[GAUGING INTEREST] H-Card Mini Group fab

[Taugeran]

H-CARD Mini:
    The h-card mini is the OSHW offspring of a wonderful forum member here, Kaerf. to summarize their efforts: they had received a few regular v1.2 H-cards with several dead chips. instead of mailing them back for RMA, they used the available schematics from MBP/BFSB for the v1.2, which normally holds 16 Bitfury ASICs, and redesigned it to hold 8 ASICS and accommodate mounting holes for standard 40mm² northbridge heatsinks.

Images courtesy of Kaerf

Album
http://imgur.com/a/5Ebiw

Blank board


Standard 40mm² heatsinks



What am I proposing?

I am gauging interest to see how many folks out there in the community with unfilled m-boards who would like to fill them with something now that MBP/BFSB are both out of stock on these cards.

Who am I?
I am a frequenter of the forums and partaker of several past group buys from distributors such as kosmokramer, Bitcoinvalet, Xian, and the well known designer of the Nanofury devices, vs3.

Ive been on the purchasing end of group buys before but this is my first venture into leading one.

[Taugeran]

Privately I have been gearing up for a small-ish run of 10 boards with

• pcbs fabbed by seeedstudio's fusion pcb service incl: free e-test and ENIG gold finish on connector for durability.
• Polyimide (Kapton) Stencil from OharaRP
• Parts lists filled with the most cost effective split across Digikey and Mouser

so i figured i would extend the opportunity to the public.

Depending on the number of interested parties, i may continue this route. Or if there are enough interested I will come up with a more formal, professional fab sequence utilizing an assembly house.


UPDATE: pricing breakdown of the bare boards and BOM will be listed in various order quantities..shortly

[Taugeran]

Reserved: for pricing breakdown and additional notes.





All interested please leave a comment with Qty desired. all comments are non binding and may be retracted at any time for any reason*


FOR ALL Parties interested in the present standing of the Bill of Materials and Estimates, here is a link to a publicly commentable spreadsheet which includes some scratchpad math used in arrival at prices.

https://docs.google.com/spreadsheets/d/1oAkDHg9hR94NZVTgnxD9AuBUls8syuhw3IGAKbLQkPc/edit#gid=42212007







*until a formal list is compiled, and BOM lists are ordered.

[Swimmer63]

I still run 2 rigs with open spots. I would be interested in trying 2 out, depending on the price.
How many GH/s per board?

[Taugeran]

I still run 2 rigs with open spots. I would be interested in trying 2 out, depending on the price.
How many GH/s per board?

well going off figures obtained from projects like the NFY: using the same 30A vrm to power half the # of chips would allow 3.75A per chip. and we know how amp hungry these guys are

the NFY using Rev.2 chips i got to acheive 2.8-3.0G with 3A@0.866V so, without having one immediately to test. i would say in the neighborhood of 22-25Ghash would be a safe bet. the ones kaerf made used the 20A version of the VRM and those produced ~20ghash. who knows we may be pleasantly surprised and get 25+ in which case half the chips, same hashrate as the regular 16 chip cards


see https://bitcointalk.org/index.php?topic=228677.msg3259455#msg3259455 for his results

[klondike_bar]

The better method would be to make 16-chip boards by the official spec, and using the newer revision of chip that should allow each 16-chip board to push ~50GH.

The 8-chip board just results in more slots on the m-board being filled which is an issue if you don't sell m-boards to go with it.  just my 2c

[Taugeran]

The better method would be to make 16-chip boards by the official spec, and using the newer revision of chip that should allow each 16-chip board to push ~50GH.

The 8-chip board just results in more slots on the m-board being filled which is an issue if you don't sell m-boards to go with it.  just my 2c

Hash rate is very dependent on amperage an voltage and with positive correlations to both.

A rev.2 16 chip board wouldn't reach 50 easily with out some serious heat extraction and over clocking

It would be no problem at all if another 30A power phase was added to the board. But then you have 70+ watts to pull out of a board

That's some of the rationale I used when deciding 8 vs 16

I'm wanting to get the max out of these chips and and doubling the amp ration they each get is one way to do so.

