FPGA development board "Icarus" - DisContinued/ important announcement

[ngzhang]

Icarus will for sell in a few days.

price?  just guess.

Hopefully <$500 per unit?

Any plans to have volume pricing?

 

maybe slightly higher, but not much. in effect, 500$ is enough for the manufacture cost, but months work is a REALISTIC problem.

the slot is to connect to a motherboard? so would not need either USB and power cable?

yes, the slot is used for connect to a motherboard. but in mining purpose, this architecture is highly NOT recommend. i design 2 of "link port" for cross boards connect. In fact, just 1 USB port could support hundreds of Icarus. (But by present firmware's coding bugs, this function can not work well. ) So, using a complex motherboard just for remove the power cable is inefficent.

This looks great. Wouldnt it be possible to run this with passive cooler?
Whats the price going to be?

highly NOT recommend. because of the chip packaging, a low thermal resistance cooling method is needed. relation to the size,  a 25mm fan is enough. now i use a NIDEC ball-bearing fan, high quality.

the slot is to connect to a motherboard? so would not need either USB and power cable?

Doesn't look like any of the traces from DIMM connector reach the USB microcontroller.  It does look like it could get power from the edge connector though.  That combined with a convenient way of mounting multiple cards would make the backplane worth it.  If backplane handled data to well that would just be awesome.

this function only need a slightly UCF code modification. certainly, you need build a mother board. the DIMM connector provide near 100 GPIOs, JTAG, and power.
attention: the FPGA's config bitstream CAN be stored on board. on this board, each FPGA has 1 64Mb SPI flash with it, so it can boot up when power on in 500ms.
this means the upper computer's work is simply transform the mining data. maybe a hacked router is fairy enough for it.
the bitstream download can be finish with a XILINX platform cable.

I'm hoping this will be the unit that puts us at $2/Mhash... so $600 would be perfection

EDIT: Math fail

maybe lower. but please, this is a good development board, not just a mining machine. I think it's much ease for resell if you want to give up mining bitcoins   .

This really is an exciting development. Board looks very nice compared to other hacks.

What is price ?

What is memory connector for ? I think it is just for the power and the data uses USB.

This is a development board. not just for mining. the DIMM connector is mainly for other purpose.

When this is intended as a bitcoin miner, why the 6 layer pcb?

Heck, I can imagine it would be doable on a double layer board with minimal I/O, if not even single layer if the power for both chips can be supplied by a single power converter. The thing I want is a pcie card with nothing on it except 2 fpgas, and the power circuit.

I am willing to bet this can be done with double layer and around USD 1/mhash.

there is enough of product as you say, isn't it?   beyond that, there is another side: a 6 layers PCB's heat dissipation is much better than a 2 layer PCB, so the FPGA chips' core temperature will lower than just using a 2 layers PCB . there are about 3 complete copper planes in the PCB, and all signals has reference planes. and it's necessary for diff-pairs routing.

Great work.
For a such heavy routing 6 layer PCB its necessary. I'm waiting for a price...
I like the idea for routing I/Os, maybe I buy one... If you will sell those..

certainly they are for sell, if you don't need a "open box can use" mining code, i can ship you 1 today.
now we are debugging around the mining control software. and this will lasts some days.

 Excellent work, Ngzhang!!


  Speculation on price; Those particular spartans are $158 for <60, the boards have a LOT of components and the PCBs not cheap. High amp regulator $30+. Maybe saved a few bukcs on a low end uart? Not seeing it but am not real familiar with them. Not sure if he has a license fee attached but will add it in case.


  builders markup included result = XLX $386, PCB+License $145, Components $110, usb and power cable $7

 =   ~$648  or $1.80/MH  

Did I miss any thing?

and i do all of this for free?  

relax, it will not reach 648$.  

中国人的骄傲。。。。。

speak english please...

