HashFast announces specs for new ASIC: 400GH/s - page 547.

mrb

legendary

Activity: 1512

Merit: 1028

Quote from: DeathAndTaxes on August 21, 2013, 12:22:56 AM

Quote from: erk on August 21, 2013, 12:10:02 AM

Nonsense, you still need to get the heat out of the water, that requires energy spent on fans and pumps, water is only more efficient if the pumps + fans use less power than air cooling fans.

With a large radiator you can use larger, slower, lower wattage fans...

...which is a problem created in the first place by the choice of concentrating 250W of heat on a single socket, requiring high airflow over a relatively small space.

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.

mrb

legendary

Activity: 1512

Merit: 1028

Quote from: DeathAndTaxes on August 20, 2013, 10:14:27 PM

That would mean more complex asembly, more components, and higher cost. Not sure how that is going to improve ROI%.

By the way, this is wrong. You need to consider the system as a whole when comparing water cooling vs air cooling.

You need at least N square centimeters of surface area to transfer 250W of heat to the air no matter the cooling solution: whether this surface area is the fins of a heatsink in an air cooling setup, or the radiator helping water exchange its heat with air in a water cooling setup. (And as you pointed out, water cooling in general uses even more surface area to allow using slower and larger fans.) Now what do you think require less material and parts:

(1) have the N square centimeters of metal sitting right on top of a chip, or
(2) have the chip transfer the heat through a water block, to water, through a pump, through a radiator.

Obviously (1), that is air cooling, requires less material and parts, and less assembly overall: the same N square centimeters of surface area is required (or less), but there is no water, no pump, no water block.

iCEBREAKER

legendary

Activity: 2156

Merit: 1072

Crypto is the separation of Power and State.

Quote from: DeathAndTaxes on August 20, 2013, 09:49:58 PM

Quote from: Vbs on August 20, 2013, 08:34:00 PM

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.

DeathAndTaxes

donator

Activity: 1218

Merit: 1080

Gerald Davis

Quote from: erk on August 21, 2013, 12:10:02 AM

Nonsense, you still need to get the heat out of the water, that requires energy spent on fans and pumps, water is only more efficient if the pumps + fans use less power than air cooling fans.

With a large radiator you can use larger, slower, lower wattage fans. Heat transfer to air is directly related to surface area. A radiator is going to have much more surface area and thus need less airflow. With a large enough radiator you could get by with no fans. Another thing to consider is that the power consumption of an ASIC (any ASIC) rises with temperature. Water cooling tends to have a lower delta T (difference in chip temp and ambient air temp).

erk

hero member

Activity: 826

Merit: 500

Quote from: DeathAndTaxes on August 20, 2013, 11:56:06 PM

Who buys watercooling gear from newegg? Sounds like someone who has never watercooled anything.

Each of KNC chips is rated at 250W. So either 250W can't be cooled with air (in which case KNC is screwed) or HF could just as easily cool their 250W chip with air. So why choose water. Simple water has 16x the thermal conductivity of air. You can cool a large wattage device more effectively and quieter using water.

By "commodity" I assumed they meant "off the shelf". Since it has the package size roughly that of a CPU it will work with any waterblock designed for a CPU. Case, fans, radiator, power supply it is all commodity gear. The only parts which are unique to HF are the ASIC boards and controller board. Now compare that to BFL cooling and power solution. Think there is even a single vendor selling an air or water cooler for a BFL single?

My hope is in their next batch they allow DIY kits. Sell me three ASIC boards and a controller board and let me build my own.

Nonsense, you still need to get the heat out of the water, that requires energy spent on fans and pumps, water is only more efficient if the pumps + fans use less power than air cooling fans.

mrb

legendary

Activity: 1512

Merit: 1028

Quote from: DeathAndTaxes on August 20, 2013, 11:56:06 PM

By "commodity" I assumed they meant "off the shelf". Since it has the package size roughly that of a CPU it will work with any waterblock designed for a CPU. Case, fans, radiator, power supply it is all commodity gear. The only parts which are unique to HF are the ASIC boards and controller board. Now compare that to BFL cooling and power solution. Think there is even a single vendor selling an air or water cooler for a BFL single?

Of course, there are plenty of vendors! If you opened up a BFL Single SC, you would find it has 2 standard 120mm fans + 2 standard CPU heatsink/fan assemblies. These are real commodity parts, available in much larger quantities than any water block.

DeathAndTaxes

donator

Activity: 1218

Merit: 1080

Gerald Davis

Who buys watercooling gear from newegg? Sounds like someone who has never watercooled anything.

