[Guide] Dogie's Comprehensive BitCrane T-110 Setup [HD]

[bobsag3]

The unit I have hashing in 24C ambient is at 91 and 93C per chip, previously was 99 and 97 per chip, and when the unit was moved to our warmer facility shutdown due to thermal throttle

I don't think thats thermal throttle - of the CPUs anyway. I put my unit in a 40C cell and the chips were at 90C, and they've told me they have their software cap at 110C. The ASICs internally throttle to keep the temps below that. Either way, you might have some dodgy mounting of your blocks?

Thats what I am starting to think. I will probably tear it apart tomorrow again and re-paste the blocks

[MrTeal]

why the hell are these so overpriced?!

Because ppl want to make money and don´t realize that bitcoin mining is no longer a get-rich-quick game.

These particular units use the probably cheapest 28nm chip available, the one from hashfast. (it is the cheapest one because bankrupt hashfast sells them below internal cost)

The PCBs need a bit more than usual mining pcbs, and the watercooled system adds a little to pricing.

However, internal costs of these miners shouldn´t exceed 0.6$/GH, so there is a massive margin.

I would guarantee that the internal price of these miners is over $0.6/GH, with everything that's included. That's only $660 per unit, and there's no way they're getting two Yoli boards, the coolers, fans, PSU and case along with assembly and test in a sub-$700 budget.

[Collider]

My estimated internal cost would also not enclude NRE such as engineering cost etc.

I was extrapolating from AM/Bitmain figures, and calculating at a high production volume (which might not be applicable here).

The hashfast chip also needs a quite expensive PCB, but is it more expensive per GH than other miners?

At the end of the day the chip is quite potent, but needs expensive peripherals to work properly.

[MrTeal]

My estimated internal cost would also not enclude NRE such as engineering cost etc.

I was extrapolating from AM/Bitmain figures, and calculating at a high production volume (which might not be applicable here).

The hashfast chip also needs a quite expensive PCB, but is it more expensive per GH than other miners?

At the end of the day the chip is quite potent, but needs expensive peripherals to work properly.

Yeah, I was just talking about the per unit cost. I have the board files and full BOM for that board. The PCB is expensive, and the BOM costs are quite high on the Yoli board, even in a reasonable volume. The board cost for the HF chips is higher; because the current density is so much higher you need many layers of heavy copper.

[dogie]

My estimated internal cost would also not enclude NRE such as engineering cost etc.

I was extrapolating from AM/Bitmain figures, and calculating at a high production volume (which might not be applicable here).

The hashfast chip also needs a quite expensive PCB, but is it more expensive per GH than other miners?

At the end of the day the chip is quite potent, but needs expensive peripherals to work properly.

Yeah, I was just talking about the per unit cost. I have the board files and full BOM for that board. The PCB is expensive, and the BOM costs are quite high on the Yoli board, even in a reasonable volume. The board cost for the HF chips is higher; because the current density is so much higher you need many layers of heavy copper.

But at the same time, you only need 2 PCBs per TH and its a small area. Doesn't it work itself out?

[MrTeal]

My estimated internal cost would also not enclude NRE such as engineering cost etc.

I was extrapolating from AM/Bitmain figures, and calculating at a high production volume (which might not be applicable here).

The hashfast chip also needs a quite expensive PCB, but is it more expensive per GH than other miners?

At the end of the day the chip is quite potent, but needs expensive peripherals to work properly.

Yeah, I was just talking about the per unit cost. I have the board files and full BOM for that board. The PCB is expensive, and the BOM costs are quite high on the Yoli board, even in a reasonable volume. The board cost for the HF chips is higher; because the current density is so much higher you need many layers of heavy copper.

But at the same time, you only need 2 PCBs per TH and its a small area. Doesn't it work itself out?

Short answer, no. Putting aside the issue of mounting a cooler, say for instance that you had four individual HF dies on their own 784 ball (29mm square) BGA instead of the current four die 1936 ball (484 pins per die, 45mm square). Even spreading the dies out, it would only add marginally to the area of the board; the power supply area and none of the logic would even have to change. The flip side of that is now instead of essentially having all the current entering and exiting the die through two sides of the die (since the other two sides have a die adjacent to them with their own power planes) you now have power being able to enter/exit through all four sides. In addition, using the larger package gives you 67% more power pads, giving a corresponding reduction in per-ball current. Those two factors combine to not only to improve efficiency since you lose less in the board, but you can reduce your number of layers and/or the copper weight per layer.

[Collider]

So basically the high density doesn´t pay off as you need to use expensive water-cooling loops and the PCBs also get more costly per GH.

I guess the design approach spondoolies follows is much cheaper per GH, and even large BE200 boards should be less expensive per GH (and easier / cheaper to cool).


Possible splitting of the dies into 4 chips might help achieve a higher efficiency, and cooling per die might also improve (or atleast stay the same) and should allow for an air-cooled setup.

With some solid engineering, I am sure other companies should also be able to achieve a high air-cooled density, eliminating the need for these costly water-cooled setups.