[1713571471]

1713571471

[1713571471]

1713571471

[1713571471]

1713571471

[1713571471]

1713571471

[bulanula]

What is this "other purpose" you keep talking about I thought these were on good for mining. Maybe good for WPA too ?

[Dexter770221]

Alot of I/Os are routed to connectors so you may develop some neaty devices. For example, if you want, there will be no problem to develop and connect module with Ethernet, monitor, PC keyboard, USB host, etc...

[ngzhang]

Alot of I/Os are routed to connectors so you may develop some neaty devices. For example, if you want, there will be no problem to develop and connect module with Ethernet, monitor, PC keyboard, USB host, etc...

exactly.

example1: build an WIFI daughter board and connect to the FX connector on the top, then implement some logic in FPGA, and then write some code, finally, Icarus can solo mining in a corner.

example2: connect some Power relay and a GSM module with it, write some code, then you can control your lamp with some SMS from your phone.

etc.

[bulanula]

Alot of I/Os are routed to connectors so you may develop some neaty devices. For example, if you want, there will be no problem to develop and connect module with Ethernet, monitor, PC keyboard, USB host, etc...

exactly.

example1: build an WIFI daughter board and connect to the FX connector on the top, then implement some logic in FPGA, and then write some code, finally, Icarus can solo mining in a corner.

example2: connect some Power relay and a GSM module with it, write some code, then you can control your lamp with some SMS from your phone.

etc.

Well ATM I think we should just focus on making this board really good for mining and mining only. All these extra parts make the device even more expensive I think. Please tell me when I can get an FPGA module of size smaller than a 5870 ( half size of a GPU ) and gets 500 Mhash/s and price is under $500 and consumption is under 20W.

[gmaxwell]

Well ATM I think we should just focus on making this board really good for mining and mining only. All these extra parts make the device even more expensive I think. Please tell me when I can get an FPGA module of size smaller than a 5870 ( half size of a GPU ) and gets 500 Mhash/s and price is under $500 and consumption is under 20W.

Your request for a pony is being processed. You are #6182382813 in line. Please stand by.

[nelisky]

So when can we expect to be able to order this ? :p

[bulanula]

So when can we expect to be able to order this ? :p

Same problem as BFL creation. No concrete offers and pricing and release dates At least this one is real.

[mtminer]

Those are pretty cool. Would love to know what the final price is going to be.

[sadpandatech]

Those are pretty cool. Would love to know what the final price is going to be.

  He gave us enough of a range on price to do some math with. Less than $648 and more than $550(since he stated $500 was a bit less than his costs).

  I am still gonna stick to within 10% or so of my orig speculation once he crunches all the math on parts and labor costs involved. Unless he plans to order 500+ chips at once and contracts with an ultra cheap assembler in hopes you forum peoples will buy them before a cheaper design can be built.....

[gmaxwell]

  He gave us enough of a range on price to do some math with. Less than $648 and more than $550(since he stated $500 was a bit less than his costs).

Too much more that his costs and it's going to lose attractiveness vs the X6500 — yes, it's a more flexible board (it looks beautiful and deserves a good price) but for people who are primarily interested in mining the amount of premium they'd pay for that over a more minimal design probably isn't great.  (the idea of increased resale value is good, but it's pretty theoretical... and probably not worth that much in present day dollars)

Fortunately it looks like all the boards out there already have pretty healthy margins— they have to because of the small quantities we're talking about... so there should hopefully be room to take some of the increased parts costs without making it uncompetitive.

[sadpandatech]

  He gave us enough of a range on price to do some math with. Less than $648 and more than $550(since he stated $500 was a bit less than his costs).

Too much more that his costs and it's going to lose attractiveness vs the X6500

  ummm, the X6500 is $580 for only 200MH.... = $2.9/MH  This Icarus build, even at $620 and 360MH will be cheaper per MH. = $1.722~/MH  Even at my guess of $648, its $1.8/MH. He has plenty of room to work with there.  If it was me I'd start em at $700, or $1.94/MH, for the first round and move down from there as bulk improved costs..