If I knew that 16 could crunch along at 50 It would be a no brainer decision. But in most real life situations a rev2 16 ASIC card with heatsinking on the vrm and each ASIC
OV'ed to 0.833 would probably pull ~35-40 maybe 43


TL;DR: a 16 chip card won't run at the same per chip hashrate as an 8 chip card using the same 30A  power supply.
This is because of thermal and electrical constraints

[allinvain]

I would be interested in a few. I just had two h-cards die on one of my rigs, and it looks like another card is hashing ridiculously poorly on a different rig so it's probably going to kick the bucket soon. I just hope that they will hash roughly equivalent to the 16 chip cards. I'd be happy with 25 GH. The thing is, they'd have to be pretty cheap to justify reinvesting in bitfury hardware rather than just saying screw it and go buy a bitmain or A1 based product.

I've come to the conclusion that v1.2 cards are just too poorly built to run 24/7.

[Taugeran]

I would be interested in a few. I just had two h-cards die on one of my rigs, and it looks like another card is hashing ridiculously poorly on a different rig so it's probably going to kick the bucket soon. I just hope that they will hash roughly equivalent to the 16 chip cards. I'd be happy with 25 GH. The thing is, they'd have to be pretty cheap to justify reinvesting in bitfury hardware rather than just saying screw it and go buy a bitmain or A1 based product.

I've come to the conclusion that v1.2 cards are just too poorly built to run 24/7.

If I might inquire how were the chips on the two dead cards doing before communication death?

[Taugeran]

BOM is underway. check 3rd OP for details

[zulover]

Sent pm

[klondike_bar]

The better method would be to make 16-chip boards by the official spec, and using the newer revision of chip that should allow each 16-chip board to push ~50GH.

The 8-chip board just results in more slots on the m-board being filled which is an issue if you don't sell m-boards to go with it.  just my 2c

Hash rate is very dependent on amperage an voltage and with positive correlations to both.

A rev.2 16 chip board wouldn't reach 50 easily with out some serious heat extraction and over clocking

It would be no problem at all if another 30A power phase was added to the board. But then you have 70+ watts to pull out of a board

That's some of the rationale I used when deciding 8 vs 16

I'm wanting to get the max out of these chips and and doubling the amp ration they each get is one way to do so.

If I knew that 16 could crunch along at 50 It would be a no brainer decision. But in most real life situations a rev2 16 ASIC card with heatsinking on the vrm and each ASIC
OV'ed to 0.833 would probably pull ~35-40 maybe 43


TL;DR: a 16 chip card won't run at the same per chip hashrate as an 8 chip card using the same 30A  power supply.
This is because of thermal and electrical constraints

I was talking about this rev2 chip: http://www.bitfurystrikesback.com/product/bitfury-55nm-rev2-samples/

864 cores instead of 756 (14% more) and presumably the same or slightly lower power usage.  All my 1.2 h-boards can be overclocked to ~32GH without heatsinks and up to 38GH with heatsinks. 14% more means you could get 36-45GH per card if overclocked.  (which at current pricing and considering a 3week production, will only be able to sell for  ~$75 each or possibly less)

[daddyfatsax]

I have 6 open slots on my v1.0 board. Would like to fill out, merely for OCD purposes. My v3 board will only take 12 cards, so hopefully the v1 board will take all 6.

[Taugeran]

The better method would be to make 16-chip boards by the official spec, and using the newer revision of chip that should allow each 16-chip board to push ~50GH.

The 8-chip board just results in more slots on the m-board being filled which is an issue if you don't sell m-boards to go with it.  just my 2c

Hash rate is very dependent on amperage an voltage and with positive correlations to both.

A rev.2 16 chip board wouldn't reach 50 easily with out some serious heat extraction and over clocking

It would be no problem at all if another 30A power phase was added to the board. But then you have 70+ watts to pull out of a board

That's some of the rationale I used when deciding 8 vs 16

I'm wanting to get the max out of these chips and and doubling the amp ration they each get is one way to do so.

