Those coin devs should be more concerned about someone creating a secret 90nm asic for $500,000 and 51% attacking their coin.

If the algo change is not very big it will take less than a day to modify for a fork. For instance, zcoin, they forked away by changing their lyra2 parameters. We would only need to change 4 lines of code to adapt. The longest part of making the change would be waiting for the bitstream to compile. But, they increased the memory usage beyond what is available on the chip - so we won't be able to mine them again until we get our HBM2 memory. But, we will, and we will once again achieve perf above and beyond GPUs. Then what are they going to do? There's no change they could make at that point where we couldn't correct for it in the same day they made the change.

So, now that I know zcoin will try to pow change if they find out I'm mining on it, why would I ever be public about it? Maybe I should go spend $500,000 secretively and be 99% of their hashrate? (I wouldn't do this. But, Bitmain and others would.)

Ya, ya, ya, I'm well aware of the things I've said... (Assuming you're saying that with the LOL because I've said exactly that myself )

-- The greed inside...

Because the only way I can get my own money printing machines is to buy enough of them from Xilinx. I can't do that alone. Collective bargaining. Right now we are one cog in a multi-cog collective bargaining operation going on with Xilinx. All of these cogs are competitors. We have all been doing the same thing in secret for some time. But, we all also have the same problem, Xilinx is slow and difficult to deal with. None of these cogs (myself included) would be able to achieve a successful PO on their own. The number of chips we would receive for sales (in batch 1) is less than 1/2 (about 40%) the total amount. Our total order volume right now is approaching $50M USD. Future batches would mostly be made of community orders (80-90%). But there are a few large players driving this and I am not one -- I'm closer to the community gpu miners than I am these guys with their own private multi-MW hydro electric dams and $20M operations. So, ya, I could let them do their thing -- they'd pay higher pricing... But would EVENTUALLY take all the profit out of it for my small-ish volumes, they'd take the gpu profits, they'd cause numerous coin forks as a result of their secret hashrates (which could follow and sustain through a fork -- causing various issues in the community). The only way I can see to continue for myself and compete is to open things to the community.

The only way anyone else in the community will buy the money printing machines is if they can print money with it. This means I have to release firmwares and have a method for people from the community who develop better designs than our own to release theirs. This also benefits us as if their design is better than ours, even with the devfee, why wouldn't we want to run it? Our margin on these is ridiculously small almost too small.

I truely believe that the community engineers (people you've not even seen talk in either fpga thread yet) will release bitstreams that have better performance than anything myself or whitefire could release. It's in my interest to provide them a platform to do their magic and provide them with a method to benefit from that.

If you want to invest $30M-$50M in me (at a $200M valuation), let me know, I'll go back to doing it from the shadows! -- But as I said, I'm pretty sure staying in the shadows would end up decreasing profitability (due to issues, forks, community backlash, etc).

-- The egalitarian inside...

Right now, there are 130, 90, 65, 45, etc nm asics that no one knows about mining various cryptocurrencies. These asics were created by individuals in secret at low cost with efficiencies far beyond GPUs. They do limited runs to reduce their risk using MPW projects, etc. These people, if they wanted to, could effectively 51% attack these currencies and may have -- there has been an increase in the number of 51% attacks on various coins. There has been an increase in the number of double spends being attempted (successfully and unsuccessfully) on exchanges. We need general purpose devices with efficiencies greater than GPUs. Devices that have power / performance and cost / performance ratios closer to ASICs than to GPUs. Having these devices on the PoW networks would secure these networks against 130, 90, 65 etc nm asics. Even 45nm asics may only be slightly better performing than the FPGA devices. This pushes the cost of secret asics down to 28nm which is still millions in NRE and increases investor risk at the same time. A 28nm asic would have a higher mask failure rate than a 130nm asic. The lower you go the smaller the spacings and the greater probability an error will creep into the design.

I'd eventually like to boot every GPU off of every PoW network and to have all the PoW networks without a 28nm asic or better to be secured by a decentralized network of highly efficient FPGAs. The miners don't even need to replace their rigs. They can continue using their gpus for now and swap them (the gpu cards) out over time slowly selling them off on ebay so as not to overload the secondary markets. There's also a GPU co-processor solution that GPUHoarder may be releasing to market. It uses a very cheap and low power FPGA to do some co-processing work for the GPUs. This device can add a little life to your GPU clusters increasing their hashrates and reducing power consumption.

We're working on it. It's slow going but i'm cautiously optimistic about what the next couple of weeks worth of meetings would hold for myself and others in this thread.  

Also, my project is unrelated to whitefire's other than we've been doing the same things independently. We're not the only 2, there are a few in this thread that have all been doing the same thing in secret for quite awhile now. His performance is much better than mine (and some of the others) for these logic based algos. I haven't seen him mention any numbers for memory algos, but I believe my numbers may end up higher than his there. This is why I'm in favor of a community shell and community bitstreams. There may be others in the community who can achieve performance even better than myself and whitefire! Let's see what the community engineers can do


Not necessarily. #1) Really short < 1m distance. #2) We're not required to follow any sort of network level protocol. One of the engineers I'm working with was operating 10Gbps over distance using 2 pins from the FPGA.