[gmaxwell]

 ummm, the X6500 is $580 for only 200MH.... = $2.9/MH  This Icarus build, even at $620 and 360MH will be cheaper per MH. = $1.722~/MH  Even at my guess of $648, its $1.8/MH. He has plenty of room to work with there.  If it was me I'd start em at $700, or $1.94/MH, for the first round and move down from there as bulk improved costs..

250 and not 200: https://bitcointalk.org/index.php?topic=40058.msg584606#msg584606

The X6500 is only at that speed due to not very well optimized mining code which hasn't been touched for sometime.  They were saying in #bitcoin-fpga that now that the board production was all sorted some time was going to get spent on the mining code. It's using the same -3 grade parts which ZTEX gets 190MH/s per FPGA on (and ZTEX's code is OSS).  While it's possible that they flubbed the power design or the clock signal quality and can't actually get it to that speed, I think thats unlikely.

It appears that Icarus is using the "slower" -2 speed grade part— which is probably the best part to use in terms of $/MH from a parts cost perspective.  But the parts cost only matter if they're passed on to the buyer.

[sadpandatech]

 ummm, the X6500 is $580 for only 200MH.... = $2.9/MH  This Icarus build, even at $620 and 360MH will be cheaper per MH. = $1.722~/MH  Even at my guess of $648, its $1.8/MH. He has plenty of room to work with there.  If it was me I'd start em at $700, or $1.94/MH, for the first round and move down from there as bulk improved costs..

250 and not 200: https://bitcointalk.org/index.php?topic=40058.msg584606#msg584606

  That still puts X6500 more expensive. ;p  Not trying to argue with ya, m8. But your orginal comment of it in comparison to the x6500 is off a bit.  Also, X6500 started out at 200. Icarus is in the same boat and has been stated as such by the developer that he is still working on optimizing the mining code.

  Point is, from the range Ng has hinted to thus far, his mark-up will be much less than others and his end-user price per MH is better...

[gmaxwell]

  That still puts X6500 more expensive. ;p  Not tring to argue with ya, m8. But your orginal comment of it in comparison tot he x6500 is off a bit.  Also, X6500 started out at 200. Icarus is in the same boat and has been stated as such by the developer that he is still working on optimizing the mining code.

  Point is, from the range Ng has hinted to thus far, his mark-up will be much less than others and his end-user price per MH is better...

It won't be better (from a purely myopic $/MH basis) unless it's equal to or lower than the x6500 unless he undercuts the x6500 price. Which is kinda sad because it looks like a much nicer board (for reasons unrelated to $/MH).

Or would you like to take a bet that a board with the _lower_ speed grade version of the same part is going to deliver higher performance?   I might take the opposite of that bet.

[DeathAndTaxes]

Icarus is in the same boat and has been stated as such by the developer that he is still working on optimizing the mining code.

Well not exactly the same boat.  The theoretical performance for a LX150 is ~200MH/s.

Ztek is getting 190MH/s on 1x LX150 or 90% of peak.
Icarus is getting 360MH/s on 2x LX150 or 80% of peak.
X6500 is getting 250MH/s on 2x LX150 or 62.5% of peak.

So X6500 has the POTENTIAL to make larger increases in performance and thus larger changes in MH/$.  Ztek is pretty much at peak performance.  Looks like he is looking at changing the voltage that might pick up another ~4 or 5 MH but it is pretty optomized.  Icarus could pick up some speed (and thus moderate reduction in MH/$) but not much.

Granted I wouldn't buy a board only on potential but the X6500 is interesting simply because it has a low cost per FPGA which potentially could delivery higher performance per $.  Still they got their work cut out for them because all that really matters in the end if MH/$.

Point is, from the range Ng has hinted to thus far, his mark-up will be much less than others and his end-user price per MH is better...

