Reviewing the Halong Mining DragonMint T1, a 10nm Bitcoin Miner

[oawan]

This isn't 10nm is it?  I thought it was 16nm

Look, if you assume a 10nm chip designed as well as the S9 chips, you should be getting about 20TH @ 1400w. 

If version rolling ASIC boost adds another 20%, you should be seeing ~24th at that power. 

My speculation is that this is a 16nm clone of the S9 but with the ASIC boost counteracting a worse chip design to bring a *very* slight increase in efficiency.

As much as people hate Bitmain, they do seem to have the best chip designs.

[1680026540]

1680026540

[HagssFIN]

Yoshi Goto (MyRig) has mentioned that it is 10nm chip and that the foundry is Samsung.

[dimaze]

There is a huge inductor on the PCB so it looks like
there is a DC/DC converter on board. Can you read
the chip number of the chips that are directly under
this inductor on the other side of the PCB?

I'll tear it down again soon and add a few more detailed pictures If anyone wants anymore details, let me know here! I'll update the OP with some additional details later.

How many years of warranty does Halong provide for these units?

6 months from the point of shipment - more information on the terms of sale page here

Yoshi did confirm it was Samsung and 10nm! More details this tweet

Halong is also the only miner manufacturer I've seen take this many payments options. In addition to the standard Wires/BTC options, they take BCH, LTC, ETH, XMR, DCR, USDT! Great to see a manufacturer accept the currency they sell miners on. A good thing!

[NotFuzzyWarm]

This isn't 10nm is it?  I thought it was 16nm

Look, if you assume a 10nm chip designed as well as the S9 chips, you should be getting about 20TH @ 1400w.  
If version rolling ASIC boost adds another 20%, you should be seeing ~24th at that power.

My speculation is that this is a 16nm clone of the S9 but with the ASIC boost counteracting a worse chip design to bring a *very* slight increase in efficiency.
As much as people hate Bitmain, they do seem to have the best chip designs.

Based on what theory?
They are Samsung 10nm chips.

These ever smaller nodes are geared for low-power applications -- NOT high power ones like mining. The overall power involved raise die temps far beyond what the Foundries processes are designed to allow for. As a result the only way to make the miners stable is to sandbag the operating parameters and that means a lot of *possible* efficiency gains go right out the window. Props for mining chip makers trying to push the envelope but as usual the Real World pushes back - hard.

The Ebang 10nm miner is another example of designs gone wrong...

As for theoretical gains AB may give -- so far not as-advertised...

[2112]

If anyone wants anymore details, let me know here! I'll update the OP with some additional details later.

Measure the DC voltage (if any) between the individual heat-sinks; while operating, but open, with additional external desktop fan providing additional cooling.

[buckrogers]

There currently are no firmware updates published on the website, however, Halong's Telegram (https://t.me/dragonmint)
published one for the T1 here:
https://download.halongmining.com/files/t1/t1_20180409_010732.swu

Hi,

Does the dragonmint T1 new firmware offer any improvements in hashing speed, or was it to update a different issue?

Thanks!

[philipma1957]

There currently are no firmware updates published on the website, however, Halong's Telegram (https://t.me/dragonmint)
published one for the T1 here:
https://download.halongmining.com/files/t1/t1_20180409_010732.swu

Hi,

Does the dragonmint T1 new firmware offer any improvements in hashing speed, or was it to update a different issue?

Thanks!

It  gives better graphics  

maybe a little better power draw.
maybe a little better hash.


say  1650 watts at k-watt meter
say  15540 gh


vs  
1679 watts
15300 gh


this particular unit  is the one I tested  on old firmware with 3 psu's

I need to  check  power again

i am now at

1640 watts
15540 hash

0.1055 watts a gh

 

[dimaze]

Updated the OP with a close up of the Zynq chip there - second to last picture there on the teardown section. Let me know if any of you want any other closeups!

