HashFast announces specs for new ASIC: 400GH/s

[minternj]

There are 2 pcie connectors on the pcb on each side, look at the pics of the board from earlier in this thread.

https://bitcointalk.org/index.php?topic=262052.msg3929934#msg3929934

[1714042648]

1714042648

[1714042648]

1714042648

[1714042648]

1714042648

[1714042648]

1714042648

[HarrisonS]

There are 2 pcie connectors on the pcb on each side, look at the pics of the board from earlier in this thread.

https://bitcointalk.org/index.php?topic=262052.msg3929934#msg3929934

2 connectors at 75W each is still only 150W. In order to run all 4 dies (600W) they would need to put 4x the rated current through those connectors.
They are using CoolIT CPU coolers to cool this thing off. CPU's use at most 130W. They are going to need to dissipate 4-5x as much energy as the coolers were designed for. The blog post says the 2 dies are running at 78 degrees, add another 2 and they'll be running way hotter.
I have the feeling that these things are going to fail pretty quickly unless they tone down the power usage.

[minternj]

It doesn't mean much at this point. Not enough information to make any conclusion. Wait until they get all 4 dies up and running as intended. Tick tock, halfway through December already..

[jspielberg]

Hopefully they will provide a substantive update shortly.  Seems like they are preferring twitter (which seems like a good tactical move considering the animosity towards them here).

When there are no updates with good news ( or even updates with problems they are trying to overcome), it appears that either they don't care, or they aren't doing anything.  Both of which we know can't be the case (or at least not what they are trying to communicate).

The troops are getting restless again....

... or maybe they don't work on weekends

[ImI]

i seriously think they dont work on sundays.

[minternj]

Well the way that hashfast_cl has been posting in here, you'd think they are pretty much done. But yes i also think they took the weekend off, probably because their pcb assembly house Sonic manufacturing doesnt work weekends. I think the Australia team is out surfing or hunting great whites too.

[RHA]

Again: PCI-e connector CAN supply 300 W.
6-pin version is rated 75 W, but 8-pin version is rated quite higher.
Additional two pins (both are ground) are used not for supply, but to let the receiver (usually a graphic card) to recognize if this PCI-e connector is able to supply more. Still they don't need to be connected.

Worse thing is with cooling. Let's read HF's own announcement: https://hashfast.com/baby-jet-thermal-testing/

Baby Jet thermal testing

    Posted on September 23, 2013
    by Simon Barber   
    in Baby Jet, Development, Rig Assembly   

Our liquid cooling solution supplier has the results from some real world testing of the solution we will be using in the Baby Jet. The HashFast GN ASIC was designed around a nominal operating point of 400GH/s, which it achieves consuming 250W at the chip. Due to manufacturing variations in the silicon some chips will consume slightly more, and others less. We chose this 250W operating point because that is what overclocked CPUs are routinely run at, so we knew common off the shelf cooling solutions would be able to handle this. We asked our cooling solution provider to do some thermal simulations of the system, and they confirmed that it should be able to handle 350W of heat. We have now got the results from a real system test. Using the exact final chassis that the Baby Jet is using, together with the final cooling solution. A small heating element, the same size as our chip was ordered, and attached to the cooler. The experimental results showed that the chip’s case temperature could be maintained at an acceptable level while the cooler dissipates 400W of heat. This means that our chip should have a *lot* of room for overclocking above it’s nominal 400GH/s.

[RoadStress]

i seriously think they dont work on sundays.

Sucker webmaster is working on sundays, while sluggish engineers don't.

[DeathAndTaxes]

On the power & cooling.   This won't be the first time in the first thread I have had to point out that a rig using X watts at the wall doesn't mean the chip is using X watts (and the chip is the only thing attached to the radiator).

Say chip uses 400W @ ~1V.  No ATX PSU has a 1V output so the board needs to convert the 12V from PSU to the voltage used by the chip.  If the VRMs are 88% efficient then to supply 400W to the chip @ ~1VDC would require 454W @ 12VDC.  Now add to that 2 radiator fans + intake fan + water pump + host/controller for another ~50W so in this scenario rig uses ~500W @ 12VDC.  Now the ATX PSU which converts 120/240 VAC to 12VDC is only ~90% efficient so 490W @ 12VDC = ~550W to 560W @ 120VAC.

[DeathAndTaxes]

I just realized this, 6pin PCI-e connectors are only rated for 75W, and you could probably get MAX 200W out of them.

The PCIe graphic card standard (which has nothing to do w/ ASICs) limits power to 75W on 6 pin connector and 150W on 8 pin connector (using no more power pins) however those are artificial limitations.   The actual connector is good for 9A per pin, 3 power pins (both of 6 & 8 pin connector), 12 V means 324W (9*3*12) per connector max or 648W for a pair.  

Now how much power a particular PSU can push to the connector is another question.  The VRMs on the board also have a max current limit so the chip can only be provided so much power but the connector is a total non-issue.

[CYPER]

I just realized this, 6pin PCI-e connectors are only rated for 75W, and you could probably get MAX 200W out of them.

