Topic: HashFast announces specs for new ASIC: 400GH/s - page 490. (Read 880461 times)

Im a huge fan of liquid cooling... ive been using various corsair models for the last few years on every pc system ive built (20-30)... for my friends & family (purely for fun).

theyre the most efficient way of moving large amounts of heat away from the very hot chips, to the edge of the case where you can deal with it much more effectively, and double the use of your fans since you dont need extra fans... and can use the ones attached to the radiator either as intakes or exhausts for the entire system while also cooling the rads.

by the way, dont forget that the cfm from the fans encounters resistance as it hits the rads, so the cfm on the other side of the rad is quite a bit lower.

i think the warmed up air will certainly be more than a 3 degree delta... but whatever it is (10, 15.. im not sure), i really dont think its too much for the rest of the system to care.

i also dont think that anyone is going to be running these chips at 250 watts!   absolutely everyone will be overclocking & overvolting them as much as they can take... so if they muster it, they will be outputting 350-400 watts !

i dont know how hot hf will allow them to run, but im guessing around 100 degrees is the practical upper limit... heck, if they can quickly modulate the volts/clocks, i suspect they can let them run even hotter.  (115?)

Choosing the twin PS was FU #1.  Absurd.  Eats up space & makes all the rest of the compromises necessary.  Almost.
There's no reason why the twin PS need to live in the back.  They could live in the front, and, if they're already converted with the patented backwards-fan (Cypherdoc, the semi-official ex-paid spokesman, claims that the PSs airflow is reversed, intake where the exhaust usually goes), think the rears of the PS facing the front.  Ugly power cord, but that's kosher.
*But a much saner solution is a single PS, 1600W, and go to town with the free space.  Put the silly thing smack dab in the middle of the case, there's nothing there anyway, and the PS will put out 160W of heat at full tilt.  Easy.

its academic whether it would win the d'Airflow prize d'Elegance....   its simple... once the twin psus were chosen - they take up all the rear space -  there just isnt enough radiator room to have the radiators at the rear to cool all three chips, ergo, they go at the front.   Simple.  a no brainer.  there is no other option.. !

That is the concept.  crumbs likes to scare people with the hot term hot like the air will be blasting hot.  In reality the air would be nowhere even near 40C.  250W per radiator and 50 CFM means a 3C rise in temp.  So 25C in and 28C out.


It is almost certainly OEM version of H60 or one very similar.  Thanks for the links.  While our resident troll is unlikely to stop trolling they will be useful for those who wish to be informed.

One way to look at it is in purely air-cooled setup the air will be warmed up by exactly the same amount.  So air flows over processor heats up and flows over the downstream components.  If slightly warmed air would be so dangerous then no air cooled system would work.  For example many GPU designs cool the ram after the GPU itself.  The air reaching the ram will already been warmed.

So why didn't hashfast just use air cooling?  Simple it is a matter of surface area.  The amount of airflow needed to transfer a given amount of heat (in this case 250W nominal) is inversely related to the surface area.  So larger surface area, lower airflow needed.  The heatload in the chip is very concentrated.  All 250W are produced in less than 1cm2.  Even with a large copper heatpipe type cooler you are talking about a lot of needed airflow.  Also (like seen in KNC design) you run into potential issues of trying to route airflow evenly across all modules. 

By moving the heat to the radiator you greatly increase the surface area.  Same amount of heat but now it can be cooled easier with lower cfm fans and with less hotspots and air routing issues.

You're missing the point.  They more than doubled the space inside the cabinet (read double the rackspace/moar monyz to host) and &still& arrived at a botched design.  I'm not saying it will not hash, simply that it is less efficient, bulkier, needlessly complex & due to higher internal temps, more failure-prone.  There's nothing to debate here -- the cooling system, considering its expense & complexity, is a total joke.  If a kid designed it, i would try to help, but this was done by adults, PROFESSIONALS FFS!

Edit:  As far as "their supplier had it" -- i suspect Seasonic gave them a killer deal on the 750W jobs, and they took it.  Works for the BJ, 2 for this one, and a huge quantity deal.  Nothing to do with finer points of design.

cedivad is refusing to listen to reason these days.