Measure the DC voltage (if any) between the individual heat-sinks; while operating, but open, with additional external desktop fan providing additional cooling.

I'll try to accommodate this. Unit wasn't friendly with me when I had the fans disconnected or even connected, will try a workaround.

Hi,

Does the dragonmint T1 new firmware offer any improvements in hashing speed, or was it to update a different issue?

Good question! I actually lost hashrate with the new firmware. It went from slightly above to below spec with the update. My recommendation at this point is if you have a unit that is hashing around spec, don't touch it with any firmware updates just yet. Halong indicated they're working on firmware updates to alleviate this issue. The GUI had welcome additions, such as individual chain ASIC status charts and a firmware checker, along with autotune capabilities. I'll throw those up here in the OP shortly.

[2112]

Measure the DC voltage (if any) between the individual heat-sinks; while operating, but open, with additional external desktop fan providing additional cooling.

I'll try to accommodate this. Unit wasn't friendly with me when I had the fans disconnected or even connected, will try a workaround.

Thanks in advance. If you can't do this with the unit open then just measure the voltages on those heatsinks that can be reached through the case openings with the case closed and just the fans temporarily moved aside.

[intron]

Measure the DC voltage (if any) between the individual heat-sinks; while operating, but open, with additional external desktop fan providing additional cooling.

I'll try to accommodate this. Unit wasn't friendly with me when I had the fans disconnected or even connected, will try a workaround.

You can probe just a few random heat sinks while
all boards are normally mounted in the frame. If it
is anything else but 0V, we can run some calculations.

[dimaze]

Didn't get any DC readings off the first row of heatsinks. Will try to play around with it more after I play with the new beta firmware!

https://download.halongmining.com/testing/t1_20180415_075807.swu

Had some nice high numbers on the wind up minute of hashing!

https://i.imgur.com/iIXHg37.jpg

Pointed my T1 to ck's pool and it's currently running tuning on performance mode. Follow the T1's progress here:

http://ckpool.org/users/bc1qw0el3xg4zjuukjtmq9d8u9wr6yn57ts896k83u


Moderator's note: This post was edited by frodocooper to remove inline image tags.

[philipma1957]

http://ckpool.org/users/1JdC6Xg3ajT3rge3FgPNSYYFpmf53Vbtje

I set mine to balanced   and I am waiting for results.


Moderator's note: This post was edited by frodocooper to remove an unnecessary quote.

[dimaze]

http://ckpool.org/users/1JdC6Xg3ajT3rge3FgPNSYYFpmf53Vbtje
I set mine to balanced   and I am waiting for results.

Good call! We'll be able to compare after a bit. I'm off to lunch now, will check how it's been when I am back, including what the power draw is. GUI 50 minutes in shows it averaging around 16-16.25 at least, poolside 1h appears to be catching up to that figure. Promising results so far.

[philipma1957]

to make it clear I named the worker over


http://ckpool.org/users/1JdC6Xg3ajT3rge3FgPNSYYFpmf53Vbtje


so give it an hour to settle in for hash rate

power on k-watt is

1646 watts



"workername": "1JdC6Xg3ajT3rge3FgPNSYYFpmf53Vbtje.firm415",
   "hashrate1m": "9.43T",
   "hashrate5m": "4.05T",
   "hashrate1hr": "416G",
   "hashrate1d": "17.7G",
   "hashrate7d": "2.53G",
   "lastshare": 1523808733,
   "shares": 359619,
   "bestshare": 861686.0,
   "lns": 359619.1,
   "luck": 1.01,
   "herp": 361876.5063849979
  }
 ]
}

[dimaze]

After about 3.5 hours of uptime on performance mode, the pool hashrate is 16.4-16.5T at the hourly indicator. Looks to be fluctuating between those numbers. GUI shows similar readings. Looks like my chain #1 is the performer, chain 2 has 22x the HW errors, and chain 3 has 11x. Had the other two chains performed like chain 1, it would be closer to 17TH!

