HashFast announces specs for new ASIC: 400GH/s

[RoadStress]

I still haven't heard back from Samantha about employment... I'll probably forward my resume to Amy later today and try and figure out what's going on and if they still want me for support.

They must be busy taking picture of PCBs.

[1714788230]

1714788230

[1714788230]

1714788230

[1714788230]

1714788230

[DPoS]

Comment jockey, although I also offered to go down to CIARA to report back on the progress on the boards and assembly there

sounds needed...   btw I confused thinking they were in TX.. I have no clue where they are

[Gyrsur]

Comment jockey, although I also offered to go down to CIARA to report back on the progress on the boards and assembly there

sounds needed...   btw I confused thinking they were in TX.. I have no clue where they are

pic anywhere? sabber. ;-)

[DeathAndTaxes]

You don't want to mix-n-match cables. If you have 2 power supplies feeding to a single board and one power supply dies, the board will try to pull double the load from the other power supply. Bad things happen.

Well HashFast didn't design the system for redundant power supplies (it would require much higher output power supplies = higher cost) but pulling double load from one power supply won't necessary "bad things".  It would cost more but if each power supply had sufficient capacity to handle the full load then the system could operate just fine on one.  When both PSU are operating the load on each one will be half.  If one shuts down the other will handle the full load.  You gain redundancy, and higher efficiency at the expense of more power supply cost.  If one PSU fails the other will run at 100% load until the first PSU is replaced.



  

[DeathAndTaxes]

I think though, if you've got the miner in a DC and it need a hardware power cycle you've got more problems than just reaching into the back of the rack (you've actually got to go to the DC or pay some dude to do it!

Just use a switchable PDU.  They work great for home "farms" as well.

And anyway, DC costs are going to bite into the miserable ROI. Who would even think of doing that unless they had some free space?

Agreed.   

[DeathAndTaxes]

First of all, you were incorrect....  It's not 1400, and also not 1300.  It is, as i have suggested, 780, as could be easily calculated from the product description here.  Each chip is 400 GH/s, there are 3 chips for 1200 GH/s total.  At advertised .65W/GH, that totals to 780W.  Any other number you arrive at directly contradicts the company claims.

Try to learn something for once.   The efficiency of the chip isn't the entire power requirement of the entire system.

The reported efficiency AT NOMINAL HASHRATE is ~0.65 J/GH.  So at 400 GH/s that is 260W @ ~1VDC (actually voltage depends on chip spec and hasn't been reported but likely is 0.7VDC to 1.0VDC).  An ATX power supply delivers high current on the 12V rail, however no AISC runs at 12V and thus the 12VDC has the be converted to ~1VDC and conversion means energy "lost" as heat (otherwise the VRMs would be cold to the touch).  A quality high current 12 VDC to sub 1 VDC regulator is going to be ~90% efficient so that is 288W @ 12VDC per board. 288W @ 12VDC in = 260W @ ~1VDC out + 28W as heat.

Sill 288W isn't the whole story either.  The watercooling loop requires power as well.  How much?  Not sure but lets guestimate.  Each module includes 2 fans likely high RPM so lets go with 12W ea and a pump say another 6W.  So 30W per cooling loop plus 288W per board is 318W total.   Lastly there are a pair of exhaust fans.  Lets guesstimate 12W ea.  318*3 + 12*2 = 978.  So 978 not 780 watts per rig.  

At nominal hashrate (400 GH/s) & 0.65 J/GH.
Per chip -  260W @ 1VDC
Per board - 288W @ 12VDC
Per board including cooling - 318W @ 12VDC
Entire System including exhaust fans - 978W @ 12VDC

However that is only at nominal 400 GH/s hashrate.  Hashfast has indicated they believe the chips can be pushed harder.  How hard?  Well I certainly don't know and HashFast won't even know for sure until they get the final silicon but lets say they can run at 450 GH/s at nominal voltage and with a 10% overvolt can be pushed to 500 GH/s.  In Silicon devices, power consumption increases by the square of the voltage increase.  So 10% higher voltage = 21% higher power or 0.78 J/GH.
 
500 GH/s @ 0.78 J/GH  (AS AN EXAMPLE ONLY)
Per chip -  390W @ ~1VDC
Per board - 433W @ 12VDC
Per board including cooling - 463W @ 12VDC
Entire System including exhaust fans - 1,413W @ 12VDC

So designing around a 780W power supply would limit the potential of the device.   Maybe 500 GH/s per chip isn't possible but maybe 440 GH/s or 465 GH/s is.  HashFast stated there were no power supplies in the range they were looking at (1300W to 1400W DC) from vendors they were interested in.   Also stated was that even if one was available it was more expensive than two smaller units.  

SeaSonic's "X series" for example maxes out at 1250W and a 1250W unit has higher price than a pair of 750W units (1500W total).

[jjiimm_64]

Thanks D&T.   I had to ignore him after a while... 

[DeathAndTaxes]

Thanks D&T.   I had to ignore him after a while... 

I had to as well after that last post I doubt he will listen but at least others will get the correct info.

[Minor Miner]

Thanks D&T.   I had to ignore him after a while... 