He's in a dissociative fugue state, unable to think rationally, so great is his pique and so massive his sense of entitlement.

Let cedivad bawl like a spoiled little girl told she's not getting a pony for Christmas.

Let him waste money on lawyers and time on premature FTC complaints.

Have a laugh at his expense as the brat sets his hair on fire and runs around screaming bloody murder.

When cedivad sobers up, he'll be embarrassed by his pathetic behavior.  

As he should be.  Mature adults don't melt down and go postal at the first hint of bad news or slight disappointment.

First, let's dispose of the obvious stuff -- the only discussion that goes on about passing air through the radiator is weather the fan is pushing or pulling it.  Stuffing hot air inside the case is the solution of the last resort.  It's graceless.  It's an admission that you have not planned your layout, or that you have run into such trouble that you wound up defeating whatever advantage water cooling has offered.  It's a noob mistake, and even noobs don't make it.  It's akin to venting your cAr's radiator through the passenger compartment & out of the back windows -- sure, it works, but Concours d'Elegance is suddenly out of your reach, and your GF is pissed.

there's lots of reasons why a 4U was more appropriate in this example than the 2U.  

the two obvious ones are, cooling... and power supplies.

1.  cooling.  a 4U allowed them to use three independent radiators - with 12cm fans.  had they stuck with 2U, they wouldve had a smaller radiator, and smaller fans... which wouldve been much less cooling capacity (and noisier... have you ever heard 6cm fans!?)

2.  power.  theyve gone for twin seasonic atx psus.  these are excellent power supplies... and presumably, well priced... but bulky.  In 2U, achieving the power required, and hitting the pricepoint, wouldve been very difficult.

for those two reasons alone, the switch from 2U to 4U makes sense.

Could it have fitted in a 3U instead of a 4U...?   Maybe... but i bet their suppliers already had a 4U raring to go, which wouldve saved time and, as we all know all too well... bitcoin time, is money!

i dont think its a bad setup...  bringing the cool air in through the rads will be good for the 3 hot chips...   blowing it over the dc/dc converters isnt ideal, but we're not talking about hot air here.. we're talking warm air.

if ambient is say 25, even after the air has been warmed up through the rads, its probably around 40 tops...  i doubt that would do any harm to the other components.   and its being exhausted out the back from the two rear fans...

considering the compromises of having to use two atx psus, which are bulky...  which forced the other design considerations (piggy backing one asic board on top of another) and switching the airflow around from the more obvious direction...really isnt so bad.

even corsair, which incidentally uses the same coolit liquid cooling system as these hashfast boxes...  sometimes recommends to draw cold air into the radiator and exhaust it from the case fans... exactly the same as hashfast is doing, so i really dont see any harm in it.  most liquid cooled pc's operate that way.     in fact i recognise those radiators.. they look remarkably like H60s.   Also, the hot chips run very very hot and need the most cooling, whereas the dc/dc converters arent going to run as hot and wont need so much cooling.

http://www.corsair.com/en/media/catalog/product/cache/1/image/9df78eab33525d08d6e5fb8d27136e95/h/6/h60_new_viewc.png

people have discussed air direction through the rads since the dawn of liquid cooling and either direction has their pros and cons, but it is certainly not uncommon to draw cold air through the rads, blow warmer air over the rest of the system, and then exhaust via the system/case fans.

eg: http://forum.corsair.com/v3/showthread.php?t=98980

You are arguing for the sake of arguing.  Hideous, wasteful design is hideous and wasteful.  If they could have put 3 chips into 2U, puting the same number into 4U (more than double the space) is unqualified defeat.  Fail.  Either that or the 2U mockup was just hype.

Further, going by your wattage, we certainly don't need twin PS.  Seasonic (if that's the fetishistic choice) makes a fine 1250W model
If you're planning to clock up your chips, OTOH, i'm afraid you'll have to cool the VRMs, also.
Just stop.  This doesn't make any sense,it's as elegant design=wise as a pig on ice.