[ravin]

Great review once again dogie. Added the miner to the hardware comparison list
https://docs.google.com/spreadsheets/d/1Gtj8ts3zEUZPwSBZQnYZwi__zGfg-0SKVIdH_17GqYw

With some solid engineering, I am sure other companies should also be able to achieve a high air-cooled density, eliminating the need for these costly water-cooled setups.

Very true for AM, Bitmine chips. Hashfast has designed everything to be watercooled I guess. May be a good way to reduce the airconditioning but use some clever hotair ducts :-)

[-ck]

I don't supposed you've seen source code for the (cgminer GPL3 based) driver for this hardware?

[PatMan]

Excellent guide again dogie - well done  

As a follow up to ck's post, and as has been discussed recently here:  https://bitcointalk.org/index.php?topic=738936.0  I'd like to formerly request that you include a new marking section to your guides, based on the manufacturers compliance with cgminers GPLv3 License. I believe a manufacturer who uses ck's code for their products but refuses to release the source should be marked down in some way. Open Source software should & must be kept Open Source, and any manufacturer who does not release the code, even after ck requests them to do so, should be frowned upon as to the reasons why, for obvious reasons.


I'm not sure how you could/should implement this, but I do believe it will aid the reader of your guides into making a more informed and morally correct decision as to which manufacturer to invest their BTC into.
Thanks for your time - and keep up the excellent work!

Peace  

Edit: This, of course, goes for any Open Sourse GPLv3 software - not just ck's.

[Rabinovitch]

I'd like to formerly request that you include a new marking section to your guides, based on the manufacturers compliance with cgminers GPLv3 License. I believe a manufacturer who uses ck's code for their products but refuses to release the source should be marked down in some way. Open Source software should & must be kept Open Source, and any manufacturer who does not release the code, even after ck requests them to do so, should be frowned upon as to the reasons why, for obvious reasons.

Excellent idea. 

[padrino]

why the hell are these so overpriced?!

Because ppl want to make money and don´t realize that bitcoin mining is no longer a get-rich-quick game.

These particular units use the probably cheapest 28nm chip available, the one from hashfast. (it is the cheapest one because bankrupt hashfast sells them below internal cost)

The PCBs need a bit more than usual mining pcbs, and the watercooled system adds a little to pricing.

However, internal costs of these miners shouldn´t exceed 0.6$/GH, so there is a massive margin.

This may be dated but I can tell you that one of the reasons Hashfast didn't get out of their hole is they were trying to sell things well above market price, I and a few other people had to scrap Hashfast projects using hundreds/thousands of chips/boards because until three weeks ago they were selling at very high prices... I wonder if they committed to Hashfast and almost lost it until things finally started being adjusted..

[padrino]

I am surprised they just don't speak up on the cgminer issue, I for one would be surprised if they made any changes to cgminer to support this hardware as it's just Hashfast Yoli boards.. If they are talking to it via socket they are in compliance...

With that said if they did make any changes they need to step up and do the proper (and legal) thing and release, although I can't imagine what they would have done since nothing was needed..

[dogie]

I don't supposed you've seen source code for the (cgminer GPL3 based) driver for this hardware?

The reason why I left it to you for the list is I couldn't tell you if I'm looking at Cgminer source code or Cgminer API - software is not my thing I'll get them to sort this if its not done already.

[PatMan]

I am surprised they just don't speak up on the cgminer issue.....

Me too actually, they've been online quite a few times since the first post so they're aware of the concerns. But then, they've been ignoring ck's request since day one, so .........I just bumped it again 

[dogie]

I am surprised they just don't speak up on the cgminer issue.....

Me too actually, they've been online quite a few times since the first post so they're aware of the concerns. But then, they've been ignoring ck's request since day one, so .........I just bumped it again 

I've emailed them now, they'll come sort it.

[PatMan]

Nice one dogie.

Have you had any thoughts about adding a marking system for the Closed/Open Source compliance yet?

Peace  

[dogie]

Nice one dogie.

Have you had any thoughts about adding a marking system for the Closed/Open Source compliance yet?

Peace   

Yeah it will be added, just need a list from CK on who he doesn't have source from yet.

[ManeBjorn]

Hi,
I recently interviewed the CEO of BitCrane and he said it is available on GitHub.  I am going to make sure he doesn't just mean that cgminer is available but their port of it.
I have started asking this of companies that I interview now.  I feel there needs to be more awareness of it as you do.
The interview will go live later this week or next week once I get a few more questions answered and verify info.
Thanks again for you hard work.

I don't supposed you've seen source code for the (cgminer GPL3 based) driver for this hardware?

[dogie]

Hi,
I recently interviewed the CEO of BitCrane and he said it is available on GitHub.  I am going to make sure he doesn't just mean that cgminer is available but their port of it.
I have started asking this of companies that I interview now.  I feel there needs to be more awareness of it as you do.
The interview will go live later this week or next week once I get a few more questions answered and verify info.
Thanks again for you hard work.

I don't supposed you've seen source code for the (cgminer GPL3 based) driver for this hardware?

They've not forked anything, they're using the original HashFast code and so don't need to repost.