I had to as well after that last post I doubt he will listen but at least others will get the correct info.

A lot of people appreciate your well thought out replies and explanations.   It is good to visualize what the back of this 42U is going to look like (and require).

[Syke]

Well HashFast didn't design the system for redundant power supplies (it would require much higher output power supplies = higher cost) but pulling double load from one power supply won't necessary "bad things".  It would cost more but if each power supply had sufficient capacity to handle the full load then the system could operate just fine on one.  When both PSU are operating the load on each one will be half.  If one shuts down the other will handle the full load.  You gain redundancy, and higher efficiency at the expense of more power supply cost.  If one PSU fails the other will run at 100% load until the first PSU is replaced.

Each board pulls about 300 watts over 2 PCIe 6-pin connectors. When one power supply dies, that board will start pulling 300 watts over a single 6-pin connector. How many power supplies cables can safely supply 300 continuous watts over a single 6-pin connector?

[crumbs]

First of all, you were incorrect....  It's not 1400, and also not 1300.  It is, as i have suggested, 780, as could be easily calculated from the product description here.  Each chip is 400 GH/s, there are 3 chips for 1200 GH/s total.  At advertised .65W/GH, that totals to 780W.  Any other number you arrive at directly contradicts the company claims.

Try to learn something for once.   The efficiency of the CHIP isn't the entire power requirement of the entire system.

Each chip uses ~0.65 J/GH.  @ 400 GH/s that is 260W @ ~1VDC.  However no chip runs at 12VDC and thus the 12VDC has the be converted to ~1VDC.  High current DC to DC regulator is ~90% efficient so that is 288W @ 12VDC per board.  The watercooling system requires power as well.  How much?  Not sure but lets guestimate.  Each includes 2 fans likely high RPM so lets go with 12W ea and a pump say another 6W.  So 30W per cooling loop so 318W for board & cooling.   954W per rig.  

At nominal hashrate (400 GH/s) & 0.65 J/GH.
Per chip -  260W @ 1VDC
Per board - 288W @ 12VDC
Per board including cooling - 318W @ 12VDC
Entire System - 954W

However that is only at nominal 400 GH/s hashrate.  Hashfast has indicated they believe the chips can be pushed harder.  How hard?  Well we won't know until they get final silicon but lets say with a small voltage increase 10% they can push the chips to 520 GH/s.  In Silicon devices, power consumption increases by the square of the voltage increase.  So 10% higher voltage = 21% higher power.
 
AS AN EXAMPLE ONLY  520 GH/s @ 0.78 J/GH
Per chip -  408W @ ~1VDC
Per board - 454W @ 12VDC
Per board including cooling - 484W @ 12VDC
Entire System - 1363W @ 12VDC

So designing around a 780W power supply would limit the potential of the device.   HashFast has indicated there were no power supplies from vendors they liked which had a high efficiency power supply in the 1300W to 1400W range and that even if one was available the price tends to be higher than two smaller units.  SeaSonic for example maxes out at 1250W and a 1250W unit is more than 2x 750W ones.

Yo are the only one suggesting designing around a 780W PS.  Go back and read my post again, and try to learn something for once.  I guestimated 900W, and pointed out that a 1000W power supply, available from Sea Sonic, would suffice for the stock specs.  For extra credit, try to ell me where i went wrong.

If you're a modder & wish to overclock, go out and buy (again) a Sea Sonic 1250, or any of the quality 1600W PS.  Elementary. You get even MOAR watts.  

As an example, running the 3 chips @767GH/s would require 6000W & a liquid nitrogen cooling system.  I'm surprised they have omitted the hardware allowing for *that* eventuality.  Cutting corners, i suppose

Just stop.  The make-believe math makes you look ridiculous -- there are no chips.  The numbers you're quoting could be off by a factor of 10, if Hashfast's case design is any indicator

Edit:  Case design -- the renderings, that is, is the only thing we have to judge Hashfast by.
No chips.
No boards.
No photos of *anything*.
Just some 3=D renders & a stolen Intel CPU pic.

[DeathAndTaxes]

I didnt say it was totally inaccessible, but no one wants to through the back, and wade through the cable mess, move cable management arms and get your fingers cut just to turn a machine on or off, or reset it.

What datacenter reboots servers manually?  



I would recommend using a y-cable to connect one switched PDU power drop to both power supplies.



Login to web interface, click power off, power on.  Tada no need to leave your office chair.  If someone was super crazy they would integrate cgminer with the API in popular switched PDUs to auto power cycle when miner appears to be down.  

Note: photos are for NEMA 5-15 outlets (120V, 15A).  Most likely datacenter is going to run on 208V in the US.  If someone was going to install a dedicated branch circuit at home for miners no reason to not go 240V.  Double the power for the same current, higher efficiency, and lots of used PDU on ebay for cheap.   Still the same concept applies at 120V, 208V, or 240V.

