[ANN] Bitfury ASIC sales in EU and Europe

[Omen1855]

Wonderful, looking forward to higher orders. But can't complain this is fantastic.

[punin]

NEW REV2 CHIPS HAVE ARRIVED!!

NOTE: THIS IS AN ENGINEERING BATCH. PLEASE DON'T HOARD. 32 CHIPS MAX PER CUSTOMER.
Please don't send PM requests for samples. Samples are available through the web shop.

Chips are pin compatible with the previous version. Expect better performance and lower power consumption. We are currently testing the chip ourselves, so exact performance is still unknown. Please share your experiences and measurements in this thread!!!

...

+ 1

Nice ... this is interesting .. will there be reels for public sale ... ? Later ?

As otherwise I don't see the point if there is no support in chips for DIY.

I think we'll be working mostly with trays as they're much easier to work with (and test) in the quantities usually needed by DIYers. One full box of trays is 2600 chips.

[RoadStress]

Will someone translate what 864 rolled hash cores means? What's the speed and what's the difference between rolled and un-rolled cores?

[KNK]

I think we'll be working mostly with trays as they're much easier to work with (and test) in the quantities usually needed by DIYers. One full box of trays is 2600 chips.

'Mostly' means that there still will be reels I guess, but are the boxes vacuum sealed or it's each tray separately or neither and chips baking will be necessary?

[KNK]

Will someone translate what 864 rolled hash cores means? What's the speed and what's the difference between rolled and un-rolled cores?

An unrolled core would calculate one hash at each clock, so running at 1MHz one core is 1MH/s
A rolled core calculates the hash in several clock cycles - for Bitfury chips this is 65 cycles, so single core running at 1Mhz will produce close to 15.4 kH/s
If the chip runs at 200MHz - 864 cores are equal to 200*864/65 = 2658.46 MH/s or 2.66 GH/s
For comparison the first revision had 756 cores which at 200MHz is 2.33 GH/s so 15% increase is expected

[Taugeran]

Will someone translate what 864 rolled hash cores means? What's the speed and what's the difference between rolled and un-rolled cores?

An unrolled core would calculate one hash at each clock, so running at 1MHz one core is 1MH/s
A rolled core calculates the hash in several clock cycles - for Bitfury chips this is 65 cycles, so single core running at 1Mhz will produce close to 15.4 kH/s
If the chip runs at 200MHz - 864 cores are equal to 200*864/65 = 2658.46 MH/s or 2.66 GH/s
For comparison the first revision had 756 cores which at 200MHz is 2.33 GH/s so 15% increase is expected

OMG I've never been so happy to be wrong in all my life

[Taugeran]

Will someone translate what 864 rolled hash cores means? What's the speed and what's the difference between rolled and un-rolled cores?

An unrolled core would calculate one hash at each clock, so running at 1MHz one core is 1MH/s
A rolled core calculates the hash in several clock cycles - for Bitfury chips this is 65 cycles, so single core running at 1Mhz will produce close to 15.4 kH/s
If the chip runs at 200MHz - 864 cores are equal to 200*864/65 = 2658.46 MH/s or 2.66 GH/s
For comparison the first revision had 756 cores which at 200MHz is 2.33 GH/s so 15% increase is expected

OMG I've never been so happy to be wrong in all my life

Except when I though my wife wasn't pregnant.


------------------


Is there a rough formula to transform osc6 bits into Mhz?

[RoadStress]

Will someone translate what 864 rolled hash cores means? What's the speed and what's the difference between rolled and un-rolled cores?

An unrolled core would calculate one hash at each clock, so running at 1MHz one core is 1MH/s
A rolled core calculates the hash in several clock cycles - for Bitfury chips this is 65 cycles, so single core running at 1Mhz will produce close to 15.4 kH/s
If the chip runs at 200MHz - 864 cores are equal to 200*864/65 = 2658.46 MH/s or 2.66 GH/s
For comparison the first revision had 756 cores which at 200MHz is 2.33 GH/s so 15% increase is expected

Thank you very much! Can't wait for the populated DYI cheap boards!

[KNK]

OMG I've never been so happy to be wrong in all my life

Is there a rough formula to transform osc6 bits into Mhz?

I don't know what you mean with the first quote, but for the second:

There is no relation between bits and MHz that can be calculated, because it also depends on the voltage. If you try my fork of cgminer it will periodically show the approximate MHz for the chip based on the time it takes to complete a job. The formula I have provided above is used in the code to show the expected GH/s for the chip based on the 'guessed' MHz

Here is how it looks like on my two chips board


EDIT: For 0.8V core voltage (if unchanged i.e. does not drop because of higher current drawn), the difference between 54 and 55 bits is ~8MHz

[punin]

Will someone translate what 864 rolled hash cores means? What's the speed and what's the difference between rolled and un-rolled cores?