I had answers to these questions but a few things have changed. I'll answer the questions I feel comfortable answering at this point.

3 ) USA
4 ) Original price we were going to launch at is $3995. This will change over the next couple of weeks.
5 ) 1 unit minimum order
6 ) Crypto, Credit Card, Paypal (CC/Paypal only with manual review / authorization), and Bank wire
7 ) Yes -- first batch -- no problem
8 ) No, it means that the only encrypted bitstreams you can use that make use of the efuse key can be ours. Once our key is burned into the device it's permanent and irreversible. You can still use the devices for any unencrypted designs... Or designs that make use of other keys (BBRAM). We will be allowing any community developer who wishes to design secure code for use on these devices through our encrypted platform. They will be able to publish their bitstreams and the bitstreams will be available for anyone who's using our boards to download and operate via our software. I should be really clear here, we will be colllecting a dev fee on behalf of these devs who release code. The dev fee will be set by the dev. We will take a small portion of their set fee for maintenance / operation of the software and shell platform.
9 ) The only other use prevented is loading secure / encrypted bitstreams. This limitation applies anytime you burn the efuse. If you wanted to use some other secure / encrypted bitstream, that provider would require you to ship your device to them or buy a device from them that is loaded with their key in the efuse. After they do that, you would again be limited to only using encrypted bitstreams provided by them.

.. I'll also add, our margin on these will be fixed. We're not planning a "how much you got?" sales method like baikal, bitmine, and others who regularly shift their pricing according to profitability. My goal right now is to get the cost / performance ratio to be better than a gpu. I'm not saying that we'll be able to sell them for as low of a price as a GPU -- but that when you divide cost by hashrate you'll get a number that is better than that of a gpu. For example, if I purchased a GPU to mine lyra2z it would cost about $400 per Mh/s. If I purchased a FPGA to mine lyra2z it would cost about $115 per Mh/s. We would like to get all of our algo numbers to be far more cost effective on fpga than gpu. On top of this, we're expecting an even greater power to hashrate efficiency. A gpu may use 100-200 watts per 1Mh/s of lyra2z. The fpga would use 6 watts per 1Mh/s of lyra2z.

These devices could be used to provide some ASIC resistance against 'secret' asics. Once we start releasing efficiency numbers you'll start to get an idea that any ASIC developed at greater than 28nm will be at best be on par efficiency wise as the FPGA. There are numerous groups that are making 130-65nm asics in secret and mining on them. I would not be surprised to learn that a group has already redesigned their high level cryptonight asic and are preparing to tape out for CN7. We need to close the efficiency gap between high level ASICs and general purpose hardware that everyone is using for mining.

Update: Going to place sales / pre-orders on hold following meetings this week with Xilinx and Intel.

Who's "we"? You mean "Bitmain"?

Edit: Apologies, I went back and saw your original post on May 8.


Yes, their 10nm FPGA are terrible... I'm starting to wonder if it's a vaporware product...  

All jokes aside, I don't think your friend and us are talking about the same thing. If the situation at Intel is as bad as he believes, comparatively, the situation at Xilinx has to be a lot worse.

I would guess this reason is more than likely why Xilinx is hesitant to let anyone else in the crypto community have their chips. Bitmain is already "partnered" with Xilinx.

As far as I'm aware they are at least in some limited fashion shipping the stratix 10 mx series. I'll confirm next week for sure. But I do remember hearing about problems they had in 2017 and how they wouldn't be resolved until 2018 because they had to redesign the stratix 10. I believe what we are now seeing is that redesigned chip. As far as I know, altera had never planned 58G serdes in the stratix 10 -- unless I missed it in the literature of the time.

I think our definitions of flood the market differ. You're thinking about it from a consumer level -- It will eventually be that way -- But not in the next 12 months. If intel had the same practices as Xilinx, the Xeon 4116 system I just built would have cost me $100,000... Not $6,000... And I would have had to design my own motherboard, because they wouldn't have allowed me to produce their reference design... I don't think it's unreasonable to expect a $500-1000 unit cost on a large Stratix or Ultrascale+ part with HBM2 once production is moving 100K units a month. If Xilinx has their way, that part would cost $5,000-10,000 at "volume". If Intel increases their market share by expanding the market, and making these chips available at the enterprise level for hardware acceleration of databases, web servers, etc.... It's not unreasonable to think they could move 100K units a month at a $500-$1000 cost.. Amazon is currently leasing Xilinx fpgas. Do you think AWS would be interested in including FPGA logic in every single one of their servers? What about tencent? What about Baidu? What about every other provider on the planet? There's a storm on the horizon and it's going to be beautiful to watch.

The value of these devices in the server markets is in the 100s of billions. Far beyond their value in crypto markets, defense, etc, IMO. Xilinx has been very short sighted in their attempt to make profit. The world needs low cost FPGA. Not having it is preventing major advancements in computing.

