Will fund ASIC board for mining community. Need Hardware devs.

[eturnerx]

Don't you dare say FPGA in front of ArtforZ, he will scold you into oblivion.

I don't see him here

[1714787807]

1714787807

[1714787807]

1714787807

[AnnihilaT]

I have a rudimentary design right now. It is the circuit in gate array format that I designed. It is not a final design by any means, but it allows me to identify the key bottlenecks of the system, calculate gate propagation, maximum clock speed, etc. So what I have from this is; 8MHz clock, 1 hash per clock, and 100 pipelines. This could fit on low volume production run for 27$ / ASIC from TSMC. You then need to add the cost of shipping, putting it on a PCB board and add a controller, + design cost for the lithography. So basically, 800 Megahash/s, for 57$ or so (+ design costs spread across the production line). And that is without the optimization to the main bottleneck (a 7 input adder) that I am working on. So it is encouraging.

You can easily add 4-5 ASIC on each PCB board improve the Hash/$ ratio. So, with 5 ASIC per board, you would get 4GIGAHASH/s for 165$ (+ design cost). (the price of a 5850). Essentialy it makes gpu mining obselite.

This sounds great but my understanding is that those design costs can be quite significant.... 800k to 2 mil USD.  Is that about right or can it be done cheaper? 

Following this thread with interest!

[Bloody Bell]

does anyone have a rough estimate of cost right now? I though the cost to produce makes it a lot more expensive than GPU mining...

It depends on the technology used. There is a rather large one-time, non-recurring expense, and then some price/unit for the actual manufacturing. Of course the more you can pay upfront, the better technology and lower per unit cost you get later. I made some research yesterday (and I was just about to open such a topic ), but unfortunately the manufacturers are pretty secretive about their prices.

But it seems to me that the lower end starts somewhere around $100k one-time cost which gets you a structured asic. This will give higher performance than an average fpga, but far from a real asic, and also the unit cost won't be that low. (but much cheaper than an fpga) As far as I know, ArtForz is about to order such a design.

For a real ASIC however you will need to pay a lot more upfront. (a million bucks? couldn't find exact quotes anywhere), but you could absolutely own the network with dirt cheap ICs. The technology is the simplest possible: only digital gates, no analog stuff, you don't even need memory blocks, also no complicated IP is neccessary, everything can be done and tested on FPGAs. The thing could easily run at least 100-200 MHz, even with an older technology (you don't want to pay for a 45nm process), and probably a few full pipelines could fit in. That means several hundres of MHash/sec on each chip, possibly more. The chip can be in a very simple case, with low density pinouts, so it could be soldered very easily, even with DIY methods. But that would take a lot of time with for example a thousand chips   If you had the money to manufacture the IC, making the PCB is pocket change. And bamm, you have a lot of panels, each of them with around a hundred chips, each chips putting out 100s of MH/sec. There is a topic somewhere discussing whether the CIA could commit 50% of the full network power. Well, they certainly could.

There is a free fpga design out there already, that's a good starting point. What it lacks is a mean to distribute the work and collect the result, as it currently has some hack relying on the debug features of the fpga and its dev panel. But it wouldn't take really much time to come up with a full design that can be quickly converted to any format the manufacturer needs.

Also, if you want to be safe, you probably want a design that can be somewhat reconfigured, so if the mining project fails, you still have an IC that can be used for other tasks requiring fast SHA-256 computation.

[Findeton]

Search for the term FPGA - this has already been discussed at length in there. The FPGA people already have working implementations. The big impediment to ASIC implementation is the cost not the talent on these forums.

I think there as been a misunderstanding. If not, then I apologize. But, I do not intend to simply release the design, as an "open source" file. I intend to finance the construction of a batch of ASICs chip based on a design, and sell them at cost-to-produce. So that miners can use them.

So you would release  the design as open source and also sell them at almost the cost-to-produce?

[comboy]

I would just like to repeat here, that for the sake of safety of our money it would be really nice to build into protocol hashing based not only on sha256, but maybe 2-3 other algorithms. This makes bitcoins much much safer. It could start at block number 300k or something like that.

It would of course make ASICs quite useless (unless somebody produce them for this new version).

