Timetravel10 fits in a single VU13P.  You partition the FPGA into 16 blocks, and store about 14 partial bitstreams for each block.  Then you do a dynamic partial reconfiguration from DDR4 to build the pipeline at the start of each block based on the current algorithm sequence.  Yielding one hash per clock (i.e. 500MH/s @ 500MHz).  You need 16 blocks because some functions like Groestl and Echo require 2 blocks.  The FPGA can reconfigure itself in 0.25 seconds.  The problem with Timetravel and X16R/X16S is the long time it takes to load the DDR4 bitstream table via USB.  And you lose it if there is a power outage and must reprogram the DDR4 on each FPGA.  This where utilizing the PCI bus would be an advantage.

This is correct.  The Keccak launch is primarily to iron-out power and thermal issues, determining unit-to-unit variance in over-clocking capacity, as well as auto-detecting FPGA's attached to the PC, and various other proofs of concept for scaling up operations.  The Tribus launch on June 15 is the first bitstream that generates significant profit. 

Some quick updates:

- The launch is still on schedule for May 30.  The initial launch will be for the VCU1525 as Bittware is still working to increase the power limit of their board
- It is now time for the lucky people who already have VCU1525's to private message me, so you can receive some pre-release software in the next couple of days for initial testing and to get some feedback on how hard you find the set up process
- An attempt was made (long ago) to negotiate getting a % of hardware sales with Xilinx/Bittware but this was not successful
- The Molex acquisition was a surprise; hopefully this won't affect Bittware's participation in crypto mining
- Keccak will be launched on May 30, and Tribus will be released on June 15, followed by Phi1612/Skunkhash (which are very similar); then Lyra2z, and following that it will be either CN7 or Neoscrypt; suggestions are welcome as to desired algorithms


If you already have a VCU1525 (a real one, not AWS instance), then please message me ASAP to receive your pre-release software.

I get the general idea but I still have a hard time seeing how to do that with the USB-UART pins.

As far as 'resale' of these FPGA cards, there is also a tremendous future for AI mining.  In fact, if tens of thousands of people were mining crypto with these boards, and an AI mining server became available, they are so powerful that you might end up with a 'Skynet' type superintelligence, or the Kurzweil 'singularity' event. 

My point is that if, for some reason, mining crypto became unprofitable, someone could re-sale access to the FPGA hardware for AI, weather prediction or any other compute-heavy tasks.

I had a conference call with Xilinx before announcing this project and they agreed to transition the VCU1525 into a production product for a slightly higher price; the production version will be released in June/July, in much higher quantities.

Regarding the question about bitstream compatibility, the answer is no, a bitstream built for VCU1525 will not work on the XUPP3R.  The clock & USB UART are on different pins, so I must 'build' two versions of the bitstream, one for each card.  Although the change in the code is tiny, it still takes the tools 5+ hours to re-run place & route, with multiple runs needed to get a success.

Both the VCU1525 and XUPP3R have a circuit that will shut down the FPGA power supply if the temperature exceeds limits.  This doesn't prevent you from trying to pull too much current for a short time though.

Another important note for people considering this hardware.  The VCU1525 has a fixed core voltage of 0.85V.  The XUPP3R from Bittware is more flexible and you can change the core voltage from 0.74V to 1.11V, which allows a much greater flexibility in terms of over-clocking the FPGA.  For memory intensive algorithms, a higher voltage will produce faster hash rates, vs. for power limited algorithms, a lower voltage would likely produce better hash rates.

If you have a Zetex 1.15y with four Spartan LX150, a tiny modification to the code will change it from SHA-256 to SHA-224 and it is quite profitable mining Bismuth.  Coinosaurus pool can give you the JSON protocol to submit shares since Bismuth pools don't use stratum.

Regarding the algorithms I spent months analyzing every single one to determine which were most suitable to FPGA's.  I checked them all.

We all knew the Antminer-Z was coming.  But it may not be too late for equihash; hopefully all the equihash coins will now fork to equihash-2, which the FPGA's would adapt to, and the Z9 becomes a paperweight like the Cryptonight miners.

For Ravencoin and Bitcore, the DDR4 is not used for hashing at all, but rather to store hundreds of different FPGA configuration bitstreams, which allows the entire FPGA to rapidly reprogram itself on every block based on the algorithm sequence for that block.  Cryptonight7 and Equihash, on the other hand, would use the external memory for actual hashing.  Whether CN7/Equihash gain from additional RAM depends on how those algorithms are ultimately implemented.

My Bittware board has no external memory, and I will add either standard PC DDR4, or Bittware QDR-SRAM, as needed in the future; only Ravencoin (DDR4), Bitcore (DDR4) and Equihash (QDR) need external memory.  Cryptonight7 would, although I haven't examined it in depth.  Phi, Keccak, SHA-224, Tribus, X17, Lyra2v2, Neoscrypt, Xevan; no external memory needed.
 

The VU9P FPGA is a 16nm device, not 28nm.