[DeathAndTaxes]

Well HashFast didn't design the system for redundant power supplies (it would require much higher output power supplies = higher cost) but pulling double load from one power supply won't necessary "bad things".  It would cost more but if each power supply had sufficient capacity to handle the full load then the system could operate just fine on one.  When both PSU are operating the load on each one will be half.  If one shuts down the other will handle the full load.  You gain redundancy, and higher efficiency at the expense of more power supply cost.  If one PSU fails the other will run at 100% load until the first PSU is replaced.

Each board pulls about 300 watts over 2 PCIe 6-pin connectors. When one power supply dies, that board will start pulling 300 watts over a single 6-pin connector. How many power supplies cables can safely supply 300 continuous watts over a single 6-pin connector?

Just about all decent brands (and certainly SeaSonic).  The 6 pin = 75W, 8 pin = 150W is a PCIe standard, it doesn't represent the limit of what is possible using the underlying hardware.  The connector itself can handle 288W continually, Molex rated specs for 6 pin Minifit Jr connector is 9A per pin (13A with high current pin).  So 6 or 8 pin PCIe connector is fine for 3 conductors *12 Volts * 9 A ea = 288W  and real world is probably significantly more.  Don't do this at home but I stress tested a connector at 420W for over an hour and it was only warm to the touch.  The wiring itself (3ft, 16 gauge, 12V, 20A per conductor) can handle a lot more so it isn't the bottleneck. 

The only difference between a 6 pin and 8 pin connector is the 8 pin has a pair of ground sense pins to compliant devices (i.e. GPUs) to know which cable is connected on power up.   If you look carefully at many power supply cables you will see there are TWO 8 pin connectors in series on the same cable which plugs into the PSU.  Even under normal usage that means 150W + 150W = 300W on the wires.

[bobsag3]

I didnt say it was totally inaccessible, but no one wants to through the back, and wade through the cable mess, move cable management arms and get your fingers cut just to turn a machine on or off, or reset it.

What datacenter reboots servers manually?  



I would recommend using a y-cable to connect one switched PDU power drop to both power supplies.



Login to web interface, click power off, power on.  Tada no need to leave your office chair.  If someone was super crazy they would integrate cgminer with the API in popular switched PDUs to auto power cycle when miner appears to be down.  

Note: photos are for NEMA 5-15 outlets (120V, 15A).  Most likely datacenter is going to run on 208V in the US.  If someone was going to install a dedicated branch circuit at home for miners no reason to not go 240V.  Double the power for the same current, higher efficiency, and lots of used PDU on ebay for cheap.   Still the same concept applies at 120V, 208V, or 240V.

Yeah I made the mistake of having my first 80A pulled as 120v. Nope Nope Nope 240v from here on out.

[Erin HashFast]

Good Morning!

I saw yesterday someone had mentioned they were having a hard time getting though. We are here, however, the amount of incoming phone calls/emails has tripled in the past 30 days!  If you call in and get voicemail, PLEASE leave a message with your name, phone number and order number(s). We are calling/emailing everyone back in the order calls/emails have been received.

Thank you for your patience,
Erin & Cara

[joshv06]

Good Morning!

I saw yesterday someone had mentioned they were having a hard time getting though. We are here, however, the amount of incoming phone calls/emails has tripled in the past 30 days!  If you call in and get voicemail, PLEASE leave a message with your name, phone number and order number(s). We are calling/emailing everyone back in the order calls/emails have been received.

Thank you for your patience,
Erin & Cara

Hey Cara, Josh here. Any way we can get an update here on the forum of how things are going and if we're still on schedule for ~next week?

Thanks!

[Erin HashFast]

Good Morning!

I saw yesterday someone had mentioned they were having a hard time getting though. We are here, however, the amount of incoming phone calls/emails has tripled in the past 30 days!  If you call in and get voicemail, PLEASE leave a message with your name, phone number and order number(s). We are calling/emailing everyone back in the order calls/emails have been received.

Thank you for your patience,
Erin & Cara

Hey Cara, Josh here. Any way we can get an update here on the forum of how things are going and if we're still on schedule for ~next week?

Thanks!

Yes, actually. We anticipate making an official announcement today.

Regards,
Erin

[Minor Miner]

Good Morning!

I saw yesterday someone had mentioned they were having a hard time getting though. We are here, however, the amount of incoming phone calls/emails has tripled in the past 30 days!  If you call in and get voicemail, PLEASE leave a message with your name, phone number and order number(s). We are calling/emailing everyone back in the order calls/emails have been received.

Thank you for your patience,
Erin & Cara

Hey Cara, Josh here. Any way we can get an update here on the forum of how things are going and if we're still on schedule for ~next week?

Thanks!

Yes, actually. We anticipate making an official announcement today.

Regards,
Erin

Great, you can save me calling back by answering the following questions in your official announcement:
1.  When does it look like novemeber sierras will start and finish shipping?
2.  Have your chips arrived and been tested and are within modeled specs?

thanks

[crumbs]

...
Yes, actually. We anticipate making an official announcement today.

Regards,
Erin

Erin, what does that mean?  You don't know if you'll be making an announcement, but you anticipate that you will?

Regards,
crumbs

[cedivad]

Yes, actually. We anticipate making an official announcement today.

Let's see if it's a vapourware type announcement or what.