Pages:
Author

Topic: [ANN] Bitfury ASIC sales in EU and Europe - page 5. (Read 250506 times)

KNK
hero member
Activity: 692
Merit: 502
December 21, 2015, 03:02:56 AM
If I'm thinking right about how they described the J/GH relationship at higher hashrates, 6-7W per chip is well below the expected upper range (potentially over 20W?). I'd guess the footprint is changed to something a fair bit larger.

I tottaly agree on this.
There is simply no way to cool  about 20W from small QFN ( 8x8 if I remember  correctly)
How you got that 20W?
At that power the chip should provide 284GH/s not 184GH/s as stated ... maybe you mistyped 1 with 2 in your calculations
full member
Activity: 126
Merit: 100
December 21, 2015, 02:09:23 AM
If I'm thinking right about how they described the J/GH relationship at higher hashrates, 6-7W per chip is well below the expected upper range (potentially over 20W?). I'd guess the footprint is changed to something a fair bit larger.

I tottaly agree on this.
There is simply no way to cool  about 20W from small QFN ( 8x8 if I remember  correctly)
legendary
Activity: 3416
Merit: 1865
Curmudgeonly hardware guy
December 20, 2015, 07:01:57 PM
Are those posted numbers, or assumed numbers?

So, a bit of a question.

Quote
The design target was 40 gigahash per second with power efficiency of 0.06 joules per gigahash. On average, the measured power efficiency of tested engineering samples of the new BitFury Chip ranges from 0.055 joules per gigahash to 0.07 joules per gigahash
...
J / Gh metric in working modes starting from 55 Gh/s up to 180 Gh/s follows an almost linear relationship of 0.0011, while at 40 – 55 Gh/s measured slope converges to plateau.

So, let's see if I follow what they're saying. At 40-55GH the efficiency is pretty flat around 0.06J/GH (with deviation as noted above), and above this the J/GH increased with a slope of 0.0011

So the J/GH could be estimated by .0011(G-55) + 0.06 for any given hashrage "G" (greater than 55).

Which means at 55GH/0.06J they're drawing 3.3W and 100GH gets us about 0.11J/GH for an 11W chip. The 140GH air-cooling max sees 0.15J/GH and 21.5W dissipation. That's a pretty hot chip; I wonder how big it is to handle 22W under air cooling. That's getting into the A1 neighborhood, about where the A3222 ran if I'm remembering right. Quite a bit hotter than the BM1384, whose practical air-cooling limit was about 13W. Also the consideration that if it's doing that 21.5W at a tidy low voltage, say 0.6V, that's 36A per ASIC to get top clock.

I could be wrong, but that's how I understood the press release. How comfortable is QFN48 at 13W? BM1384 in a larger QFN56 aren't too uncomfortable with it, but if I'm figuring right your 13W would only get about 110GH and the 140GH actually clocks in at something like 20W. They only state the 0.06J/GH as the efficiency in the lowerbound 40-55GH operating, not across the whole range up to 180GH.
KNK
hero member
Activity: 692
Merit: 502
December 20, 2015, 05:56:16 PM
6-7W per chip is at the guaranteed speed of 100GH/s
~10W is the maximum for air cooling and 140GH
~13W for 184GH/s and immersion cooling
the minimum for 50-60GH is ~4W ... it seems it is not impossible to have the same QFN48 after all ... let the dreams come true Wink
legendary
Activity: 2128
Merit: 1005
ASIC Wannabe
December 20, 2015, 04:35:21 PM
If I'm thinking right about how they described the J/GH relationship at higher hashrates, 6-7W per chip is well below the expected upper range (potentially over 20W?). I'd guess the footprint is changed to something a fair bit larger.