Amp pull is between 6.8-6.9, that's about 1632-1656 watts.

https://i.imgur.com/3nZwyia.jpg
https://i.imgur.com/t69OAgA.jpg
https://i.imgur.com/IalK0cJ.jpg

With my preliminary testing, I highly recommend people upgrade to the firmware I've updated in the OP, I'll be putting these screenshots up there too.

I'll continue to test the other settings. Excited with this new firmware!


Moderator's note: This post was edited by frodocooper to remove inline image tags.

[philipma1957]

I have had some trouble  getting it working.

Note to self stop  using Apple safari

Works fine with Apple - Chrome- Yosemite.

I am tuning low power efficient  setting with this miner

http://ckpool.org/users/1JdC6Xg3ajT3rge3FgPNSYYFpmf53Vbtje

looks like power dropped at least 200 watts  to 1400

still waiting on true hash rates  but 1400 watts x 4 = 5600 watts which means it will be able to allow 4 units on a 30 amp pdu.

It would be nice to see it do 14250gh at 1400 watts

https://i.imgur.com/iDwS7Fr.png

https://i.imgur.com/UYiNYlf.png

https://i.imgur.com/1GMfVCI.png

https://i.imgur.com/RUWl5xN.png


Moderator's note: This post was edited by frodocooper to remove inline image tags.

[dimaze]

I've always been a chrome man myself

I just shifted my unit to efficiency too. It's autotuning, and I'll check back in the evening with results. Same pool target as my above test for anyone that wanted to track it live.

[xdrpx]

Never looked at the reviews of the s7 or s9 despite them being several years old now??
The t1 has same layout and lazy/brain-dead heatsink design that makes for them being LOUD due to all of the turbulence created by so many leading/trailing edges on that mass of individual sinks.

ref the Canaan Avalon repair guide for example of correct thermal design.

That's interesting, since you've mentioned about the S9. I went ahead and found HagssFIN's post on the guide / teardown but the PCB there wasn't visible since I guess the user wanted to retain the warranty seals. I went ahead and look it up on youtube and was amazed to see that it looks exactly similar to the Halong Mining DragonMint T1, including the outer casing.

Note: I tried googling the Canaan Avalong repair guide since the hyperlink you've provided is incomplete. I came up with https://bitcointalk.org/index.php?topic=3191702.0 and the technical setup page over here - https://canaan.io/faqs/avalonminer-721/page/2/ (for the Avalon 741 and 721). What's interesting was as compared to the Halong Mining dragonmint T1, the Avalon 841 has one continuous heatsink without many tiny ones attached to the PCB. Also, the 16nm ASIC chips right beneath them https://imgur.com/cYXplGP are clearly visible. Looks like an interesting design difference between the two.

[philipma1957]

Yeah  safari  is dying  fuck apple.

I have to stop using it and move over about 100 passwords to chrome.


This software  will be nice if I settle in with easy  watts for pdu's  1400 watts  or lower works really well


@xdrpx

The  T1 board uses the exact same heatsinks as the s-9.

The board is a bit different in layout.



the efficient setting  lowers fans to 60%  in an 85 f garage

noise becomes okay.

still tuning,but  this firmware looks like a winner for me on low hash setting.


It rocks at  efficiency setting

13.85 -13.90
1325 watts

you tube link

https://www.youtube.com/watch?v=K_ksMluPUZI&t

[dimaze]

Efficient mode settings I had fluctuating between 14.2-14.3TH, and drawing 5.3-5.4 amps -- 1272-1296 watts over the course of over a day.

That's between 88.95W/TH and 91.27W/TH, with a 90.11W/TH median at least on this boot and the beta firmware provided.

https://i.imgur.com/XNN6RJO.jpg
https://i.imgur.com/0IXCFvL.jpg

I'm pretty eager to see what type of specs this pushes out as the firmware is fine tuned.