Bitcoin Forum

I compiled a table comparing the voltage converters supplying VCCINT to the Spartan-6 LX150, across all board manufacturers. The table below lists the maximum current that can be provided to each FPGA, as well as the efficiency of the converter at 8A (for 12V input, and 1.2V output):

ngzhang's ICARUS	15A or 20A(*) per FPGA	Texas Instrument TPS40192 or TPS40193 (http://www.ti.com/lit/ds/symlink/tps40192.pdf)	unknown efficiency
Enterpoint Cairnsmore1	12A per FPGA	Micrel MIC26950 (http://www.micrel.com/_PDF/mic26950.pdf)	87% efficient (from datasheet)
btcfpga ModMiner	10A or 13.3A(**) per FPGA	Delta D12F200 (http://www.delta.com.tw/product/ps/dcdc/std/download/data_sheet/DS_D12F200_11042008.pdf)	84% efficient (from datasheet)
fizzisist's X6500	10A per FPGA	Murata OKR-T/10-W12-C (http://www.murata-ps.com/data/power/okr-t10-w12.pdf)	83% efficient (from datasheet)
Ztex 1.15x/1.15y	8A per FPGA	Alpha and Omega AOZ1025D (http://www.aosmd.com/pdfs/datasheet/AOZ1025DI.pdf)	87% efficient (measured; datasheet says 75%)

(*) Some batches of ICARUS boards use the TPS40192 (rated 15A), others use the TPS40193 (rated 20A).
(**) The nice thing about the ModMiner is that you can decide to plug only 3 FPGAs in the mainboard, to give 13.3A to each FPGA, as they share a common 40A-rated voltage regulator.

eldentyrell: can you comment on how much current does your TML bitstream pull?

If chip is designed to deliver 20A it doesn't mean that whole DC/DC converter will. Many depends on coil, both efficiency and max current. ngzhang says that is 12A continous current for Icarus.
Micrel at 1.2V and 12A (12Vin) output have 83.5% efficiency. 87% is max at 3A output current.

Very useful information. Thanks!

I quoted efficiency when outputting 8A, for all converters, regardless of their max output current, so that we can compare them all fairly.

Efficiency values are nearly useless without real world measurements on the FPGA boards itself. Values from data sheet are marketing. Maybe it makes sense for DC/DC module as ModMiner uses but not for self layouted with external low side FET (high side FET is integrated) and coil as ZTEX uses. Let's concentrate on Mhs/W values, as long as all available bitstreams are around 200MHs per FPGA its is good comparable. In my eyes Features as "Layout improvement for FPGA cooling" are more important until there are cheap FPGAs with metal head-spreader available. 12A @1.2V are challenging for the lousy LX150 package for long time.

Hpman

You mean FET?

Your right, FED is more responsible for the fiat money :).

Ok. This is what I was told by Stefan too. A lot depend on the external layout and components. So the efficiency numbers I quoted would make sense only for the ModMiner and X6500 as they both use pre-made modules.

For that micrel also should be good. There are integrated FETs and if cainsmore engeneers take good care choosing coil, efficiency shouldn't be worse than specified.

No idea where the 75% come from. According to my measurements on a 1.15x board (w/o fan, current at 3.3V counted as inefficiency) the efficiency is about 87%. (6.7A @ 1.23V vs. 9.4W at 12V).

The 75% comes from an estimation of reading the efficiency graphs on page 6 of Alpha and Omega AOZ1025D (http://www.aosmd.com/pdfs/datasheet/AOZ1025DI.pdf).
Anyway I updated the post. Thanks for measuring efficiency!

The data sheet says that this is the efficiency with an AO4722 external MOSFET. But I do not use this thing.

At 12V input voltage the efficiency of the AOZ1025DI mainly depends on that MOSFET and the coil.