The PCIe graphic card standard (which has nothing to do w/ ASICs) limits power to 75W on 6 pin connector and 150W on 8 pin connector (using no more power pins) however those are artificial limitations.   The actual connector is good for 9A per pin, 3 power pins (both of 6 & 8 pin connector), 12 V means 324W (9*3*12) per connector max or 648W for a pair.  

Now how much power a particular PSU can push to the connector is another question.  The VRMs on the board also have a max current limit so the chip can only be provided so much power but the connector is a total non-issue.

Now can you please provide data for the maximum 24/7 safe current per 18AWG wire.
My own research led me to believe it is much lower than 9A.

[itod]

i seriously think they dont work on sundays.

+1

[DeathAndTaxes]

Now can you please provide data for the maximum 24/7 safe current per 18AWG wire.
My own research led me to believe it is much lower than 9A.

It depends on the length of the wire, and the acceptable voltage drop.

At 3 ft circuit legnth, 18 AWG is good for 9A with a 2.83% voltage drop.  As for safe, at 2.83% voltage drop that would only 1.5W dissipated as heat.  Still the point was somewhat academic I doubt HF will be pushing anywhere close to that much current but if they did the limiting factor wouldn't be the artificial PCIe graphics standard for 6 and 8 pin connectors.

[arorts]

Hashfast Countdown!

http://www.timeanddate.com/countdown/to?iso=20131231T235959&p0=224&msg=Hashfast+Shipping+Deadline!

That should be deliver "deadline *".  The "deadline" (until they decide to retroactively change it after the fact) is based on delivery date not shipping date.  

* Of course the deadline for pre Aug 15th orders has long since passed, the company however has fraudulently decided to steal customers funds by not providing refund in violation of their own pre Aug 15th TOS.

Correct. Delivery. Not shipping BS. I have a copy of their TOS and archive version in case they decide to lie and screw people. They should be serious about their TOS.

[arorts]

Irony, it now seems you all are seemingly disappointed IF you recieve your miner BEFORE New Years....

Don't the Terms of Agreement/Service state you have to notify HashFast with some letter that you have not recieved your order "within a reasonable period of time"?

If you don't receive your unit by 11:59:59 PM on Dec 31st, HF, shipping carriers and the world has a proof record of non-delivery by that time. Then you can immediately send your letter since you necessarily needed to wait until that time in case you want a refund. Some people might be totally ok receiving unit later. It's up to people to individually make that refund choice if they are entitled to it, not a forced refund on people.

[cedivad]

However filling that letter will almost certainly be equivalent to deciding to proceed legally against HF, because i can't find any scenario where they will decide to refund the BTC we paid even if they failed to ship.

And yet, thinking that HF will fail to ship by the end of the year is daydreaming.

[jspielberg]

So... Does anyone have experience with water based coolers like those that are being used?

Are they loud?  Do you guys think they will be as loud as the avalon's?
I am currently very happy with my fanless bitfury setup.

[minternj]

So... Does anyone have experience with water based coolers like those that are being used?

Are they loud?  Do you guys think they will be as loud as the avalon's?
I am currently very happy with my fanless bitfury setup.

The water pump is pretty much silent. The fans that move air past the radiators are what will make noise. You can always swap them out for quieter fans. Without knowing what model fans this system comes with we wont know how loud they are. It really should be a very quiet system though.

[aerobatic]

So... Does anyone have experience with water based coolers like those that are being used?

Are they loud?  Do you guys think they will be as loud as the avalon's?
I am currently very happy with my fanless bitfury setup.

The water pump is pretty much silent. The fans that move air past the radiators are what will make noise. You can always swap them out for quieter fans. Without knowing what model fans this system comes with we wont know how loud they are. It really should be a very quiet system though.

if its anything like the typical pc installation the fans will be temperature sensitive and spin faster when its hot and go slower (and quieter) when its not hot.   since its likely that most people will want it to overclock to the max, and run very hot i think its safe to assume the fans are going to be spinning pretty fast all of the time.  but if the noise is really a problem you can always under clock it and have a quieter but less profitable system.  (no one's going to seriously do that are they?)

the only reason you can run your bitfury's without a fan is because they're not running that fast, or your ambient air is very cool.

you could always switch out the fans they give you with quieter ones, but they probably won't perform as well so you'd almost certainly be crippling the performance if you do that.  you make a choice, fastest possible, whatever the noise... or, low noise and lower performance.

im not expecting either hashfast nor cointerra systems to be as quiet as knc systems.. simply because the amount of watts of heat that needs to be dissipated is a lot more... that said, i suspect they're still quieter than avalons.

in noise per gigahash i am expecting both the hashfast and cointerra systems to be very efficient (considering their performance).  but in absolute terms, they probably won't be silent!   cointerra's has a larger radiator (almost the full size of the rear) so that may give them an edge in not needing as fast a fan speed, but until we see both systems running... and hear them... we won't know.

and anyway, both these systems deserve to be run in a data center!

[3DX4D]

that said, i suspect they're still quieter than avalons.

A god damn freight train is quieter than my neighbours B1 Avalon.  LOL