The Antminer S7 is about 10w/chip. Its likely these would be a similar footprint, providing ~200GH
legendary
Activity: 3416
Merit: 1865
Curmudgeonly hardware guy
December 20, 2015, 03:52:12 PM
If I'm thinking right about how they described the J/GH relationship at higher hashrates, 6-7W per chip is well below the expected upper range (potentially over 20W?). I'd guess the footprint is changed to something a fair bit larger.
KNK
hero member
Activity: 692
Merit: 502
December 20, 2015, 03:39:33 PM
Rectified mains is almost PFC 1 (where the "almost" is because of diode losses) but it'd be completely useless. Add a capacitor filter across it to get a regular voltage that doesn't swing from peak to zero 120 times a second, and suddenly your current waveform is evil

There are many ways to skin a cat ... ad some super caps ... one for every 4-5 chips ... the rest will be done from the string itself, if you use the built-in current mirror.

If you're referring to the one in Georgia, Georgia is about 80% hydro and 20% geothermal.

Didn't know that, thanks for the info. Then neither of them (several kV) is good enough for direct rectifying for miner use.

No, we stopped our business but the design is for sale so other people may build a new board with the new Rev3 chip. I can't guarantee it will work, I don't have the technical knowledge. But if the Rev3 chips has the same string design as Rev2 big chance it will work with some small adjustments to the design.

wttbs, your design is exactly what i meant with
My guess is that they are using 12V strings, because at that voltage the power is distributed/balanced directly from the chips - no other components

Only if the footprint is the same.... and similar comms. I have a bunch of h cards with the chips pulled. Possibly a few tweaks to get the voltage right, man I would be ready to rock. But one can only dream right?
That is the problem - 6-7W per chip seems too much for the same footprint, but if you let me share your dream ... I also have some boards, which if the chip has the same pin-out (and footprint) will work great at 3.3V instead of 12V with just a minor modification

EDIT: 3.3V @0.4V per chip or 5V @0.6V per chip
sr. member
Activity: 453
Merit: 250
December 20, 2015, 02:58:38 PM
Only if the footprint is the same.... and similar comms. I have a bunch of h cards with the chips pulled. Possibly a few tweaks to get the voltage right, man I would be ready to rock. But one can only dream right?
legendary
Activity: 2210
Merit: 1109
December 20, 2015, 02:06:14 PM
Do you plan to make new board with the new chips ?

No, we stopped our business but the design is for sale so other people may build a new board with the new Rev3 chip. I can't guarantee it will work, I don't have the technical knowledge. But if the Rev3 chips has the same string design as Rev2 big chance it will work with some small adjustments to the design.
legendary
Activity: 1484
Merit: 1004
December 20, 2015, 01:59:16 PM
Do you plan to make new board with the new chips ?
legendary
Activity: 2210
Merit: 1109
December 20, 2015, 01:42:04 PM
We have a M-board + H-board(s) based on the Rev 2 Bitfury. Very efficieny miner and some our miners are running stable over a year 24/7 without any problem. If anyone is interested in the design send me pm, maybe with some small changes to the design you can make it work for the rev 3chip.

https://bitcointalksearch.org/topic/yiazo-ybf-bitfury2-miner-30-up-to-380ghs-647102
legendary
Activity: 1484
Merit: 1004
December 20, 2015, 12:45:07 PM
Hope to see the H-Board back in action!  Grin
legendary
Activity: 1600
Merit: 1014
December 20, 2015, 12:43:22 PM
If you're referring to the one in Georgia, Georgia is about 80% hydro and 20% geothermal. Also more like 90/10.
f
legendary
Activity: 2128
Merit: 1005
ASIC Wannabe
December 20, 2015, 12:06:18 PM
Good chip info. I never worked with the 55nm stuff, didn't even own any until a few months ago so I don't really know much about how they behaved.

55nm was great gear. the system provided chip-by-chip feedback that only spondoolies has been able to exceed (and took an extra year to produce). cooling the h-boards was easy with 120mm fans and you could easily apply stick-on heatsinks sourced from ebay/alibaba/etc to allow pushing well past 25GH/board (required pencil mod)
legendary
Activity: 1666
Merit: 1185
dogiecoin.com
December 20, 2015, 10:45:22 AM
According to their press releases the datacenter is running on green power - solar or probably mostly wind. The solar is almost at constant voltage, so a string built for that voltage directly, will not need any power management parts, but ... no sun no fun. The wind turbines are usually working at 600-700V AC, which is too much for a string, just because it will be too long for the SPI communication - there is a 17ns per chip propagation delay (for the 55nm chip), which makes the longest chain 300-400 chips max if it is a single string for both power and communication.