Each of KNC chips is rated at 250W. So either 250W can't be cooled with air (in which case KNC is screwed) or HF could just as easily cool their 250W chip with air. So why choose water. Simple water has 16x the thermal conductivity of air. You can cool a large wattage device more effectively and quieter using water.

By "commodity" I assumed they meant "off the shelf". Since it has the package size roughly that of a CPU it will work with any waterblock designed for a CPU. Case, fans, radiator, power supply it is all commodity gear. The only parts which are unique to HF are the ASIC boards and controller board. Now compare that to BFL cooling and power solution. Think there is even a single vendor selling an air or water cooler for a BFL single?

My hope is in their next batch they allow DIY kits. Sell me three ASIC boards and a controller board and let me build my own (with flourinert).

mrb

legendary

Activity: 1512

Merit: 1028

Quote from: ?? on ??

Not sure if you've seen this yet but check out this link: https://hashfast.com/design-of-the-rig/

Thanks for the link.

They explain they wanted to rely on commodity computer processor cooling parts. "Commodity" is air-cooling and heatsinks designed for CPUs up to 100-150W, at most. A heatsink or water cooling block designed to dissipate 250W of heat is hardly "commodity"...

Heck, check Newegg: there are 27 products in the "water cooling system" category vs 300 products in the "processor heatsink/fan" category. And Newegg restricts the purchase quantity of most water cooling systems to 5 per customer. Found a model you like, and want to buy units to keep as spares for your Hashfast farm? Can't even buy more than 5 from Newegg! Have to resort to sourcing from multiple resellers, shipping hassles, etc... I get the feeling that HashFast is out of touch with the reality of what is a "commodity" part in computer hardware.

Ytterbium

full member

Activity: 238

Merit: 100

Quote from: ?? on ??

Quote from: mrb on August 20, 2013, 11:00:21 PM

Yes the energy density is the same. But my point remains: it is technically easier to cool 4 x 63W chips than 1 x 250W chip. So I am waiting for Simon's reply...

4 x 63W chips means four fans and four heat sinks versus 1 water cooler and heat sink for the 1 x 250W chip. Some of AMD's server chips run at 80W with mobos having 4 sockets and they all require fans and heatsinks.

Right, and they require cheaper fans and heatsinks as well.

Multicore chips need a huge amount of bandwidth between the cores. Each core can actually read eachother's cache at full cache access speed. They need to do that to implement thread memory locks - as soon as one chip changes a spot in memory, all the other chips need to be able to see it. (Although this is different in some multicore cellphone chips to save energy)

bitcoin miners barely need any bandwidth at all. If you split up the chip you could use water cooling on all of them and potentially pump more power through them and get more hashrate for the same amount of silicon. You would have the same w/mm^2 of silicon, but each one could have more total heatsink area.

jspielberg

sr. member

Activity: 490

Merit: 255

Quote from: Ytterbium on August 20, 2013, 11:03:36 PM

How old are you people, 3? What an embarrassing exchange.

I assume they were just thread crapping... the ignore list just gets a bit longer.

Ytterbium

full member

Activity: 238

Merit: 100

Quote from: klee on August 20, 2013, 06:21:31 PM

Quote from: ASIC-K on August 20, 2013, 06:20:35 PM

Quote from: klee on August 20, 2013, 06:19:32 PM

Quote from: ASIC-K on August 20, 2013, 06:13:58 PM

oh they dont exist, but im saying dont start comparing shit when its not out yet. asshole.

Another biased shithead here! surprise surprise
Fuck ur mother anytime

im not biased you dick head. i have orders with knc, xcrowd and bfl (still waiting)

can we just stop while we are ahead.

Biased AGAINST HF!! what a retard

Quote from: ASIC-K on August 20, 2013, 06:22:21 PM

im not biased! i wanted to order a baby jet but didnt have the BTC. ugh. ok im quitting anyway.

How old are you people, 3? What an embarrassing exchange.

mrb

legendary

Activity: 1512

Merit: 1028

Quote from: DeathAndTaxes on August 20, 2013, 10:14:27 PM

Quote from: mrb on August 20, 2013, 10:06:24 PM

Simon, regardless of what you think is doable, you are taking unnecessarily high risks to develop such a high-TDP chip. It would have made a helluva lot more sense to go with, say, 4 chips of 63W each. The die would be 1/4th the size. 1/4th the cost too since you would have 4x more chips per wafer. A little more PCB space would be used (big freaking deal). But this would have taken away a lot of the risk of cooling a single hot chip.

The energy density (W/mm2) would still be the same. 1/4th the wattage over 1/4th the area. They had decided to go water cooling for efficient heat transfer. That would mean more complex asembly, more components, and higher cost. Not sure how that is going to improve ROI%.

