Technobit HEX8A1 260GH/s Coincraft based miner reviewed

[2GOOD]

In the past months I presented you 3 boards from Technobit and today I'm happy to give you my opinion about the latest build from the Bulgarian Company - HEX8A1.

The miner in focus today is based on Bitmine Coincraft A1 28nm chip delivering more than 30GH. So far this is the first 28nm chip available for DIY projects and third party designs, and as it seems the board from Tehnobit is the first in production and shipping.



There are several differences in the design from the past generation of HEX boars, the first thing you will notice is the size of the cooler - it's huge compared to older versions. The board itself is equipped with 8 Coincraft chips, squeezed between two heatsinks and two 92mm fans. Also the molex connectors are replaced with two 6-pin PCI-Express connectors. All of these changes are due to more powerful chips requiring better power modules and cooling.


HEX16B (BitFury) and the new HEX8A1 boards side by side

Performance:
At the Technobit's site, HEX8A1 is advertised as 260GH miner in turbo mode, so let's see how it handles various settings.



Indeed with moderate overclock the board is hashing with 260GH/s, but the first thing you will notice when this beast is powered on is the noise, we'll get to that later. Let's focus on performance and power consumption. According to Bitmine's pre-production specifications, the chip can deliver 40GH in Turbo mode with power usage of 1W/GH as you can see from chart bellow that is not the case.



The power supply used for the test is FSP600 50ARN 88+, it is rated to "reach up to 88% efficiency" that's what the second power column is representing Wall power consumption minus 12%, the efficiency of the board is calculated with these numbers. I tried an CoolerMaster B600 (85%) for comparison but it draws around 32W more power at 260/1000.

At lower settings 230/860 the efficiency is 1.09 W/GH and if you take out the cooling it gets around 1 W/GH for the board itself, of course if we go at lower settings let's say 200/860 we'll achieve even greater efficiency but IMO it's not worth it at least not now, maybe in a year time...

About the voltage I tried 850 and lower settings but it wont's hash or unstable. It is also important to mention that the real chip freq is multiplied by 4 so when you say 250 it's really 1000Mhz core chip frequency.



Each chip has 32 Engines, you can have 1 or 2 faulty per chip but thats not a problem. If you try to "undervolt" more keep an eye on the Engines if there is not enough juice you will get lower engines which is a good indicator, although with lower voltage and one or two chips with 28 engines I get some strange (better) results in efficiency perspective. it's all about finding the right ratio for your device - freq/voltage. I recommend 960/1000 it's very stable and you will get results from 255 to 265 GH/s depending on your chips.

Cooling:
This thing is LOUD !!! No seriously, it's pretty much a vacuum cleaner. Despite the huge sink the board needs powerful fans to cool down the 8 chips in a small area. According to Technobit the chips need to be close for the chain to work properly so no way around it.

 

The smaller fan is known from the latest versions of HEX16A2 and HEX16B:
Sunon PF92251V1-000U-A99
Dimensions:    92x92x25 mm
Power: 4.7W
Speed: 4500 rpm
Air Flow: 75 CFM

The big fan is:
Sunon PMD1209PMB1-A.(2).GN
Dimensions:    92x92x38 mm
Power: 12.5W
Speed: 4900 rpm
Air Flow: 120 CFM

Both fans are very dangerous and there are some reports for injured fingers from the previous boards. I'm happy to see that Technobit have taken precautions with the fan grills. The noise is about 60dB at 15cm and 55dB at 1m i.e. LOUD

At 260/1000 the sink temperature close to the chips is 46C at 26C room temperature, with the help of a fan regulator I lowered the fan a little bit and the temperature raised to 50C but noise was significantly lower.

If you are planning to mine at home I recommend to get the new 4 chip version of the board for half the price. It will be equipped with 2 x 2W Sunons and probably much quieter solution.

Note: According to Coincraft A1 specification about 70% of the power is being dissipated through the ground pad and about 30% through the top of the package.


Pricing:
Note: Updated Price Chart
With the latest GOX FUD the bitcoin price is going down, however I believe that Gox will "fix" their withdraw problem "soon" As always investment in bitcoin equipment is risky but let's see what can you order/buy from various mining manufactures. Technobit are selling HEX8A1 for 999 Euros full price which seems normal to me.



As you can see from the chart the main competitor to HEX8A1 is Bitmain's AntMiner, especially if you live in a country where Customs are not a problem for the rest of us importing an Chinese miner can be a hassle. On the other hand Bitmine are selling fully equipped miner with PSU and everything you need. Even more interesting is the newcomer - The UK based company BitcoinUltra, they have good price and ETA, but no working prototype. I assume they will be using A1 chips too but that's only a guess. The race is on and much more to come from BitFury and ASICMINER with their crazy low price per GH, even BFL can ship something soon - doubt it

Software:
Technobit's HEX line still lags proper mining software for Windows where you can only mine with their special HEXMiner, the driver is in the inf folder. The setup is similar to HEX16A so you can check nemercry's video tutorial for Windows.