VCU1525 has 2 x QSFP28 connectors, so you link 2 boards with 2 x 100 gigabit ethernet cables, and two boards is the max that can be daisy chained.  With the Bittware XUPP3R board, since it has 4 x QSFP28 connectors, there is no limit to the daisy chain length.  These specialized ethernet cables have nothing to do with internet, they are solely so data can flow from one FPGA board to the next board.  The cables are around $40 each.

Personally I think GPU mining will still be around for a while.  I believe the market is going to rise so dramatically this fall that all existing GPU rigs will be making a lot of money, even with rises in difficulty.

Clarifying the projected hash rates
X17: 2 cards daisy chained get 600MH/s total
X16R: 2 cards daisy chained get 600MH/s total
Xevan: 4 Bittware cards daisy chained get 600MH/s total

I'm extremely busy trying to hit the May 30 launch date so I don't have a great deal of time to follow this thread, and I have so many PM's I can't possibly read/respond to them all.  However I did glance at some of the common questions and I thought I would give some short answers:

- I've been contacted by many big farm owners and so on, and for them as well as anyone who is hesitant, please just wait for proof.  The only real proof will come when guinea pigs like Ruplikminer get their hardware in the end of May and start hashing and reporting back.  Don't worry about a shortage of cards; Xilinx and Bittware are preparing for high volumes
- Of course I considered renting Amazon AWS F1 (VU9P) FPGA instances and mining with them since you can make 4x more than the rental cost of the FPGA.  However by all accounts Amazon has some sort of watchdog software and if it catches you mining, your instances gets shut down.  Most people report being shut down within 8 hours if they try renting FPGA's with Amazon and mining with them
- I'm trying to help the crypto community move forward and as such I'm happy to help as many other FPGA developers as possible.  Digitalcruncher came very close to my Keccak hash rate with 24 Keccak instances @ 550MHz.  What most people might forget is you don't need to calculate the full 256-bit hash in the FPGA.  All you need is the top 48-bits to see if it is less than the target, pass the winning nonce back to the PC and the PC will regenerate the full hash using the nonce.  So start with the 48-bit hash fragment you are interested in, and propagate backwards to eliminate calculations; this eliminates most of the final round of the algorithm, and the same optimization applied to the 1st round yields similar benefits.  Then, for S-boxes always use compressed combinational logic forms that have been published, and try running them through an Espresso logic minimizer like Logic Friday, which can sit there for 2 days optimizing your S-box (vs. Vivado which will not spend more than a few minutes trying to optimize it during synthesis)
- USB data transfer rate is 12MBaud so it is fine to use USB2 ports.  Technically you can have way more than 8 FPGA cards on a single PC, with lots of PCIe-USB adapter cards; eventually the load on the CPU and RAM will become high, so for a 16-card rig with two power supplies, a heavier duty CPU and RAM might be necessary and I can't vouch for the stability of such a big rig
- I hear Avnet only ships to North America.  The VCU1525 is available from Digikey for around $4500 USD and they ship internationally.  Bittware is really the best of the companies and is the only company that will actually help you configure (or even pre-configure) your card and if you have problems they will help you 1-on-1 because they are a smaller company and they will be mining with some of their own FPGA's (and Bittware will ship internationally no problem).  Plus the Bittware card can do Xevan and Equihash better than the VCU1525.
- All stratum communication is done by the PC with a fork of tpruvot's CPU miner which has been adapted to communicate with the FPGA's.  The PC-side code is a messy hack right now, so I have enlisted a good friend in Spain to help clean it up for the launch and make it more user friendly and resilient.  I have been contacted by three super-expert FPGA programmers and after the initial algorithms are launched I hope to make it more of a group effort to get the remaining algorithms implemented faster with their help, so if I were to die/disappear the project would continue
- You can calculate how many FPGA's each coin can handle before miners overwhelm the coin and profits drop.  You have around 2.9 BTC of daily rewards available on Keccak and over 20 BTC of daily rewards available on Phi1612, enough to support many hundreds of FPGA's, however as more algorithms are released, we will start hitting the huge coins and I don't foresee a dramatic and sudden profit drop as usually happens with ASICs
- It takes 40 seconds to switch the FPGA from one algorithm to another using Xilinx Vivado Lab Edition; this can be done remotely if you live away from your farm, but it is not automatic like for GPU's
- It is possible by using data from initial algorithms to project the hashrate within +/-10% for future algorithms, and in that light the expected rates (per card) are about 300MH/s for X17 & X16R, 25MH/s for Neoscrypt, 600MH/s for Lyra2v2; 150MH/s for Xevan (Bittware card only for Xevan!).  For Equihash it is much harder to calculate the projected hash rate.  I don't think Ethash would be profitable enough to be worth it.  Those numbers are just projections though, and their profits are in the same range as the initial algorithms being released, with X16R and Xevan looking the best at around $75/day
- X17 and X16R require two FPGA cards daisy chained together with 2 x 100G ethernet cables, one FPGA does half the function, the other FPGA does the other half
- Xevan requires FOUR FPGA cards daisy chained together with 6 x 100G ethernet cables; this is only possible with the Bittware card
- The QDRII+ SRAM modules for the Bittware board are very expensive: $2250 for one 76MB module, four modules max per board (288MB max).  This is the way to do Equihash but I have not yet been able to calculate if the price of the memory is worth it for the increased hash rate

Stay tuned for the end of the month launch and I hope that answers most of the important questions.