There is a reason that Intel puts 4 cores on a single socket chip rather than having consumer grade boards run 4 sockets with single core processors.

Yes the energy density is the same. But my point remains: it is technically easier to cool 4 x 63W chips than 1 x 250W chip. So I am waiting for Simon's reply... If my point wasn't clear enough: an air-cooled 4 x 63W-chip Hashfast Baby Jet could almost assuredly be designed and shipped before a more complex water-cooled 1 x 250W-chip system.

PS #1: the reason Intel doesn't use 4 sockets on consumer grade boards is because sockets are expensive (4 sockets at $10 each would double the price of a consumer grade $40 mobo...) plus space is constrained in a desktop form factor (whereas nobody care really if a 400 Gh/s miner delivered in October is 1U vs 3U).

PS #2: I also disagree with BFL's choice of the PCIe form factor. I am just saying it could be worse if they had decided to host a single 300+ watt chip on the card.

Mobius

hero member

Activity: 988

Merit: 1000

Quote from: Kouye on August 20, 2013, 10:22:43 PM

Quote from: DeathAndTaxes on August 20, 2013, 10:03:37 PM

Quote from: Kouye on August 20, 2013, 10:01:21 PM

Quote from: DeathAndTaxes on August 20, 2013, 09:49:58 PM

If it 90% efficient then to supply 277W @ 12VDC requires 308W @ 120VAC.

Talking about serious PSU, here. The kind that cost more than HF chips. Grin

?

I was not aware that commercial PSU had reached a 90%+ efficiency on a ~100% load.
Seems to be acheived, and not that expensive, so I'll just shut up.

http://www.techpowerup.com/reviews/?category=Power+Supplies&manufacturer=&editorschoice=1&recommended=1&pp=25&order=score

DeathAndTaxes

donator

Activity: 1218

Merit: 1080

Gerald Davis

Yeah PSU have gotten much better in the last decade. Finally. The "80Plus" program has actually done some good. It sucked that at one time ATX PSU had a sweet spot of maybe 80% efficiency in a narrow band of 50% to 60% of peak power. No reason that it took this long to finally get decent PSU. Thank the rising wattage requirements and electrical rates.

Kouye

sr. member

Activity: 336

Merit: 250

Cuddling, censored, unicorn-shaped troll.

Quote from: DeathAndTaxes on August 20, 2013, 10:03:37 PM

Quote from: Kouye on August 20, 2013, 10:01:21 PM

Quote from: DeathAndTaxes on August 20, 2013, 09:49:58 PM

If it 90% efficient then to supply 277W @ 12VDC requires 308W @ 120VAC.

Talking about serious PSU, here. The kind that cost more than HF chips. Grin

?

I was not aware that commercial PSU had reached a 90%+ efficiency on a ~100% load.
Seems to be acheived, and not that expensive, so I'll just shut up.

DeathAndTaxes

donator

Activity: 1218

Merit: 1080

Gerald Davis

Quote from: mrb on August 20, 2013, 10:06:24 PM

Simon, regardless of what you think is doable, you are taking unnecessarily high risks to develop such a high-TDP chip. It would have made a helluva lot more sense to go with, say, 4 chips of 63W each. The die would be 1/4th the size. 1/4th the cost too since you would have 4x more chips per wafer. A little more PCB space would be used (big freaking deal). But this would have taken away a lot of the risk of cooling a single hot chip.

The energy density (W/mm2) would still be the same. 1/4th the wattage over 1/4th the area. They had decided to go water cooling for efficient heat transfer. That would mean more complex asembly, more components, and higher cost. Not sure how that is going to improve ROI%.

There is a reason that Intel puts 4 cores on a single socket chip rather than having consumer grade boards run 4 sockets with single core processors.

Quote

I see this as a very poor strategic decision made by Hashfast. (And Cointerra, and KnCMiner). Only BFL seems that they will get it right, since their 350W Monarch card will split the workload across 10-30 chips.

BFL get it right? The PCIE card form factor is very tough. BFL will need to cool 350W using a single blower fan in a very compact space. This is a task that even AMD got wrong ... twice (6990 & 7990) resulting in 6 month delays on those cards. If you said Avalon, AsicMiner, or BitFury (easily cooled boards with multiple chips over large surface area) you might have had a point. I don't think "right" is the proper word (it is a design choice) but you may have had a point. BFL? You showed your true colors there.

zedicus

legendary

Activity: 966

Merit: 1004

CryptoTalk.Org - Get Paid for every Post!

Hey simon how about just putting some guidelines like "this chip wont overclock past 540" and a mechanism to monitor or log chip speed and shut it down if it get past a certain threshold, maybe something besides a thermal sensor..

then you dont have to offer a 10 second warranty and you can claim you have hardware that you support and stand behind.