If Hashfast wanted to throw the BabyJet batch 1 customers a bone, they should allow them to "upgrade" their October BabyJets to November Sierras for the difference in price ($7080-$5600=$1440). Their marginal cost on the upgrade (4U vs BJ chassis, extra PSU, two extra closed loop coolers and modules) will be less than the $1440 they would charge.

Yes you see that larged finned component which is taller than the processor waterblock.  The purpose of the fins is to increase the surface area relative to a flat component.  Given the height of the fins (43mm) and the dimensions of the component (~100mm by 20mm) we are talking a pretty large surface area for something which has a staggering heat load of 12.5W.   Yup less than a CFL light bulb.


That is right PER CHIP the nominal wattage is 250W.  The DC converters will therefore produce a heat load of a staggering 25W.  Actually on second thought there are two regulators per board so it is more like 12.5W.   12.5W per regulator with large finned heat sink producing a large surface area.   A low airflow is sufficient to keep temps <50C over ambient.

Relatively large surface area?  And if 10% is 25W, my crusty calculator tells me that 100% is 250W, is that what the total draw of those three chips is?  RLY?  Then Y the dueling Seasonics?  Or do you mean moar liek 150W?

Edit:  Stop defending stupid, ugly designs.  No thought went into this.  Zero.  If a 2U solution was possible, a 4U solution is monstrously wasteful.
Unless their 2U solution was nothing but a bluff.

Lower isn't low.  Your statement was it was would "hot and very low velocity".   

Standard heat capacity of air is 0.569 W C/CFM  Chip thermal output is 250W.  So we are looking at rise in air temp of C = 142/CFM.  At 50 CFM per radiator it is <3C rise in air temp over ambient.   Obvious your concern is that a system which will operate at ambient without issue is going to ignite at 3C over ambient.

The reality is the airflow will be slightly warmer and still sufficiently fast to cool components with very load heat loads.   If chip uses 250W and the DC converter is 90% efficient we are talking 25W on heatload over a relatively large surface area (enhanced by the large heat sink fins on the module).  The rest of the components on the board would be a rounding error in terms of heat output.  It is a complete non-issue.  

Stop it with e delta T crap.  Why do i assume the velocity would be low?  Are you serious?  Do you understand anything about fluid dynamics or airflow?  Think, ffs.  Do i have to explain that airflow would be slower inside a big wide case than it is at the fan?

Rosewill is garbage?  Everything but Seasonic is garbage?  You're finding problems where there are none.

 

Not sure what you are talking about, not having my layout in front of you.  As i mentioned before, you seem to know absolutely nothing about this stuff.


If they were planning to put 3 chips into a 2U case, the same number of chips in a 4U case seems like a lame solution, no?  Lowered expectations much?

Well "hot" is all relative.  Hotter than ambient?  Sure.  Hot enough to make cooling ineffcient unlikely.  If component is rated for 60C (most are 85C) and the intake air is 27C that is a huge delta T.  More than enough for sufficient cooling.  Why do you assume the velocity will be low?  With pusher and puller fans on radiator plus exhaust fans you base this assumption on very low velocity based on what?

You are forgetting that those components, the ones rated for 60C, aren't only conveniently heated by the radiator exhaust -- they generate heat themselves, and need cooling.  Like the VRMs, which cook.  With air cooling, there's enough air spillover to cool them, with this hotbox setup, the air traveling over those VRMs is both hot & very low velocity.  Dumb as dirt.

Well at least this case design has exhaust fans.  KNC didn't see the need... pretty dumb and/or cheap of them.  But not as idiotic as this design.

Single low efficiency, high cost, no name brand piece of garbage PSU vs rock solid nearly perfect reviews SeaSonics.


and disrupt the airflow to the radiators which is the single most important components.  So nice ice cold power supply (which is designed to use case air) and worse airflow to the radiators. 


Because it is 50% more per chassis and rack than 2 modules.  Still if it bothers you buy 2 and split it into two units.  Problem solved.

Then again you have no intention of buying one, in any configuration, at any price you are just here to troll.