More answers

- Power: 100W to 160W per card depending on the algorithm.  So how much money can you make from one 15A electrical outlet (1500-1600W)?  One 8 x GTX1070 rig makes around $13/day using the full capacity of the outlet; the same outlet can power 8 x VCU1525 and make $400+ per day, so your profit vs. electricity goes up 20x to 30x

- I'm hoping to release the first software at the end of May; keep in mind this does require many extra steps vs. mining on my own; encrypting the bitstreams, device DNA protection, thermal monitoring & overload shutdown, troubleshooting guide & utility, as well as detailed step-by-step instructions and testing on several other computers other than my own

- I have already received inquiries from people who want to help.  There is a great deal of work to do to port all the available GPU algorithms into the FPGA.  One of the most tedious tasks is test-vector generation, which is the process of inserting printf() statements into CPUMiner while connected to a live mining pool to dump extremely detailed intermediate hash values to a file to use to verify the FPGA implementation.  The release date schedule can be sped up dramatically if more people choose to help.  Test vector generation doesn't require any FPGA knowledge, it just requires C-programming knowledge

- Price; you could always mine with one KU090 for $975, otherwise you can sell 8 GPU's from an existing rig and replace all of them with just one VCU1525/XUPP3R card (for around the same price) and your profit per day goes up dramatically and your electrical bill goes down 8x

At today's prices
- one 8 x 1080 Ti rig mining Ravencoin makes $21.87 per day, and burns 1520W
- Selling those 8 x 1080 Ti cards and replacing them with one VU9P FGPA card can make $50+ per day, burns 150W

I am using Windows 10.  Originally on my GPU rigs I ran linux, but first I had of lot of freeze-ups, second, all the latest (and fastest) miner builds were only released for Windows, so I upgraded all my GPU's to Windows.

For the FPGA's, the development tools and drivers are mostly geared towards Windows.  To run my software, it is expecting Windows 10 (but probably works on Windows 7), you need to install the FTDI D2XX drivers, and you need to install Vivado Lab Edition (free from Xilinx) to download the bitstreams into the FPGA's.

Some answers:

- The exact part number for VCU1525 is: DK-U1-VCU1525-A-G
(this is the version with a fan; it also comes in a fanless version which needs some type of external air movement system, not recommended; definitely go for the version with a fan)

- I realize even a single VU9P FPGA card might be out of range (pricewise) for some people.  The Avnet KU040 is $975:
https://www.avnet.com/shop/us/products/avnet-engineering-services/aes-ku040-db-g-3074457345630043740/
This FPGA is much smaller than the VU9P.  It can mine Keccak at 3GH/s, SHA-224 is profitable as well, and I am trying to fit Tribus into it which would run at 600MH/s and be quite profitable, but it is a tight squeeze to fit Tribus in there and I can't promise it yet.

- The smallest/cheapest board is the Nexys Video
https://store.digilentinc.com/nexys-video-artix-7-fpga-trainer-board-for-multimedia-applications/
(also available from Digikey and Avnet)
This board has an even smaller FPGA and can only do Keccak and SHA-224 (Bismuth), with an ROI of around 200 days.  So it isn't the most profitable, but it does have the amazing advantage that it needs no fan only can get by with a passive heat sink. 

- Some people have asked me about future boards.  There is the upcoming VCU1526 (VU13P FPGA), XUPP3R-VU13P, as well as the VU35P/VU37P FPGA's with high bandwidth memory (most useful for ethereum).  I do have some plans on supporting upcoming boards, it does require changes to the software, but for the near term the only four boards I am supporting are the VCU1525, XUPP3R-VU9P, KU040 and Nexys Video.  The reason I am focusing on those boards is they have the best ROI.  The absolute biggest FPGA available (VU13P) used in some of the 'upcoming' boards is so much more expensive that the ROI is actually worse.  It's kind of like the Titan V GPU; faster than a 1080 Ti, but the extra price isn't worth it.

My plan is to release Keccak at the end of May, then Tribus/Phi/Skunkhash in June/July.  More algorithms to follow every month.  So yes you can buy the hardware then use my software, which has a 4% fee built in.

Yes, very simple, you just need to make sure you have enough USB ports on the motherboard, otherwise you can use a PCIe-USB adapter card.  Depends on how many FPGA cards you want to run.

If you mean 'fork' to a different algorithm, I can change the software in the FPGA just as fast as they can fork their coin.  You can't fork away from FPGA's.