As with the previous HEX board I highly recommend that you get one cheap TP-Link TL-MR3020 and flash it with the firmware from Technobit's website. You can also order it directly from them at slightly higher price, but pre-flashed and ready to rock. In the end you are paying a thousand euros for the miner, so 30 more for the router is nothing. Of course if you have a linux host it will best to just compile cgminer with the latest patch from Technobit or download my build for Ubuntu x64 cgminer4_0.2.2

Here is how to compile

Code:

git clone https://github.com/ckolivas/cgminer.git
cd cgminer
wget https://dl.dropboxusercontent.com/u/8082585/cg/2786526d9f1acb7bd903ece3c822c2c9203aa0b7.patch
git checkout 2786526d9f1acb7bd903ece3c822c2c9203aa0b7
patch -p1 < 2786526d9f1acb7bd903ece3c822c2c9203aa0b7.patch
./autogen.sh --enable-hexminer8 
make

I start cgminer with the following line:

Code:

sudo ./cgminer -c ./cgminer.conf --hexminer8-set-diff-to-one 0 --hexminer8-chip-mask 255 --hexminer8-voltage 1000 --hexminer8-options 8:260

where cgminer.conf is in the same folder and with your regular settings.



Pros:

• EU Based
• Efficiency - 28nm
• Availability
• Software voltage control

Cons:

• Noise
• Loud
• No thermal sensor
• as usual no cgminer/bfgminer for Windows

Conclusion:
Yet another good product from Tehnobit, even that the noise is a huge drawback for most of the miners. The efficiency of the board is good, not as good as promised by the manufacturer but still better than most of the miners on the market. In the end the Bulgaria based company once again did a great job with this challenge and what surprises me the most is that Technobit are shipping before Bitmine.ch

Full gallery can be found here: http://imgbox.com/g/eKqOvbI9Uv

best
2GOOD

[Gator-hex]

230/860mV = 226GH 279W 1.086W/GH 920MH/W

Are you sure that's right? What I got was...

230/860mV = 195GH 218W 1.12W/GH 894MH/W

maybe it was 900+mV?

[marto74]

Great review as always.
Thank you

[2GOOD]

230/860mV = 226GH 279W 1.086W/GH 920MH/W

Are you sure that's right? What I got was...

230/860mV = 195GH 218W 1.12W/GH 894MH/W

maybe it was 900+mV?

Interesting,
I'll double check that on different devices and let you know

EDIT:
Device 1 seems OK:


Power consumption:
@Wall: 276W
-12%: 243W
Speed: 227GH/s
=> 1.07W/GH 934MH/W

Device 2 is NOT OK:
1st run 250W @WALL = 205GH no HW
2nd run 267W @WALL = 210GH with HW

It's just not stable at these settings, it runs fine at 260/1000 @wall 428W - 264GH/s -0 HW
*cgminer on tplink

best
2GOOD

[Gator-hex]

Mine run best efficiency at around 220/900 = 215GH/250W, but they were from the pre-production run, maybe the end of Feb production run are better? I don't see any capacitors/polyfuses in your photos so assume that's what you have there? 

Nice review and photos. What camera are you using?

[marto74]

actually the yellow elements are poly fuses Gator, not capacitors.

[Gator-hex]

actually the yellow elements are poly fuses Gator, not capacitors.

Stops overcurrent? I've got a couple that are not attached on one board. Should I worry? Seems to run okay but one board has more H/W errors uses 40W more than the other. I noticed one Hex8A1 didn't come back straight after the power supply faulted. I assume it was these things doing their job then. Fixed that by putting a Sunon 4500 fan on the 5V line. Useful for something besides making noise, helps regulate my PSU!

Anyone who doesn't want to listen to fans as loud as their vacuum cleaner, you can run 3x Arctic F9 (92mm/1800RPM/43CFM) in the low 900s mV

[marto74]

Gator
what is the fan model.
We can test it in order to make a version for not OC-ed boards

[Gator-hex]

Gator
what is the fan model.
We can test it in order to make a version for not OC-ed boards

It's P/N: AFACO-09000-GBA01 the standard F9 3pin model I use. Cut back the yellow and wire in the red/blank.
http://www.arctic.ac/eu_en/products/cooling/case-fan/arctic-f9.html

There's been some good bulk buy deals on them recently and I've been paying about £3/€4/$5 for the standard one.
Also they don't eat fingers. 

[2GOOD]

Mine run best efficiency at around 220/900 = 215GH/250W, but they were from the pre-production run, maybe the end of Feb production run are better? I don't see any capacitors/polyfuses in your photos so assume that's what you have there?

Nice review and photos. What camera are you using?

Actualy "device 1" is from the pre-production with polyfuses and "device 2" is a later version of the board, so IMO it all depends on the chips. The new version seems to be more resistant to heat but thats an early observation.

Have you tried 230/900 !?

The photos are taken (as usual) with Nikon D5100 - minimal color correction applied in post.

Thank you

[Gator-hex]

230/860mV = 226GH 279W 1.086W/GH 920MH/W

Are you sure that's right? What I got was...

230/860mV = 195GH 218W 1.12W/GH 894MH/W

maybe it was 900+mV?

Mine run best efficiency at around 220/900 = 215GH/250W, but they were from the pre-production run, maybe the end of Feb production run are better? I don't see any capacitors/polyfuses in your photos so assume that's what you have there?

Actualy "device 1" is from the pre-production with polyfuses and "device 2" is a later version of the board, so IMO it all depends on the chips. The new version seems to be more resistant to heat but thats an early observation.

Have you tried 230/900 !?

230/900 is not good for me, one machine has 10% errors the other 2% and the hash rate is 5-10GH lower too. Probably needs more mV.

Seems to be a lot of variation in these chips. I see from Bitmine's own 1TH video they are using ones running good ones at 200GH @ 830mV!

http://youtu.be/91MuwGOgRFo?t=44s (need to go full screen and HD720p to see it.)

I did hear a rumor they made a later production run with wider tracks to improve efficiency. I thought thats what you had to get those numbers.  

[aopa38]

2GOOD

Good review.

Can these boards be stacked together? Did you try it?

We have a number of them coming and I'm trying to figure out a layout for them.

Thanks

[Gator-hex]

Can these boards be stacked together? Did you try it?

Not really, you cannot stack them vertically because you have 2 powerful fans back to back trying to suck the air out of each other.

You cannot stack them horizontally because they placed the heatsinks in a cross pattern.

I did try to warn them early on about this.

Best buy some shelving units to stand them on and stagger them to stop the heat from one being sucked into the next.

[2GOOD]

230/860mV = 226GH 279W 1.086W/GH 920MH/W

Are you sure that's right? What I got was...

230/860mV = 195GH 218W 1.12W/GH 894MH/W

maybe it was 900+mV?

Mine run best efficiency at around 220/900 = 215GH/250W, but they were from the pre-production run, maybe the end of Feb production run are better? I don't see any capacitors/polyfuses in your photos so assume that's what you have there?

Actualy "device 1" is from the pre-production with polyfuses and "device 2" is a later version of the board, so IMO it all depends on the chips. The new version seems to be more resistant to heat but thats an early observation.

Have you tried 230/900 !?

230/900 is not good for me, one machine has 10% errors the other 2% and the hash rate is 5-10GH lower too. Probably needs more mV.

Seems to be a lot of variation in these chips. I see from Bitmine's own 1TH video they are using ones running good ones at 200GH @ 830mV!

http://youtu.be/91MuwGOgRFo?t=44s (need to go full screen and HD720p to see it.)

I did hear a rumor they made a later production run with wider tracks to improve efficiency. I thought thats what you had to get those numbers.  

On the video it's SYS_CLK=830Mhz which corresponds to 207.5 MHz on HEX8A1, I assume Bitmine's voltage is fixed since there is no information for it on the video.

And they have only 1 dead engine in the whole set.

[Biffa]

Can these boards be stacked together? Did you try it?

Not really, you cannot stack them vertically because you have 2 powerful fans back to back trying to suck the air out of each other.

You cannot stack them horizontally because they placed the heatsinks in a cross pattern.

I did try to warn them early on about this.

Best buy some shelving units to stand them on and stagger them to stop the heat from one being sucked into the next.

Can't you just undo the heatsink and rotate it?

[marto74]

nooo
as far it is in conteact with rechtange area of the ASIC's

[Biffa]

nooo
as far it is in conteact with rechtange area of the ASIC's

My bad, just looked at the pics a bit closer and see what you mean. Shame. Do you plan on changing the design?

[Gator-hex]

nooo
as far it is in conteact with rechtange area of the ASIC's

My bad, just looked at the pics a bit closer and see what you mean. Shame. Do you plan on changing the design?

They did originally have both heatsinks running in the same direction. That would have allowed for stacking them on their sides and using cable ties in the fan holes to secure them together and putting one large fan on one side an blowing over both heatsinks.



 

[marto74]

nooo
as far it is in conteact with rechtange area of the ASIC's

My bad, just looked at the pics a bit closer and see what you mean. Shame. Do you plan on changing the design?

They did originally have both heatsinks running in the same direction. That would have allowed for stacking them on their sides and using cable ties in the fan holes to secure them together and putting one large fan on one side an blowing over both heatsinks.



 

Yes But cooling of the top side of the chips was not enough
Sorry

[Zelek Uther]

Thanks, very professional review.