Vbs

hero member

Activity: 504

Merit: 500

Quote from: Simon Barber on August 20, 2013, 09:37:05 PM

350W is a rounded up number for the whole system, including the power lost in the 2 stages of power supply, and fans etc. The chip itself draws 250W @ nominal.

Depends on how hard you cool and overclock your Sandy Bridge E. The company that is assembling our systems specializes in overclocking. They run Sandy Bridge Es overclocked to 350W (CPU power alone, not whole system), using the same cooling system we are using. We are also using a heatspreader.

Metal migration is a well understood phenomenon. We have followed all the fab's rulesets for electromigration so that the current levels we're going to see will not be a problem (even current distribution, and thicker metal layers). Currently in the simulator for EM our chip passes the test for a 5 year lifetime, but fails the 11 year test - and that is running somewhat overclocked, at about 540GH/s.

Thanks for the info. What's the size of the package/heatspreader and what are you using between it and the chip? Solder? TIM? W/mK?

Quote from: DeathAndTaxes on August 20, 2013, 09:49:58 PM

Quote from: Vbs on August 20, 2013, 08:34:00 PM

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 excluding water cooling there chips doesn't run on 12V. A good 12VDC to ~1VDC PSU (on the board) is at best 90% efficient.

So if the chip uses 250W @ ~1VDC then it will require at least ~277W @ 12VDC. Now the ATX PSU which converts the 120V AC isn't 100% efficient either. If it 90% efficient then to supply 277W @ 12VDC requires 308W @ 120VAC.

Add in 20W for pump and radiator fans plus a margin fro safety and ~350W is more than reasonable.

Once again this isn't unique to just HF, it applies to all ASICs by all companies.

Board Wattage = Chip Wattage / DC PSU efficiency
System Wattage = ASIC Board Wattage + Controller Wattage + Fan Wattage + Auxiliary Wattage
Wall Wattage = System Wattage / ATX PSU efficiency

The wattage at wall is NEVER going to be the wattage at the chip level.

Yes, you are quite correct, two power stages at η around 0.9 are enough to have 19% in power losses, plus the rest of the cooling extras.

Still, the specs say "Under 350 watt power draw*" and "* Real silicon power consumption may vary from simulation results by +/- 20%", so we're talking apples and oranges here.

mrb

legendary

Activity: 1512

Merit: 1028

Quote from: Simon Barber on August 20, 2013, 07:49:11 PM

Quote from: Vbs on August 20, 2013, 07:31:47 PM

Interesting so, KnC is a 250/900 = 0.28 W/mm^2 chip and HashFast is a 350/324 = 1.08 W/mm^2 chip, requiring a cooling solution that can transfer 5.6x the heat per mm^2. I really hope your cooling solution holds up 24/7!

Out of comparison, an ATI 7970 is a 250/365 = 0.68 W/mm^2 chip and an nVidia GTX Titan is 250/561 = 0.45 W/mm^2 chip.

Hashfast's chip at nominal (400GH/s) is expected to consume about 250W of power, not 350W. i.e. 0.77W/mm^2. Overclocked Sandy Bridge E in shipping commercial products with the same cooling system runs 350W with a 425mm^2 die, 0.82W/mm^2.

Simon, regardless of what you think is doable, you are taking unnecessarily high risks to develop such a high-TDP chip. It would have made a helluva lot more sense to go with, say, 4 chips of 63W each. The die would be 1/4th the size. 1/4th the cost too since you would have 4x more chips per wafer. A little more PCB space would be used (not a big deal). But this would have taken away a lot of the risk of cooling a single hot chip.

Bitcoin mining being an embarrassingly parallel algorithm, there is absolutely no point in trying to made a die as large / as hot as possible, when you can spread computations across chips with virtually no downside.

I see this as a very poor strategic decision made by Hashfast. (And Cointerra, and KnCMiner). Only BFL seems that they will get it right, since their 350W Monarch card should split the workload across 10-30 chips.

If I was an investor in Hashfast, I would have asked you "why do you take the risk of aiming at the highest TDP possible for your gen1 chip?". Please explain your choice.

DeathAndTaxes

donator

Activity: 1218

Merit: 1080

Gerald Davis

Quote from: Kouye on August 20, 2013, 10:01:21 PM

Quote from: DeathAndTaxes on August 20, 2013, 09:49:58 PM

If it 90% efficient then to supply 277W @ 12VDC requires 308W @ 120VAC.

Talking about serious PSU, here. The kind that cost more than HF chips. Grin

?

Topic: HashFast announces specs for new ASIC: 400GH/s - page 547. (Read 880816 times)