If I knew that 16 could crunch along at 50 It would be a no brainer decision. But in most real life situations a rev2 16 ASIC card with heatsinking on the vrm and each ASIC
OV'ed to 0.833 would probably pull ~35-40 maybe 43


TL;DR: a 16 chip card won't run at the same per chip hashrate as an 8 chip card using the same 30A  power supply.
This is because of thermal and electrical constraints

I was talking about this rev2 chip: http://www.bitfurystrikesback.com/product/bitfury-55nm-rev2-samples/

864 cores instead of 756 (14% more) and presumably the same or slightly lower power usage.  All my 1.2 h-boards can be overclocked to ~32GH without heatsinks and up to 38GH with heatsinks. 14% more means you could get 36-45GH per card if overclocked.  (which at current pricing and considering a 3week production, will only be able to sell for  ~$75 each or possibly less)

Ah I gotcha now

[Taugeran]

After some back and forth last night with the very enlightening vs3 has given me some insigh and ideas on how the BOM can be lowered further. Though this will require maybe 1-2 evenings reworking capacitor pads and quantities.

Summary:

Since these cards are based off one of the first production designs which itself was heavily over-engineered particularly the decoupling capacitor arrays around each chip and the groups of 8 through out Vcore power plane

[mikerbiker6]

I don't quite understand why 16 chips won't reach 50GH/s. How is this not scalable?
If you double the powersupply and use active cooler too get rid of the heat you could reach 50GH/s right?

[allinvain]

I would be interested in a few. I just had two h-cards die on one of my rigs, and it looks like another card is hashing ridiculously poorly on a different rig so it's probably going to kick the bucket soon. I just hope that they will hash roughly equivalent to the 16 chip cards. I'd be happy with 25 GH. The thing is, they'd have to be pretty cheap to justify reinvesting in bitfury hardware rather than just saying screw it and go buy a bitmain or A1 based product.

I've come to the conclusion that v1.2 cards are just too poorly built to run 24/7.

If I might inquire how were the chips on the two dead cards doing before communication death?

To be honest I'm not sure as I was not monitoring them actively before. But I do remember them hashing at around 25 GH or so. Another card died just the same. I ended up giving the card away to some dude in the UK. He found out that "16 Capacitor mising and 1or 2 Resistor blowup." I hope you use robust resistors and caps.

[Taugeran]

I don't quite understand why 16 chips won't reach 50GH/s. How is this not scalable?
If you double the powersupply and use active cooler too get rid of the heat you could reach 50GH/s right?

yes doubling the power supply to 60A would allow 16 chips to achieve 50 Gh. But TI doesn't have a pin compatible regulator capable of 60A to replace the TPS53355

[klondike_bar]

I don't quite understand why 16 chips won't reach 50GH/s. How is this not scalable?
If you double the powersupply and use active cooler too get rid of the heat you could reach 50GH/s right?

My experience with The rev1 chips on v1.2 H-boards was that with a good board you could get up to 25-30GH/s stock.  
With a small pencil mod AND either decent airflow OR heatsinks 30-34GH/s was achievable.
With a moderate pencil mod AND decent airflow AND some heatsinks on the backside 35-38GH was achievable.
With all the above plus a bit of luck, or some very dedicated tuning, 38-42GH was achievable.  At this stage the VRM is about 150% its maximum amperage specs but could handle it.

The rev2 chips are about 15% more powerful at the same wattage presumably.


I recently had a 38GH board (3rd slot) stop hashing and fail to show up in .stat.log.   my assumption is that either the TPS53355 burnt out or the inductor burnt out. Haven't made much headway on the diagnostic process

[BitcoinValet]

I can get you a quote for pcb and assembly from a US (South FL) facility. We already quoted someone from my first group buy for these, but he just needed a few and the price breaks are at 100-500-1000 units.

I also have 600-800 R2 chips and 100 R1 chips if anyone will be be needing them for this. They will likely be the last bitfury chips that I get, and there are some NF6 group buys that may be needing them, first come first served. You can still buy them from bitfurystrikes back but they have higher prices for smaller quantities, and you are limited to 32 chips for small orders of less than a tray.