If you're referring to the one in Georgia, Georgia is about 80% hydro and 20% geothermal.
legendary
Activity: 3416
Merit: 1865
Curmudgeonly hardware guy
December 20, 2015, 10:08:49 AM
Rectified mains is almost PFC 1 (where the "almost" is because of diode losses) but it'd be completely useless. Add a capacitor filter across it to get a regular voltage that doesn't swing from peak to zero 120 times a second, and suddenly your current waveform is evil - that is exactly why datacenter PSUs have good PFC. If you tried to run multiple megawatts of capacitor-filtered rectified mains, well, good luck. Maybe if they used an active PFC unit in standalone as a constant-voltage DC supply? That could be interesting.

Higher voltage tends to be better efficiency in part because of resistive losses. The same power in a high voltage means much less current; relatively fixed-voltage losses like diodes or BJTs would decrease power loss linearly but a lot of things are resistive losses (power FETs, inductor windings etc) which lose power proportional to current squared. Quadrupling the voltage reduces current to one fourth, which cuts resistive losses to one sixteenth. Same applies to the efficiency gain of running supplies on 240 instead of 120 mains, especially for active PFC supplies; less current is required through input boost components to get the required power to the output regulator.

Good chip info. I never worked with the 55nm stuff, didn't even own any until a few months ago so I don't really know much about how they behaved.
KNK
hero member
Activity: 692
Merit: 502
December 20, 2015, 07:26:07 AM
Except without good PFC the utility provider would murder them.
Datacenter PSUs usualy have good enough PFC and rectified mains is almost PFC 1
The higher the voltage, the better is efficiency - compare some high power PSU spec it goes from 83% to 90% for the same class for 12V and 48V.

According to their press releases the datacenter is running on green power - solar or probably mostly wind. The solar is almost at constant voltage, so a string built for that voltage directly, will not need any power management parts, but ... no sun no fun. The wind turbines are usually working at 600-700V AC, which is too much for a string, just because it will be too long for the SPI communication - there is a 17ns per chip propagation delay (for the 55nm chip), which makes the longest chain 300-400 chips max if it is a single string for both power and communication.

My guess is that they are using 12V strings, because at that voltage the power is distributed/balanced directly from the chips - no other components, while for 48V you will need to use the current mirror (which is built-in in the chip) and that means some more external components even if cheap.

I would be very happy to see those external components made internal for the next revision (the 16nm one which may be considered current), because it is possible (for most of them) and will even work better in that case. The only problem is the communication again, because for 12V the string will be close to the limit and the best voltage in that case will be 5V
legendary
Activity: 3416
Merit: 1865
Curmudgeonly hardware guy
December 19, 2015, 06:45:24 PM
Except without good PFC the utility provider would murder them.
sr. member
Activity: 427
Merit: 251
- electronics design|embedded software|verilog -
December 19, 2015, 06:38:29 PM
I would not be surprised to see thier in-house hardware operating with 48VDC power supplies for density
--
Or rectified mains;)
legendary
Activity: 2128
Merit: 1005
ASIC Wannabe
December 19, 2015, 12:55:12 PM
Will be reopening the shop soon! The chip is crazy and amazing. Anyone interested in H-cards with 16nm chips?

Thats a name i haven't heard in a long time...

Id love new h-cards! But at the same time I think an assembled, standalone unit similar to the S1 or S5+ would be great. keep the lightweight, barebones style, just include a basic frame to prevent wobbles and a pair of fan brackets

I think it comes down to chip cost and overclocking feasibility. Gen1 could be pushed an extra 40% under the watch/modification of advanced users, or run with almost no added cooling (heatsinks or fans) at stock.
Pages:
Jump to: