Pages:
Author

Topic: Where are the 28nm FPGAs? (Read 5246 times)

member
Activity: 89
Merit: 10
May 01, 2013, 03:06:32 AM
#54
I think Artix-7 would be the best one/most cost efficient to mine BTC on, Virtex is a bit of overkill, and a lot more expensive.

Zynq would probably be amazing as well though (I think)
It seems to be a lot harder to get bitcoin miners to place and route successfully on Artix-7 than on Kintex-7.

Do you know, is that just because the Artix devices are smaller, or have Xilinx 'cost-optimised' the routing resources in Artix?

The usable chip area of the Artix-7-200T looks like a giant 'H', think of a building signage 'H' as in 'HOSPITAL', with two unusable 'parks' thrown in for good measure, a 'Central Park' and a 'Golden Gate Park'.

As far as I have been told, the usable chip area of the Kintex-7 is a huge rectangle.

Get it?

Thanks - the Zynq seems to have the same hole in the middle issue as Artix Sad
sr. member
Activity: 448
Merit: 250
April 30, 2013, 09:09:29 PM
#53
I think Artix-7 would be the best one/most cost efficient to mine BTC on, Virtex is a bit of overkill, and a lot more expensive.

Zynq would probably be amazing as well though (I think)
It seems to be a lot harder to get bitcoin miners to place and route successfully on Artix-7 than on Kintex-7.

Do you know, is that just because the Artix devices are smaller, or have Xilinx 'cost-optimised' the routing resources in Artix?

The usable chip area of the Artix-7-200T looks like a giant 'H', think of a building signage 'H' as in 'HOSPITAL', with two unusable 'parks' thrown in for good measure, a 'Central Park' and a 'Golden Gate Park'.

As far as I have been told, the usable chip area of the Kintex-7 is a huge rectangle.

Get it?
member
Activity: 89
Merit: 10
April 30, 2013, 06:31:39 PM
#52
I think Artix-7 would be the best one/most cost efficient to mine BTC on, Virtex is a bit of overkill, and a lot more expensive.

Zynq would probably be amazing as well though (I think)
It seems to be a lot harder to get bitcoin miners to place and route successfully on Artix-7 than on Kintex-7.

Do you know, is that just because the Artix devices are smaller, or have Xilinx 'cost-optimised' the routing resources in Artix?
legendary
Activity: 2772
Merit: 1028
Duelbits.com
April 20, 2013, 06:18:15 PM
#51

People are idiots or they're pumping asicminer share prices for their own good.

My bet would be on 2nd as asicminer share pumpers are pretty hardworking and religiously devoted group, doubt anyone is idiot to pay that price for hashing power that mines 0.6 btc at this difficulty and as such needs more than 4 months to meet ROI without difficulty rise.
sr. member
Activity: 247
Merit: 250
April 20, 2013, 04:14:06 PM
#50
These projects always interest me. Anyone direct me on material to read to learn to design PCBs for FPGA miners?

staff
Activity: 4284
Merit: 8808
staff
Activity: 4284
Merit: 8808
April 15, 2013, 08:01:13 PM
#48
The really high end FPGAs are targeted at people doing large scale logic simulations and other stuff that _cannot_ be done in a smaller part and so they have a premium price, they also often have very fast specialized IO hardware and other stuff that mining doesn't need.  I wouldn't expect them to be the most effective for mining. I expect mining to be most cost effective on whatever the largest or second largest low end devices is in any product line— keeping down the per chip costs and also dodging premium product pricing.

Of course, nothing beats actually simulating a design. Smiley
phk
newbie
Activity: 28
Merit: 0
April 15, 2013, 07:50:27 PM
#47
While the artix family is probably the most cost effective, a few kintex boards would probably be better/more power efficient than a ton of artix  boards.

Well, if you want to do that math, you need to use the power estimation tools from Xilinx/Altera.

As far as "few" vs. "ton", I think roughly you need (2) low-grade for (1) medium-grade.

In the case of Xilinx 7 devices, at $300 vs. $3k, that math is kind of easy.
full member
Activity: 126
Merit: 100
April 15, 2013, 07:14:21 PM
#46
While the artix family is probably the most cost effective, a few kintex boards would probably be better/more power efficient than a ton of artix  boards.
phk
newbie
Activity: 28
Merit: 0
April 15, 2013, 06:25:30 PM
#45

I don't think it's going to be that much of a problem, If it is I can/will use the Kintex, but trying to keep it as low cost as possible.

Someone should sit down and do the math.  I have a feeling that Artix/Cyclone is the families to go with.  In the next step up, you're buying more than just gates, and these hashing functions really don't need anything.
full member
Activity: 126
Merit: 100
April 15, 2013, 05:52:32 PM
#44
I think Artix-7 would be the best one/most cost efficient to mine BTC on, Virtex is a bit of overkill, and a lot more expensive.