Hinted but I would like to see a firm price list w/ volume pricing.

Of course if BF Labs 1GH/s for $500 isn't a scam well that pretty much kills all these projects.

[sadpandatech]

Icarus is in the same boat and has been stated as such by the developer that he is still working on optimizing the mining code.

Well not exactly the same boat.  The theoretical performance for a LX150 is ~200MH/s.

 Quite right you are. By same boat, I was referring more to the initial public sales release aspect.

Ztek is getting 190MH/s on 1x LX150 or 90% of peak.
Icarus is getting 360MH/s on 2x LX150 or 80% of peak.
X6500 is getting 250MH/s on 2x LX150 or 62.5% of peak.

 In my haste I had neglected to view the Ic/X from this perspective of potential. Thanks for taking the time to sum that up.

So X6500 has the POTENTIAL to make larger increases in performance and thus larger changes in MH/$.  Ztek is pretty much at peak performance.  Looks like he is looking at changing the voltage that might pick up another ~4 or 5 MH but it is pretty optomized.  Icarus could pick up some speed (and thus moderate reduction in MH/$) but not much.

Granted I wouldn't buy a board only on potential but the X6500 is interesting simply because it has a low cost per FPGA which potentially could delivery higher performance per $.  Still they got their work cut out for them because all that really matters in the end if MH/$.

 Yep and yep. =)

Point is, from the range Ng has hinted to thus far, his mark-up will be much less than others and his end-user price per MH is better...

Hinted but I would like to see a firm price list w/ volume pricing.
Of course if BF Labs 1GH/s for $500 isn't a scam well that pretty much kills all these projects.

  He did a bit more than hint when he replied to my price speculation with. "relax, it will not reach $648.  "
  Volume price for parts is very easy to evaluate but we've no idea what he would want per license if he chooses to keep it as a private 'for sale' project.

  Frigging BF Labs, the 10 ton unicorn in the room for sure.  They better put some fire under it one way or the other. I'm about 1 quote away from going ahead with my endeavours without them.  



    Cheers m8

[sadpandatech]

 That still puts X6500 more expensive. ;p  Not trying to argue with ya, m8. But your orginal comment of it in comparison to the x6500 is off a bit.  Also, X6500 started out at 200. Icarus is in the same boat and has been stated as such by the developer that he is still working on optimizing the mining code.

  Point is, from the range Ng has hinted to thus far, his mark-up will be much less than others and his end-user price per MH is better...

It won't be better (from a purely myopic $/MH basis) unless it's equal to or lower than the x6500 unless he undercuts the x6500 price. Which is kinda sad because it looks like a much nicer board (for reasons unrelated to $/MH).

Or would you like to take a bet that a board with the _lower_ speed grade version of the same part is going to deliver higher performance?   I might take the opposite of that bet.

  Unless I am really missing something in my math here, it IS better from a $/MH at the price we have to go by assuming Icarus price.
  Can you show me how $580/250 is cheaper than $648/360?

  X6500= $580/250MH= $2.32 per MH
  Icarus= $648/360MH= $1.80 per MH

  If, Ng's reply to me was misunderstood and the buyer price is higher, it would have to be  $835.2 to be the same price as X6500 per MH.  That's  $2.32*360MH=$835.20

  I'm not sure where we are misunderstanding each other at here. I am completely non-biased to either of these two projects. I think every one of the guys working on FPGA builds are absolutly amazing!

  Kind regards,
      Derek

[gmaxwell]

 Unless I am really missing something in my math here, it IS better from a $/MH at the price we have to go by assuming Icarus price.
  Can you show me how $580/250 is cheaper than $648/360?

  X6500= $580/250MH= $2.32 per MH
  Icarus= $648/360MH= $1.80 per MH

  If, Ng's reply to me was misunderstood and the buyer price is higher, it would have to be  $835.2 to be the same price as X6500 per MH.  That's  $2.32*360MH=$835.20

  I'm not sure where we are misunderstanding each other at here. I am completely non-biased to either of these two projects. I think every one of the guys working on FPGA builds are absolutly amazing!

