whatever the reason my point is that KNC needs 4 chips to acheive the same hashing power not because they are taking a "more and cooler" aproach but because that is where they maxed out at.  If they were 4x as efficient they would be using a 400 GH/s chip as well.
full member
Activity: 210
Merit: 100
But I agree, it's nice to see different designs popping out, we followed Bitfury's approach too for our ASIC and we're confident that the road to success is huge grids of small and cool chips.

$6250   700W       85GH/s      8.23W per GH   $73 per GH
$5600   350W      400GH/s      0.88W per GH   $14 per GH

Cointerra:                                                $8/Gh
Butterfly Labs:                                         $8/Gh
XCrowd:                                                  ~$2/Gh

If we're comparing chips that don't exist, let's try to include all of them.
sr. member
Activity: 490
Merit: 255
But I agree, it's nice to see different designs popping out, we followed Bitfury's approach too for our ASIC and we're confident that the road to success is huge grids of small and cool chips.

$6250   700W       85GH/s      8.23W per GH   $73 per GH
$5600   350W      400GH/s      0.88W per GH   $14 per GH

D&T - are you comparing a 110nm asic to a 28nm asic?  Would probably be more realistic to at least compare against bitfurys 55-ish nm chip.
donator
Activity: 1218
Merit: 1079
Gerald Davis
But I agree, it's nice to see different designs popping out, we followed Bitfury's approach too for our ASIC and we're confident that the road to success is huge grids of small and cool chips.

$6250   700W       85GH/s      8.23W per GH   $73 per GH
$5600   350W      400GH/s      0.88W per GH   $14 per GH
full member
Activity: 222
Merit: 100
If Hashfast had designed 4 x 63W chips as I suggested, there would be 4 heatsinks occupying a larger space, hence allowing the use of larger, slower, lower-wattage fans.
More wafer cutting, more chip packaging, more complex boards, more supporting components, more fans, more heatsinks, more assembly, etc.  A larger number of smaller chips gives you more flexibility and makes cooling less of a challenge but the system isn't going to be cheaper or more importantly assembled faster.   Waterblocks are a significant portion of any water loop so if you are going to watercool less blocks is always better (i.e. water cooling two 5970s is going to be cheaper than 4x5870s).

Not really, in fact some IC packages cost much less than others just because they're being used more frequently. A package like the one used by them that can withstand such high heat transfer will be very, very expensive, I would say around $20-$30 just that. Waterblock cooling is an useless overhead dictated only by the need to transfer away so much power in such a tight space, it could be avoided by using many smaller chips that are just air cooled.

Also don't forget that a wafer is a round thing: the bigger the squares you fit on it, the more area you loose on the edges, here again a smaller chip could have been cheaper, not more expensive.

But I agree, it's nice to see different designs popping out, we followed Bitfury's approach too for our ASIC and we're confident that the road to success is huge grids of small and cool chips.
sr. member
Activity: 308
Merit: 250
decentralizedhashing.com
Thanks guys, the page changed so much in the last couple weeks I lost track.
qwk
donator
Activity: 3542
Merit: 3413
Shitcoin Minimalist
sr. member
Activity: 462
Merit: 250
Why are all references of Uniquify removed from the HashFast page?

to make room for the sexy girl pics beggin you to BUY BUY BUY
legendary
Activity: 1764
Merit: 1002
Why are all references of Uniquify removed from the HashFast page?

they're not:  https://hashfast.com/uniquifystatemen/
sr. member
Activity: 308
Merit: 250
decentralizedhashing.com
Why are all references of Uniquify removed from the HashFast page?
legendary
Activity: 3878
Merit: 1193
9. Start shipping  :                               20-30 October

Do You still support initial claim about shipping dates or shipping date is just a wish we all want to come true?

Of course the October ship date is a pipe dream. If they really believed they could really ship in October, they'd offer refunds at that time, not 2 months later like they are doing.
sr. member
Activity: 490
Merit: 255
I am not sure if I see the problem. 

If it gets too hot for the cooler to handle, the chip will heat up,  the thermal control software will disable like 1/2 of the chip, and the temperature will come back down.  Maybe they will put 2 chips in each unit in that case... that is HF's problem to figure out. 