Rolled core requires multiple clock cycles to complete the SHA256(SHA256(x)) (61+4 IIRC). 864 cores are working in parallel, so with clock of 220MHz the chip produces around 3 billion nonces per second. The difference between these two is in optimization of the design. In a nutshell rolled design is more resistant to errors in wafer production process and has other benefits as well.

[punin]

I think we'll be working mostly with trays as they're much easier to work with (and test) in the quantities usually needed by DIYers. One full box of trays is 2600 chips.

'Mostly' means that there still will be reels I guess, but are the boxes vacuum sealed or it's each tray separately or neither and chips baking will be necessary?

We are thinking of testing the chips at the packaging facility after which reel packaging could be made available. Cut tape will not be made available, as it ruins the vacuum.

[intron]

Will someone translate what 864 rolled hash cores means? What's the speed and what's the difference between rolled and un-rolled cores?

In a unrolled architecture the operations that make up
a certain calculation are laid down individually in silicon
in a pipelined fashion. This means every clock cycle another
calculation can be started and (after the pipe delay) every clock
tick another result is produced.

In a rolled architecture there is only a single 'main calculation
block' and partial results are fed back into that block until a
result is produced. Results are produce less frequent, but
the silicon area taken up is also smaller, so more loops can
be implemented on the same die.

When you roll up a calculation to a tight loop taking up less
area, changes of a defect of that block are smaller. However
there is a small penalty in silicon area for defining the stop condition
(i.e. 'is the calculation ready or not?').

[Taugeran]

Definitely want to see etches of the chip like over at http://zeptobars.ru/en/read/bitfury-bitcoin-mining-chip


maybe send them a single ref sample?



genuinely curious how you guys fit 108 more cores on the die while still using the same 55nm node O.o

[Taugeran]

OMG I've never been so happy to be wrong in all my life

Is there a rough formula to transform osc6 bits into Mhz?

I don't know what you mean with the first quote, but for the second:

There is no relation between bits and MHz that can be calculated, because it also depends on the voltage. If you try my fork of cgminer it will periodically show the approximate MHz for the chip based on the time it takes to complete a job. The formula I have provided above is used in the code to show the expected GH/s for the chip based on the 'guessed' MHz

Here is how it looks like on my two chips board


EDIT: For 0.8V core voltage (if unchanged i.e. does not drop because of higher current drawn), the difference between 54 and 55 bits is ~8MHz

the first quote was me being a tad sarcastic. I had always been under the impression the gentleman/woman behind bitfury design had no intention of revising the silicon to correct the errors inherent in Rev.1

[punin]

Definitely want to see etches of the chip like over at http://zeptobars.ru/en/read/bitfury-bitcoin-mining-chip

maybe send them a single ref sample?

genuinely curious how you guys fit 108 more cores on the die while still using the same 55nm node O.o

It's called Bitfury magic. He has a magic power that enables him to do such things when he moves his hands over the computer keyboard.

[BeepBeep2]

Definitely want to see etches of the chip like over at http://zeptobars.ru/en/read/bitfury-bitcoin-mining-chip

maybe send them a single ref sample?

genuinely curious how you guys fit 108 more cores on the die while still using the same 55nm node O.o

It's called Bitfury magic. He has a magic power that enables him to do such things when he moves his hands over the computer keyboard.

I wish bitfury was on 40nm
Would be massive raping of 28nm ASIC designs 

[KNK]

I wish bitfury was on 40nm
Would be massive raping of 28nm ASIC designs 

Even at 55nm it is better than most of the 28nm designs

[Taugeran]

I wish bitfury was on 40nm
Would be massive raping of 28nm ASIC designs 

Even at 55nm it is better than most of the 28nm designs

IF i ever win the USA Powerball lottery, I'll hop on board to finance R&D on taking ßitfury to a smaller node. 

[punin]

Definitely want to see etches of the chip like over at http://zeptobars.ru/en/read/bitfury-bitcoin-mining-chip

maybe send them a single ref sample?

genuinely curious how you guys fit 108 more cores on the die while still using the same 55nm node O.o

It's called Bitfury magic. He has a magic power that enables him to do such things when he moves his hands over the computer keyboard.

I wish bitfury was on 40nm
Would be massive raping of 28nm ASIC designs 

Be careful what you ask for

[tarmi]

punin, do you have some spare m-board for older 1.2v h-cards?

my setup with 16 cards is unstable.