Because $580/380 is $1.52.  380 is the speed we should expect from two of the the -3 parts, because thats the speed that the currently best available open source code typically gets on them.

It would be surprising if the X6500 is unable to reach the same speed ZTEX is getting with _the same part_ after adopting the same double stage pipeline design spending the couple hours brute-forcing the optimizer settings to get it to route for 200mhz. It might be the case that the -2 part also ends up just as fast due to the vulgarities of binning, but I know of no reason to expect it to be to be faster.

(and keep in mind, of these three the ztex is the only one you can click-and-buy for immediate shipment right _now_ so if you want to talk about dollars per MH _this moment_ rather than what we can reasonably expect— then Icarus and X6500 are both ~$∞/mh. )

And yes, agreed. It's all good work. And regardless of how the $/mh ultimately works out when the boards are generally available. The Icarus boards look like they'll be an excellent value due to their increase flexibility.

[ngzhang]

  heated disscussion.

Those are pretty cool. Would love to know what the final price is going to be.

  He gave us enough of a range on price to do some math with. Less than $648 and more than $550(since he stated $500 was a bit less than his costs).

  I am still gonna stick to within 10% or so of my orig speculation once he crunches all the math on parts and labor costs involved. Unless he plans to order 500+ chips at once and contracts with an ultra cheap assembler in hopes you forum peoples will buy them before a cheaper design can be built.....

less than 600$, first batch.

  That still puts X6500 more expensive. ;p  Not tring to argue with ya, m8. But your orginal comment of it in comparison tot he x6500 is off a bit.  Also, X6500 started out at 200. Icarus is in the same boat and has been stated as such by the developer that he is still working on optimizing the mining code.

  Point is, from the range Ng has hinted to thus far, his mark-up will be much less than others and his end-user price per MH is better...

It won't be better (from a purely myopic $/MH basis) unless it's equal to or lower than the x6500 unless he undercuts the x6500 price. Which is kinda sad because it looks like a much nicer board (for reasons unrelated to $/MH).

Or would you like to take a bet that a board with the _lower_ speed grade version of the same part is going to deliver higher performance?   I might take the opposite of that bet.

Icrus uses a  -2I fpga, "I " means the industrial temperature. speed/voltage spec are guaranteed across the industrial temperature range. by my testing, -2I  is very alike with -3C, and the price is also the same, bug much easier to buy in my country.

Icarus is in the same boat and has been stated as such by the developer that he is still working on optimizing the mining code.

Well not exactly the same boat.  The theoretical performance for a LX150 is ~200MH/s.

Ztek is getting 190MH/s on 1x LX150 or 90% of peak.
Icarus is getting 360MH/s on 2x LX150 or 80% of peak.
X6500 is getting 250MH/s on 2x LX150 or 62.5% of peak.

So X6500 has the POTENTIAL to make larger increases in performance and thus larger changes in MH/$.  Ztek is pretty much at peak performance.  Looks like he is looking at changing the voltage that might pick up another ~4 or 5 MH but it is pretty optomized.  Icarus could pick up some speed (and thus moderate reduction in MH/$) but not much.

Granted I wouldn't buy a board only on potential but the X6500 is interesting simply because it has a low cost per FPGA which potentially could delivery higher performance per $.  Still they got their work cut out for them because all that really matters in the end if MH/$.

Point is, from the range Ng has hinted to thus far, his mark-up will be much less than others and his end-user price per MH is better...

Hinted but I would like to see a firm price list w/ volume pricing.

Of course if BF Labs 1GH/s for $500 isn't a scam well that pretty much kills all these projects.

I think a well designed PCB, enhanced on board power regulator, and high-quality assembling is also important for a long run. and that's why people do not buy the cheapest brands of GPUs / PSUs for mining.