Somebody selling hardware dedicated just for mining is inevitable with bitcoin growth, so this is one of the last moments for such a big change (we don't want blockchain splits because of miners that invested a lot in ASICs later on). Such change could make bitcoin solid as a rock. Even broken sha256 would not be a threat.

[Rodyland]

Screw PCIe ASIC miners, what we really need (and will eventually get, if bitcoin doesn't tank) is a USB miner.  Put your wallet on the USB miner, plug it in to a computer and it starts hashing automatically. 

If I had the time, money or expertise I'd be working on something like this.  Unfortunately I have none of the above - I'm merely an ideas man.   

[itsagas]

Screw PCIe ASIC miners, what we really need (and will eventually get, if bitcoin doesn't tank) is a USB miner.  Put your wallet on the USB miner, plug it in to a computer and it starts hashing automatically. 

spend up to millions of dollars to fit one asic chip on a usb stick (and get limited hashing and zero scaling ability)  vs. build a PCIe card or array that can fit tons of chips and scale  hugely)?  

I would go with the latter.

[rasputin]

Screw PCIe ASIC miners, what we really need (and will eventually get, if bitcoin doesn't tank) is a USB miner.  Put your wallet on the USB miner, plug it in to a computer and it starts hashing automatically. 

spend up to millions of dollars to fit one asic chip on a usb stick (and get limited hashing and zero scaling ability)  vs. build a PCIe card or array that can fit tons of chips and scale  hugely)?  

I would go with the latter.

I've only seen ahtremblay talk about PCB, i.e. circuit boards. Frankly, fitting these on pc slot-in cards would be daft. External units connected by usb, or preferrably ethernet is the only way to go..

[Sukrim]

zero scaling ability

USB hubs vs. PCIe splitters... just look up the prices!

Ethernet or USB (2.0 or 3.0) would be the way to go for sure (with Ethernet, you could do PoE and USB also allows electricity transport...)

Anyways, as long as I just read about "I designed this and that and it does xxx MH/s" and noone releases even an FPGA implementation of that, I personally don't really trust it.
Doing all of Bitcoins math in just 1 clock cycle for example takes quite a lot of gates, hm?

[ahtremblay]

I apologize for locking this thread by accident. It is now unlocked.

[qed]

I have a rudimentary design right now. It is the circuit in gate array format that I designed. It is not a final design by any means, but it allows me to identify the key bottlenecks of the system, calculate gate propagation, maximum clock speed, etc. So what I have from this is; 8MHz clock, 1 hash per clock, and 100 pipelines. This could fit on low volume production run for 27$ / ASIC from TSMC. You then need to add the cost of shipping, putting it on a PCB board and add a controller, + design cost for the lithography. So basically, 800 Megahash/s, for 57$ or so (+ design costs spread across the production line). And that is without the optimization to the main bottleneck (a 7 input adder) that I am working on. So it is encouraging.

You can easily add 4-5 ASIC on each PCB board improve the Hash/$ ratio. So, with 5 ASIC per board, you would get 4GIGAHASH/s for 165$ (+ design cost). (the price of a 5850). Essentialy it makes gpu mining obselite.

Absolutely impossible.

[MoonShadow]

I have a rudimentary design right now. It is the circuit in gate array format that I designed. It is not a final design by any means, but it allows me to identify the key bottlenecks of the system, calculate gate propagation, maximum clock speed, etc. So what I have from this is; 8MHz clock, 1 hash per clock, and 100 pipelines. This could fit on low volume production run for 27$ / ASIC from TSMC. You then need to add the cost of shipping, putting it on a PCB board and add a controller, + design cost for the lithography. So basically, 800 Megahash/s, for 57$ or so (+ design costs spread across the production line). And that is without the optimization to the main bottleneck (a 7 input adder) that I am working on. So it is encouraging.

You can easily add 4-5 ASIC on each PCB board improve the Hash/$ ratio. So, with 5 ASIC per board, you would get 4GIGAHASH/s for 165$ (+ design cost). (the price of a 5850). Essentialy it makes gpu mining obselite.

Absolutely impossible.

Maybe, maybe not.  Those who wanted to mine on GPU's didn't think to highly of the claims of a single ATI 5830 getting 100 times the hash rate that a high end cpu could get either.  I'm going to reserve judgement until I see some physical hardware.

[ahtremblay]

I have a rudimentary design right now. It is the circuit in gate array format that I designed. It is not a final design by any means, but it allows me to identify the key bottlenecks of the system, calculate gate propagation, maximum clock speed, etc. So what I have from this is; 8MHz clock, 1 hash per clock, and 100 pipelines. This could fit on low volume production run for 27$ / ASIC from TSMC. You then need to add the cost of shipping, putting it on a PCB board and add a controller, + design cost for the lithography. So basically, 800 Megahash/s, for 57$ or so (+ design costs spread across the production line). And that is without the optimization to the main bottleneck (a 7 input adder) that I am working on. So it is encouraging.

You can easily add 4-5 ASIC on each PCB board improve the Hash/$ ratio. So, with 5 ASIC per board, you would get 4GIGAHASH/s for 165$ (+ design cost). (the price of a 5850). Essentialy it makes gpu mining obselite.

Absolutely impossible.

Hello, care to elaborate a little more?

[qed]

I have a rudimentary design right now. It is the circuit in gate array format that I designed. It is not a final design by any means, but it allows me to identify the key bottlenecks of the system, calculate gate propagation, maximum clock speed, etc. So what I have from this is; 8MHz clock, 1 hash per clock, and 100 pipelines. This could fit on low volume production run for 27$ / ASIC from TSMC. You then need to add the cost of shipping, putting it on a PCB board and add a controller, + design cost for the lithography. So basically, 800 Megahash/s, for 57$ or so (+ design costs spread across the production line). And that is without the optimization to the main bottleneck (a 7 input adder) that I am working on. So it is encouraging.

You can easily add 4-5 ASIC on each PCB board improve the Hash/$ ratio. So, with 5 ASIC per board, you would get 4GIGAHASH/s for 165$ (+ design cost). (the price of a 5850). Essentialy it makes gpu mining obselite.

Absolutely impossible.

Hello, care to elaborate a little more?

I have developed in Xilinx ISE and from my experience that's way off. How many gates do you have? To have 1 hash/clock it must be insanely big, and you have to multiply it by 1000 (as you said). Hoe did you guess the production cost?

Even without knowing anything about algorithms Bitcoins and other stuff, the fact is: Commercial solutions are slower and more expensive, how a person with no experience make something better, faster and cheaper?

[TheSeven]

I have a rudimentary design right now. It is the circuit in gate array format that I designed. It is not a final design by any means, but it allows me to identify the key bottlenecks of the system, calculate gate propagation, maximum clock speed, etc. So what I have from this is; 8MHz clock, 1 hash per clock, and 100 pipelines. This could fit on low volume production run for 27$ / ASIC from TSMC. You then need to add the cost of shipping, putting it on a PCB board and add a controller, + design cost for the lithography. So basically, 800 Megahash/s, for 57$ or so (+ design costs spread across the production line). And that is without the optimization to the main bottleneck (a 7 input adder) that I am working on. So it is encouraging.

You can easily add 4-5 ASIC on each PCB board improve the Hash/$ ratio. So, with 5 ASIC per board, you would get 4GIGAHASH/s for 165$ (+ design cost). (the price of a 5850). Essentialy it makes gpu mining obselite.

Absolutely impossible.

Hello, care to elaborate a little more?

I have developed in Xilinx ISE and from my experience that's way off. How many gates do you have? To have 1 hash/clock it must be insanely big, and you have to multiply it by 1000 (as you said). Hoe did you guess the production cost?

Even without knowing anything about algorithms Bitcoins and other stuff, the fact is: Commercial solutions are slower and more expensive, how a person with no experience make something better, faster and cheaper?

The performance numbers sound reasonable. I have no idea why he only gets 8MHz, but this sounds like it can be improved. For a full pipeline you need roughly 100K flipflops plus all the logic. 100 times that would be 10M flipflops plus logic, which would be a rather big ASIC, but it's certainly possibly.

I had a co-worker try an Altera HardCopy synthesis and that one ended up with 20 pipelines running at >200MHz each, yielding >4GH/s per chip.

I don't trust these cost estimates at all though. The ASIC price estimate might be sensible for >10000 volumes, but the board, host interface and assembly cost estimated at $30 seems to be impossible. The VRM alone will cost that much.

[ahtremblay]

I have a rudimentary design right now. It is the circuit in gate array format that I designed. It is not a final design by any means, but it allows me to identify the key bottlenecks of the system, calculate gate propagation, maximum clock speed, etc. So what I have from this is; 8MHz clock, 1 hash per clock, and 100 pipelines. This could fit on low volume production run for 27$ / ASIC from TSMC. You then need to add the cost of shipping, putting it on a PCB board and add a controller, + design cost for the lithography. So basically, 800 Megahash/s, for 57$ or so (+ design costs spread across the production line). And that is without the optimization to the main bottleneck (a 7 input adder) that I am working on. So it is encouraging.

You can easily add 4-5 ASIC on each PCB board improve the Hash/$ ratio. So, with 5 ASIC per board, you would get 4GIGAHASH/s for 165$ (+ design cost). (the price of a 5850). Essentialy it makes gpu mining obselite.

Absolutely impossible.

Hello, care to elaborate a little more?

I have developed in Xilinx ISE and from my experience that's way off. How many gates do you have? To have 1 hash/clock it must be insanely big, and you have to multiply it by 1000 (as you said). Hoe did you guess the production cost?

Even without knowing anything about algorithms Bitcoins and other stuff, the fact is: Commercial solutions are slower and more expensive, how a person with no experience make something better, faster and cheaper?

I never said I had not experience. In fact, I said I am a physicist with 1 year of masters in electrical engineering. I said I understand, transistor, gate logic and most if the solid state physics stuff related to electronics, since thats what I did for a few years at the university. I said, I had no experience design on a massive scale using current software and compiling the data to the TSMC format. Thats a lot different than no experience or no knowledge of the issue. The reason I know the price per units is because I have a quote from TSMC.

Anyways, I have hired a firm to help out with the design from start to finish. So things will go smoothly now.

[Dien5Daze]

Even without knowing anything about algorithms Bitcoins and other stuff, the fact is: Commercial solutions are slower and more expensive, how a person with no experience make something better, faster and cheaper?

I think you might be assuming too much about his level of experience and knowledge on the matter, lets not rush to conclusions.

I'll defiantly be watching this thread, but in a way it would be kind of a shame to see mining move away from the GPU and onto dedicated hardware, it would mean the end of the garage miner.  It seems like the people who would be most interested in buying these would be small companies and start ups, only the die-hard miners would buy hardware that could only be used for mining bitcoins, even if they were cheaper than GPUs.

But on the other hand, I wouldn't want to see this type of thing only be available to a select few either. I'm kind of torn, but honestly I would probably buy one.  

[padrino]

I never said I had not experience. In fact, I said I am a physicist with 1 year of masters in electrical engineering. I said I understand, transistor, gate logic and most if the solid state physics stuff related to electronics, since thats what I did for a few years at the university. I said, I had no experience design on a massive scale using current software and compiling the data to the TSMC format. Thats a lot different than no experience or no knowledge of the issue. The reason I know the price per units is because I have a quote from TSMC.

Anyways, I have hired a firm to help out with the design from start to finish. So things will go smoothly now.

Good choice given the nature of the thread thus far, I wish you the best with this and think you may yield some good results depending on the firm you hired.

[pokermon919]

bookmarked

[padrino]

Even without knowing anything about algorithms Bitcoins and other stuff, the fact is: Commercial solutions are slower and more expensive, how a person with no experience make something better, faster and cheaper?

Commercial bitcoin mining solutions? I didn't know there was a solution built specifically for bitcoin mining already instead of an adaptation of generally available commerical products built for differing reasons. Please enlighten us..

EDIT: I misread the initial quote however I'll leave my reply as is because the general point is the same, nothing has been purpose built that is generally available to the community, in that sense it could be done.

I have a background in hardware driven SHA-* hashing for different reasons and there is a fair amount of good research available online as well as good performance analysis of different approaches, should be an interesting project if the details are released to the community (vs. Art's close hold)