Zynq would probably be amazing as well though (I think)
It seems to be a lot harder to get bitcoin miners to place and route successfully on Artix-7 than on Kintex-7.
I don't think it's going to be that much of a problem, If it is I can/will use the Kintex, but trying to keep it as low cost as possible.
hero member
Activity: 686
Merit: 564
April 15, 2013, 05:49:49 PM
#43
I think Artix-7 would be the best one/most cost efficient to mine BTC on, Virtex is a bit of overkill, and a lot more expensive.

Zynq would probably be amazing as well though (I think)
It seems to be a lot harder to get bitcoin miners to place and route successfully on Artix-7 than on Kintex-7.

I just updated the repo with a DSP48E1 design, for Kintex 7.  400MH/s using 80% of the DSPs, and ~25% of random logic.  I don't have an Artix-7 to port to, but I think they also have DSP48E1's.  I plan to start revamping the code base to better support multi-core designs and standardize the communication modules.  Hope to get my Kintex up to at least 1GH/s by throwing in two regular hashing cores.
Cool!
legendary
Activity: 1029
Merit: 1000
April 15, 2013, 03:15:20 AM
#42
fpgaminer: your board it's KC705 with K7325T?
400 MH/s is not impressive compared to 300MH/s from cripled Spartan6 (bitfury aproach of small rolled cores). I know that you will throw a second core but IMHO still not worth a hassle Wink
full member
Activity: 126
Merit: 100
April 15, 2013, 02:15:12 AM
#41
Quote
One of my friends and I are working on designing the PCB (he knows how to do it an stuff) so once I get that, I will be adapting the opensource FPGA miner to run on it (we will be using 2x Atrix-7)
will make a thread in proj dev to keep you all updated.
I just updated the repo with a DSP48E1 design, for Kintex 7.  400MH/s using 80% of the DSPs, and ~25% of random logic.  I don't have an Artix-7 to port to, but I think they also have DSP48E1's.  I plan to start revamping the code base to better support multi-core designs and standardize the communication modules.  Hope to get my Kintex up to at least 1GH/s by throwing in two regular hashing cores.
Nice, I'll be sure to test it out on my eval board when I get a chance. Shouldn't have to change too much.
hero member
Activity: 560
Merit: 517
April 15, 2013, 01:54:39 AM
#40
Quote
One of my friends and I are working on designing the PCB (he knows how to do it an stuff) so once I get that, I will be adapting the opensource FPGA miner to run on it (we will be using 2x Atrix-7)
will make a thread in proj dev to keep you all updated.
I just updated the repo with a DSP48E1 design, for Kintex 7.  400MH/s using 80% of the DSPs, and ~25% of random logic.  I don't have an Artix-7 to port to, but I think they also have DSP48E1's.  I plan to start revamping the code base to better support multi-core designs and standardize the communication modules.  Hope to get my Kintex up to at least 1GH/s by throwing in two regular hashing cores.
full member
Activity: 126
Merit: 100
April 15, 2013, 01:46:34 AM
#39
One of my friends and I are working on designing the PCB (he knows how to do it an stuff) so once I get that, I will be adapting the opensource FPGA miner to run on it (we will be using 2x Atrix-7)
will make a thread in proj dev to keep you all updated.
sr. member
Activity: 262
Merit: 250
April 14, 2013, 11:26:34 PM
#38
If I understand this correctly, since the 28nm chips is small we  can be place more of them into the pcb board?  what we need
now is the Price of this pcb board and the Hash rates it can produce?

The size of the die inside the chip is smaller, or more logic/LUT's can be fit in the same die area as the previous generation.

The actual package might or might not be smaller. The smaller FPGA's in the Artix family is available in 10x10mm packages (0.5 mm pitch), even though high capacity Artix FPGA's more suited to mining is only available in bigger packages, 19x19mm being the smallest. However, many hobbyist/PCB designers prefer larger packages which have pins which can be be soldered by hand (some hard core hobbyist can solder BGA's using other heating devices)
full member
Activity: 126
Merit: 100
April 14, 2013, 10:45:26 PM
#37
If all you investors want to set up a bounty for a pcb design using the artix-7, kintex-7, and vertex-7 (3 diff boards) that would be great
or if you send however much BTC you wanna offer for this to me, I'll set up the bounty, either way that's all I need to start work on a 28nm mining FPGA
legendary
Activity: 966
Merit: 1000
April 14, 2013, 10:42:08 PM
#36
<-- Also interested in investing in the dev effort in exchange for finished units.
member
Activity: 112
Merit: 10
April 14, 2013, 10:08:45 PM
#35
what we need now is for someone to create a pcb design before pricing is even spoken of.
Been saying this the whole time xD
someone makes me one I can get an FPGA togethor

BUMP FOR THIS ^^ SOMEONE!!!
Pages:
Jump to: