Topic: Idea for NEW Agilmine FPGA UltraMiner (Read 1128 times)

the commercial is not so fast. Today, for ultraminer i think the people not can pay more of 300 dollars...
and 500 for ultraminer plus... but is just my opinion

Better use the most powerfull version with RAM to get a security web for algo evolutions. And payment via bank card.

Yes we will be making all the algorithm bitstream files. in future we consider release some as open source to allow community contribution.
we are still evaluating the option of adding memory to the board. thanks. we hope to finalize the negotiation with supplier and launch the kickstarter campaign soon

ship to me in Canada, then I ship to you in US

-Assuming you have all the bitstreams and dev's making them?
-Kickstarter for sure, everyone gets an extra layer of security
-Consider adding ram, and mine with RAM intensive algo's, which then puts your product in it's own class.

Monoro sounds like an interesting project 

The price is reasonable if you compare it to the other options available, the only reason I might consider going with a cheaper fpga would be if you could produce it faster and in smaller batches. I would much rather have a FPGA that can do 12 mhs on Lyra2z @ 60w delivered in under a month than one that can do 25-35 mhs @ 60w-80w 4 months from now.

Thanks, so what do you think of our price point. for ultraminer, we can potentially go with lower price with cheaper fpga, but the performance would be lower as well.
However the $1099 is probably the lowest price we can make for Virtex based UltraminerPlus

If these existed now I would jump at the chance to grab a few, unfortunately the waiting game for these type of projects can extend way past what most people are comfortable with. I think the kickstarter idea might not be a bad way to go with this.

this sounds more interesting, than 4k

we have not finalized the algos list yet, and at the end of day fpga of this price and performance should outperforms gpu in terms of ROI, and takes much less power. there are already many secretly mining using custom design fpga boards, we want to start an open fpga project that everyone can have access to.

Hes not being pessimistic, just being honest.  No doubt FPGAs are highly capable, but not profitable with the algos you have presented.

Hi i think it might be 27% now?  we are based in US, definitely want to better serve US customers. we will look into where this product should be built to reduce cost on customer side.
it is possible to add a small piece of high speed ram on board for mining lite memory intensive algorithm more efficiently, but probably not those heavy memory ones like ethereum.. we are still evaluating the benefit/cost addtion.
thanks.

Will this be distributed in the usa so that way usa customers won't have to pay 25 percent extra trump tarriffs?

Also will there be memory add on to this device so we can mine memory hungry algos?

Im interested in the ultra miner plus.

Hi.

We do accomplish 25MH/s in simulation on Kintex based devices already. The VU3P has more FPGA logic so we expect that it should see improved hash rate, but we have not done simulation testing with VU3P yet, as our expectation was that people would be more interested in the cheaper UM miner (this might be a wrong assumption on our part).
I do not understand where your cealing of 25MH/s for Lyra2Z would come from? The basic hash done in Lyra2Z is just a chained 64bit blake2b  (8*8*9 times chained for Lyra2Z) and that is trivial to get really fast. Your limits will comes in how you design the memory routing for the 768 x 8 x 8 bits work matrixes needed.

Now our simulation assume that we have no issues with power delivery or cooling. And we are not assuming that the FPGAs will be in some power-saving low voltage mode, we are giving them all they can handle, and our PCB layout design will support this reliably. But that is also why we do not want to go bigger then VU3P, as a massive combined package like VU9P has much more issues with power delivery and heat dissipation.

While we have not done enough work to say much about any Cryptonight version yet. I do not agree that integer division and square root operations would be a FPGA blocker. Just a simple Radix2 divider would not be to big and can be heavily pipelined, and then there are the hardened DSP blocks "Configurable ASIC blocks if you will" in most FPGA's that can be used for thees purposes  as well.

Thanks for you inputs, but my friend i think you are overly pessimistic  
HW resource and capability of this fpga device is very powerful and can easily adapt to the future algorithm changes. besides there are always new coins popping up, programmer can develop new algo support on fpga board with decent hash power. and have i mentioned about power difference between gpu and fpga?
Again, i believe it would be really hard to become fpga resistant.

Lyra2rev2 is already ASIC'ed.
Tribus, Keccak, 0xbtc algos are dead if speaking in terms of profit
The last coin which is used Phi1612 algo is a small shitcoim called Folm with about $100/day trading volume on exchange, and even this small coin is going to fork with algo change
On Phi2 even old RX Polaris are provide almost two times cheaper hash cost. $44/Mh vs ~$25/Mh. nvidia Pascal - even bigger difference
Zcoin - the biggest Lyra2z coin is going to fork at Dec 10th, hashing algorithm is also will be changed. Also 35Mh it's very overestimated hash rate for this type fpga chip... real achievable ceiling is about 25mh/s
On cryptonight v8 fpgas can't  achieve any reasonable hashrate due to sqrt and div which was added into V8 cryptonight codebase. in terms of hash cost fpgas can't beat AMD Polaris and  nvidia Pascal GPUs.
No doubt that payback period of these fpgas will be much higher than 1.5 year. (very optimistically) and we are not taking in count huge amount of bcu1525 and chinese copycats which are hitting network these days

Owners of all this great mining equipment will be screwed with unreasonable roi period.
I want to repeat: nowadays mainly only miming hardware manufacturers and hardware resellers earn money in crypto mining  sphere.
This is a game where  miners will always lose
In near time we'll see more and more coins forks with asic, fpga resistant algos.

Ya most good coins are being mined by asics so it makes it tough,

 Monero, zcoin, dash, appear to be best quality choices.
 
Coins I like that asics struggle with.
Ethereum, Ethereum classic.

Other coins i like, that asics haven't taken over yet and have some future potential
Bitcoin Gold, Raven Coin,

Speculative coins
Feather coin neo scrypt. bitcoin interest,

Thanks, we proposed these ideas, trying to sense how big the demand is there for those types of fpga miners, because, we will need massive order (5k+) to show xilinx in order to get this aggressive price approved. we can't sell for this price if there is only a few hundreds ordered...

Ok. I’d say go kick-starter route. $1099 will be a whole new market with many GPU miners wanting to move up.

Most of the estimate is from xilinx simulation tool. VU9P is essentially 3x VU3P inside, interposer packaging. VU3P is much more cost effective of the line, and the high speed hyperlink can connect multiple together to achieve similar performance of larger single die packaged fpga

sorry we don't have enough data provided for the interconnect co-op mining case, the performance may vary based on the specific implementation, and haven't had enough time to run through all cases.

I see. Your hash projections, are they estimates or based on actual propotype? Beacuse they look quite close to BCU1525, which is VU9P based, and I presume much faster than VU3P.

Edit: Nvm, I mis-read, your “estimates” are lower. Make sense.

Thanks Philp for the good words! once we have the sample boards ready we definitely want you be able to help us to test it, and arrange the escrow in the future for people who may need that service.

we think maybe kickstarter would be a good idea, as the fund raised there will enable us to purchase the materials to build a small sample batch and send to community members like you who want to be early birds.

I totally understand your frustration about pre-ordering, that's the reason why A1 is yet to go for mass production because we couldn't get a firm commitment from chip supplier. we don't want to take any order that we can't deliver on time. for this new models, we have confirmation from both suppliers and chip manufacturers on the date, we will take extra miles to guarantee the delivery

you also raised a good point about the bitstream support, we are considering maybe we should do open source once the product is shipped. so community can update or even enhance the performance.

UMP is most flexible, it will be a 16nm VU3P. This gives most flexibility.

UltraMiner goal is to be best hash/$. So it will be Kintex based, we are still investigating best price to performance between 16nm and 20nm.

Thanks, what other good ones you are interested in seeing we are now collecting ideas/requirement.

But which one? There are many VU5P, VU9P etc...

It will be based on 16nm Xilinx Ultrascale

His first model worked for me.  I have it turned off at the moment since the coin I was mining had the pool closed down.

I would be interested in the new model but there are multiple issues here .
bitstreams are the god of this type of  gear.
So anyone getting in on these needs to understand you may or may not get good bitstream support.

I may order one from him but I do hate the pay upfront and wait game as much as anyone else.
These guys are smart guys that know how to build and program gear.

I suggest that some one with some coin could have them build a sample bunch of 10 or 20 units. Maybe for sales.

I am always willing to demo gear or hold escrow.  If asked. BUYt I am retired and not full of cash to order a lot of them for myself.

I just spent  July 13 to Oct 27 waiting on the Acorns I got a refund and they are still not yet shipped maybe next week for them.

I did like working with the builders of the gear. and maybe I get in on this gear some how.

This is a good idea for folks like me that have high electric cost and still be able to tinker without losing money.

Get a credit card payment channel working and more algos for less crappy coins.  Only good ones I see is monero and zcoin.

The price looks very good. Which exact FPGA is the UMP based upon?

It has a resonable price for the specs. Specially Lyra2z and Phi2 seem good.

thanks! with prior experience there is no technical barrier for us to design this board. we have also aligned with supplier and manufacturer, now if the community desire a miner like describe we would be thrilled to build one. we have seen so many secret fpga miners, we want to build one everyone can see. 

Would be interested in a normal Ultraminer, just to learn and play with a new toy 
Ill keep an eye on this topic to see if it comes to a real product.

at this point we can only take small quantity order for the A1 miner, please PM me directly if you are interested. the reason being the fpga chip manufacturer is going through merger with another company, which significantly slowed down their production timeline, and since we couldn't get a firm confirmation on delivery date from them, we think it is too risky to take in mass orders from customers not knowing product can be delivered in fixed time. that's also the reason why we have learned to go with the top fpga supplier this time, they can guarantee the delivery for high volume order. sorry for the confusion i know it is a frustrating process... to us as well who have spent so much time and effort to build it. 

Why does the store page remain "Coming Soon"?
https://www.agilmine.com/store/

HI All,

This is arm_race/rapidminer, jumphit and grapefruit from Agilmine. Some of you may know me from other threads, we are a group of engineers enthusiastic about building affordable & accessible fpga mining devices for bitcointalk community members. In August we built the world's very first custom designed FPGA altcoin miner "Agilmine A1", which can efficiently mine multiple algorithm at GPU level hashing performance with only 1/5 of GPU power consumption (link below).

https://bitcointalksearch.org/topic/agilmine-your-real-altcoin-fpga-miner-updated-816-4493669

And thanks to our beloved and well trusted member philipma1957 who helped to verify that it is a true product that exists, and can achieve the claimed performance. He has been happily mining it for months! see his video reviews below:

https://www.youtube.com/watch?v=Yle13b1kgWo
https://www.youtube.com/watch?time_continue=105&v=_bet2aRAhvw

Since then we have received tons of inquiries and comments from our forum members, starting to realize that the HW capability of Agilmine A1 FPGA may not meet the requirement of our demanding customers , we started looking for another fpga chip that can yield higher performance, yet at a cost effective price, resulting in shorter ROI, but at the same time not compromising the easy-of-use feature of A1 miner. We have conducted collaboration effort with major fpga manufacturers, and contacted fpga distributors worldwide to search for an optimal solution. What is proposed here is what we believe that would work after months of work. But to make it happen, we need your feedback and suggestion, and hopefully your support behind it eventually.

Agilmine UltraMiner (UM): Price $750 USD
Standalone FPGA miner designed with Xilinx UltraScale+ Kintex FPGA chip, which is the Best Price/Performance/Watt Balanced 16nm FPGA device. The design concept is similar to the Agilmine A1 miner, expect 4x performance gain at the cost of 2x power consumption. the estimated hash rate performance see the table below.

Agilmine UltraPlus Miner(UPM): Price $1099 USD
Standalone FPGA miner designed with Xilinx UltraScale+ Virtex, which is the Highest Performance 16nm FPGA device in the world. Expect 40%-60% performance gain compared to UltraMiner. Plus, it is equipped with proprietary Hyperlink IO connector, allowing 2 or more miners connected together to mine in "CO-OP" mode, allowing more efficient implementation for mining complex algorithms like PHI2, Lyra2REV2, TimeTravel10 and Xevan. In general consider additional 25% performance gain compared to mining in standalone mode. Plus, the hw resource should allow mining CryptoNightV8. Many work still in progress, we will release more data later once available.

* No dev fee on all algo


Moving forward, we will proceed with this product development if there is enough interest. It is very aggressive low price we can get it from the fpga supplier only when the volume is high. Now I understand there is certain trust issue with regard to the fpga miner, so we are considering launching a kickstarter campaign to raise fund for production. The benefit of using kickstarter is that if we fail to deliver in the last, you will have your money back, guaranteed by the platform agreements.

Please let us know what you think about it, is it something you are interested in or you think it is a bad idea and please tell us why?

We are available on discord channel, or you can shoot me email always. Again transparency and openness is the key, we intend to build a community owned project!
https://discord.gg/R36gzXM
[email protected]

You can also join our mailist by clicking the link below we will send periodic project update to you!
http://eepurl.com/dMfjhQ

P.S. for A1 miner, we are thinking of building a small batch if you are interested in buying.