Unfortunately that's human nature. The same can be said for any system of proofing or economics (even socialism). There is a new netflix show called "Explained". The third episode on Monogamy had some interesting things to say about tribal nature and raising children. It really got me thinking about society as a whole, why we do the things we do, and how ridiculous it all is. We're all in this together. If the military budgets globally were diverted into science / research... Most of the ails of today would be no more... Unfortunately, empowering people also means losing power for those who currently hold it... It reminds me of that line from cloudatlas... "There is a natural order to this world, and those who try to upend it do not fare well.". (Oh reminds me of this too..)

Even if we dropped the price to $10 per HBM2 Stratix 10 part.. It wouldn't stop people with more money from buying more than the rest. Even if we limited sales it would just create a secondary resale market that those with more money would take advantage of.

I've been warning the same -- I've been very vocal against a lot of these ASIC companies and would refuse to do business with most all of them.

Yes, the Stratix 10 parts are equivalent to the Ultrascale+ parts. I believe we'll be able to get the same or better performance for a much lower price point. If FPGA are really going to be viable in the long term to displace GPUs the cost has to be reduced in a significant way. $3995 is ok, right now, while margins are high... But if we really wanted to displace GPUs completely both the cost / perf and power / perf will need to be better than GPUs. I'm hoping that we'll eventually (6-12 months) be able to sell a $1000-2000 board with a stratix 10 or ultrascale+ part that has HBM.

Not 100% accurate, but a good read. Thanks.


I think you reversed who the trolls were in the sia dev situation. They were trolling the community. All of their "ASICS WILL PROTECT THE COMMUNITY". Then "OMG! BITMAIN LET'S CHANGE POW!".

It depends what the algo looks like. There are already tensor bitstreams available for lease on amazon from third party companies.

The export control is no worse than the export control for a lot of other multi-purpose hardware. There's an ITAR code for your CPU, I'm sure even GPUs have one. A license per country you wish to ship to (Canada excluded). It's really not a big deal at all, and, doing production outside the USA with a non-usa corporation you can ship to wherever you want (according to whatever laws are applicable to the entity in the territory that it's registered, operating, and shipping from).

As far as I've seen, the only person who's really publicly announced or mentioned performance has been whitefire990. I was more than happy keeping the whole operation a secret and continue mining in secret. If you don't want to buy one. DONT.

Ya, Once I realized that my VCU1525 thread was #2 on google for vcu1525... I figured they'd be aware of it... I'm hoping that whoever is blocking the sales at Xilinx will wisen up and open the flood gates.. If they don't want my money, I'm more than happy to give it to intel. $50,000 for a proto chip 

Secret info? No, All you need to do is look at the information they're publishing, what they're developing, etc to determine what their goal is. As I've said before, intel could lose market cap (stock valuation) equivalent to 100% of Xilinx's stock and it would just be a "bad day". Xilinx and Intel are not in the same class. Intel owns their own fabs. Xilinx pays for fab. Intel can run as many tape outs as quickly as they want with no delay. Xilinx has to wait for TSMC scheduling. Intel can produce chips and sell them at a profit for below what Xilinx's cost to produce is. If they did get into a pricing war, Xilinx would be destroyed. Not only will they lose on cost, but they're also going to lose on performance because.... Well, vivado isn't great. Even ISE was better than vivado.

The Stratix 10 will also have HBM2 and they also have a dual port memory technology equivalent of the ultraram.

I just did. The first thing any sales rep is going to do is tell you how wonderful of a company they are and how many huge businesses in your sector they're doing business with. All you need to do is ask some leading questions the right way. They'll happily provide you information that you can use a little bit of supposition with or in some instances, just directly tell you information they shouldn't. In the case of amazon, a xilinx authorized distributor sales' agent directly told me what amazon paid (without any leading questions).

There's only so many people in this business. Sales and more specifically, the authorized sales reps, like to brag.

How do you think bitmain knows what their competitors are doing? How do you think I know what amazon paid Xilinx, or about recent purchases Bitmain made?

I'm pretty sure intel's plan is to flood the market. Why else would they be developing hybrid cpu/fpgas (like APUs) or why would they bother to create the CCIX interconnect? They're definitely going to flood the market. The question is when. Xilinx will almost surely wait and only react hoping that Intel will also try to preserve high margins. The way I see it, Xilinx has one last chance to gain market share before Intel opens the flood gates. Get your options placed on XLNX while you still can -- Looking forward to making money on their downfall.

Plus, as you and I both know, Quartus is capable of placement / routing that is orders of magnitude better than anything Vivado could do automatically without floorplanning.

I'm looking forward to my call with Intel next week! Hopefully I'll be able to get something rolling quickly and at a lower price point so we (crypto community as a whole) can avoid Xilinx all together.

Edit:

Oh ya, I find it hilarious that devs are changing their algos based on what we say in this thread. Which reminds me, I just developed this new code for cryptonight which makes use of a little hack that's able to bypass some steps to obtain a result quicker. I'm now mining monero at 100Kh/s per fpga.

I've got rolled and unrolled versions of all the x11 algos plus some others (lyra2, 256bit versions of some of the x11 algos, etc) -- in vhdl.