Step 0 have the chip fabricated.
First step, get a chip and characterize its performance.
Second step, produce system with the characterized chips at stated hashrate.
Third ship the system.
full member
Activity: 210
Merit: 100
Important to remember that thus far, we're watercooling & clocking a chip that still doesn't exist.
If HashFast is as fast & loose with guessing their own specs as they are with KNC's (estimating KNC ASIC die size from package dimensions), allow for +-500% error.  While BFL was able to stick on a huge cooler when their power specs proved optimistic, with this chip we're talking power planes hefty enough to run an arc welder & liquid nitrogen cooling.
*Ed Trice, if ever there was an opportunity...
Vbs
hero member
Activity: 504
Merit: 500
1) https://hashfast.com/shop/babyjet/ states 350W power draw (+/- 20%), but the chip only consumes 250W?  100W for cooling/misc?

Not sure why this is still a question (goes beyoond just HF).  Remember even excluding water cooling pumps, fans, and controller boards there are a lot of efficiency loss between between the wall and the chip.

"This is still a question" because Vbs is here to spread FUD about HashFast.

He's on ACTM's Board of Cheerleaders, invested more than he can afford to lose, and got married to a competitor's  stock that not only became obsolete soon after launch but is also enduring disasters WRT BTC conversion and bulk Avalon orders.

At this point, asking dumb questions which have already been answered repeatedly is all he can do to prop up his failing stillborn venture.

I'm sorry, who are you again? This kind of ad hominem argumentation is indeed lamentable, so if you have nothing substantial to discuss, please keep to yourself.

I'll reiterate my question about the power figures: Simon, the store page says "* Real silicon power consumption may vary from simulation results by +/- 20%" (I assume this means the chip can be between 200-300W), you would really have no problem if it ever comes to be 300/324 = 0.926W/mm^2? It's still 12% more than an o/c SB-E even at an extreme 350/425 = 0.824W/mm^2.

Using some quick math for forced convection water cooling, with:
Water:  50 - 10.000 (W/m^2K)
Chip area (m^2): 0.000324
t_surface = 80ºC
t_air = 30ºC

q = 10.000*0.000324*(80-30) = 162W, which is a water heat transfer figure under chip requirements, so the heatspreader has to really compensate for the so small chip area!
mrb
legendary
Activity: 1512
Merit: 1028
You didn't seriously think they were going to scrap a design and make another one because you don't like it do you.

Of course not. I commented, because I want to hear Simon's explanation about his company's choice of making a 250W chip, in order to determine how risky an investment in Hashfast would be.

The company believes they have the expertise to design a large efficient processor.  If they can there is no reason to use a larger number of less powerful chips. The less integration at the chip level the more integration that is needed at the board level. More wafer cutting, more chip packaging, more complex boards, more supporting components, more fans, more heatsinks, more assembly, etc.  A larger number of smaller chips gives you more flexibility and makes cooling less of a challenge but the system isn't going to be cheaper or more importantly assembled faster.

"More wafer cutting, more chip packaging": true, but negligible. A Baby Jet costs $5600. At most, a few extra dollars would be spent to slice and package 4 chips instead of 1 per machine.

"more complex boards, more supporting components": true, but completely negligible. Look at Avalon: each additional chip on a hash unit has merely 13 supporting components (resistors and ferrite beads, all 0402) which cost at most a few cents. An extra bitcoin mining chip added to a circuit in general does not need much supporting components because all it needs is power and a low speed serial i/o line.

"more fans, more heatsinks": false. As you pointed out it's a curve. It's a choice between more heatsink + fewer fans, or less heatsink + more fans. But both solutions, air and water, must be designed to cool 250W regardless.

"more assembly": false. The little extra time of pick-and-placing, say 3 extra bga chips, is completely negligible compared to the extra assembly time required to install the water cooling system. I guess Hashfast could save time by shipping systems with the water cooling completely untested and unassembled. But still, the extra time to pick-and-place 3 extra bga chips on a board with 100+ components remains negligible.

Air cooling is cheaper and faster to assemble. Nobody argues against that. I am surprised that you do(!)
hero member
Activity: 630
Merit: 501
Miner Setup And Reviews. WASP Rep.
They have already said they signed a NDA and can't disclose information like that. Also to do with water cooling you can get better temps meaning the chip can hash faster isn't that what everyone wants?

I would prefer to get response from HashFast but thank You Beastlymac for quick answer.
I was trying to find any info about HashFast NDA agreement originating from HF but cant. Do You suggest that they sign NDA for a whole chain operations? Its ridiculous to me.

Do they act only as an chip designers and whole logistics is handled by mysterious contractor? This brings more and more questions.

If anyone here cares about being sure Hashfast claims are base on truth - please support me.

Hashfast - please find a minute or two to answer me.


I would assume they have signed a NDA with The foundry and uniquify.
Jump to: