Bitfury Tardis (SHA256) 80 TH/s 6300w

[markiz73]

https://bitfury.com/hardware/bitfury-tardis

Max performance - up to 80 TH/s
6U server rack
6.3 kW power consumption per server  
Upgradable design
Easily configurable (5-8 hashboards) for max performance and low power consumption
Settings that support various cost optimization plans

https://bitfury.com/content/downloads/bitfury-b9-tardis-datasheet.pdf

[bsp]

Here are some prices:

Bitfury Tardis miner, 6.3 KW.
The Bitfury Tardis miner is offered with different configuration options:

5 hashboards, 66TH/s, 6,3KW – 95 mJ/GH, $55 per TH/s
6 hashboards, 71TH/s, 6,3KW -  88 mJ/GH, $58 per TH/s
7 hashboards, 75TH/s, 6,3KW -  84 mJ/GH, $61 per TH/s
8 hashboards, 78TH/s, 6,3KW -  80 mJ/GH, $65 per TH/s
Each hashboard contains 128 Bitfury Clarke chips.

[HagssFIN]

I think it should make sense for them to only offer the most power efficient version.

[fanatic26_]

I think it should make sense for them to only offer the most power efficient version.

All they are doing is downclocking the boards and adding more. Any of those versions should be able to run at the best power efficiency at the expense of some hash rate.

[HagssFIN]

All they are doing is downclocking the boards and adding more. Any of those versions should be able to run at the best power efficiency at the expense of some hash rate.

Yep, I agree they do different frequency and voltage settings with those models.

[markiz73]

S9i-14.5 TH/s with PSU   $440.00
power usage 1365W

https://shop.bitmain.com/product/detail?pid=000201809271201391568Ipb917H0635

6300/1365= 4.615
4.615*14.5=66,92TX

For some reason it seems to me that the asics from Bitmain will cost less.

[fanatic26_]

S9i-14.5 TH/s with PSU   $440.00
power usage 1365W

https://shop.bitmain.com/product/detail?pid=000201809271201391568Ipb917H0635

6300/1365= 4.615
4.615*14.5=66,92TX

For some reason it seems to me that the asics from Bitmain will cost less.

Of course they will...because they are last generation tech being sold at fire sale prices. They are much less power efficient as well. I dont think I quite understand your weird ass math either.

[HagssFIN]

Yeah that math does not make any sense..

[bsp]

I think the reasoning behind his math:

Bitfuty Power / Antminer Power = Bitfury is drawing 4.615 times more power.
And if same Antminer @ 14.5 TH/s would be 4.615 times bigger it would hash at 14.5 x 4.615 = 66.92 TH/s

He should complete that calculations saying that to get to that level it would cost only $440 x 4.615 = $2030.60 in Antminers
And the cheapest Bitfury at same power draw would be $55 x 66TH/s = $3630.00
which is (2030.60 / 3630) x 100 = 56% higher in $$$

[HerbPean]

I think the reasoning behind his math:

Bitfuty Power / Antminer Power = Bitfury is drawing 4.615 times more power.
And if same Antminer @ 14.5 TH/s would be 4.615 times bigger it would hash at 14.5 x 4.615 = 66.92 TH/s

He should complete that calculations saying that to get to that level it would cost only $440 x 4.615 = $2030.60 in Antminers
And the cheapest Bitfury at same power draw would be $55 x 66TH/s = $3630.00
which is (2030.60 / 3630) x 100 = 56% higher in $$$

/Highfive !

[Steamtyme]

All they are doing is downclocking the boards and adding more. Any of those versions should be able to run at the best power efficiency at the expense of some hash rate.

Looking at it I assumed it was set up with a static fan speed of 100% or something. I couldn't figure out why there was no change in power usage.

So are you figuring that you would be running a different firmware version dependent on how many boards you attached? And that that would be how they downclock the board as they ad more.

That makes me wonder then if they would have a version available that would allow you to run on performance not efficiency (like the 5 board model) for say a colder climate.

[sidehack]

Efficiency changes with voltage, not clock (though heat can be a factor). To change efficiency there has to be some regulator adjustment, or changing the number of nodes in a fixed-volt string.

[vincl]

i guess (as always) the pricing wont be worth it...

[agente]

8 hashboards clocked or volted like 5 hashboards system = 105th/s. 9800w. Interesting

[Steamtyme]

*snip* To change efficiency there has to be some regulator adjustment, or changing the number of nodes in a fixed-volt string.

Forgive my ignorance.

When your talking about the nodes is that in the circuitry, or would that refer to the number of hashboards connected?

[markiz73]

I think the reasoning behind his math:

Bitfuty Power / Antminer Power = Bitfury is drawing 4.615 times more power.
And if same Antminer @ 14.5 TH/s would be 4.615 times bigger it would hash at 14.5 x 4.615 = 66.92 TH/s

He should complete that calculations saying that to get to that level it would cost only $440 x 4.615 = $2030.60 in Antminers
And the cheapest Bitfury at same power draw would be $55 x 66TH/s = $3630.00
which is (2030.60 / 3630) x 100 = 56% higher in $$$

Thanks , you helped me out.
For mining, there is a huge amount of asics, the course of Bitcoin falls and in May 2020 we are waiting for halving the reward for the block.
But I don’t understand why so many ASICs?

[sidehack]

No, I mean if they're operating on a fixed serial voltage (12V, 24V or something) with N levels of chips dividing that voltage. S9 run 63 chips per board in 21 series nodes of three parallel chips, and about 9.5V per board (this varies based on model and tuning) for 0.45V per triplet (9.5/21).

If your voltage is fixed (which seems likely; Bitfury tends to do this), in order to change your voltage, you need to change how many series nodes of ASICs are dividing that voltage. If they're running the new Clarke ASIC, the datasheet calls for something like 0.30-0.35V which means, for a 12V bus, between 34 and 40 chips in series.

Hashrate increases with operating frequency because each calculation will have a specific number of math operations, and as the clock speed increases, the logic doing that math will perform each operation faster. But in order for the chip's transistors to switch faster, they need to push the electrical charges asserting those switches in less time, which requires more energy, which means a higher voltage - analogous to air pressure in a pneumatic system. Increasing voltage also increases, approximately linearly, the amount of charge transferred, which means an increase in current (which is just charge displacement per unit time - literally an electron "flow rate"). The main power losses in a chip happen during switching, and resistive power loss is proportional to both current and voltage.

Being able to operate faster requires increasing voltage, which increases the current, so power loss increases by the square of voltage. Efficiency charts for these chips have an exponential curve to them, where as the chip gets faster, the power required to eke out a bit more speed keeps getting bigger and bigger.

My first thought was the boards were somehow bussed in series, but going from 5 to 8 boards but that makes almost a 40% change in core voltage, which would push them from a severe overclock to a severe underclock (0.3-0.35 is a 15% change). Not really sure what they're doing, but in order to get the flexibility they're claiming it's possible they do have some series regulator on each board, or are trimming the output of the 48V PSU.

[Steamtyme]

Thanks for the explanation that clears it up for me. I've always been curious about the inner workings of miners, this one just seemed peculiar with the more boards/better efficiency.

Hopefully, someone picks one up and provides some updates/photos.