Ergo FPGA PoC - ~3MH/W

[Wolf9466]

Some of you may remember me from my old account, Wolf0 (which I no longer have access to, as I got rid of the email I used long ago.) I'm still developing miners. Recently, I've been working on a miner for a coin called Ergo. Ergo uses a new proof of work known as Autolykos2 ( described here: https://docs.ergoplatform.com/ErgoPow.pdf ), which requires a large dataset in memory to mine. This dataset increases in size every 51,200 blocks, with a 2 minute block time. Currently mined with GPUs, and widely believed to be FPGA and ASIC resistant, this is the first FPGA miner (that I know of). While I have no clue what I'll be doing with it yet (releasing or otherwise), it performs slightly better than a 3070 per FPGA, at less than half the wattage.

My miner is still in alpha stage - just a proof-of-concept, but I am currently mining it with a JC35 (Jungle Cat 35P) FPGA board which features two XCVU35P FPGAs, each with 8GB of HBM2 memory. I am also mining it with an Osprey ECU50 (aka U50C).

I have exceeded 3 MH/s per watt - screenshot is here (NSFW):  https://i.imgur.com/FsXfZiR.jpg

Breakdown for those who don't feel like zooming in with the screenshot:
FPGA0: 56.14W @ 172.0MH/s
FPGA1: 55.54W @ 172.0MH/s
FPGA2*: 68.96W @  169.2MH/s

*This FPGA is an Osprey ECU50 (aka U50C) which I cannot lower the voltage on.

For FPGA0: Rounding the watts up to 57W: (172 / 57 == ~3.018 MH/s/W)
For FPGA1: Rounding the watts up to 56W : (172 / 56 == ~3.071 MH/s/W)
For FPGA2: Rounding the watts up to 69W: (169.2 / 69 == ~2.452MH/s/W)

I still need to make more progress with the efficiency of the ECU50 (aka U50C) FPGA, but this may take me some time to get to, as the JC35 FPGAs (FPGA0 & FPGA1 are a JC35) are my primary target.

Core clocks are 400Mhz for FPGA0 & FPGA1, and 425Mhz for FPGA2. Memory clocks for all three FPGAs are 1100Mhz. These can go lower or higher to gain efficiency and/or hashrate.

[alucard20724]

As usual, I have a screenshot (and as usual, it is NSFW, so I did not inline the image): https://i.imgur.com/aIyhu2b.png

I feel like i just got trolled by a fury lover ... NSFW..

[GKumaran]

As usual, I have a screenshot (and as usual, it is NSFW, so I did not inline the image): https://i.imgur.com/aIyhu2b.png

I feel like i just got trolled by a fury lover ... NSFW..

Had it not been fury, I would have questioned the legitimacy of the author 

[deadsix]

As usual, I have a screenshot (and as usual, it is NSFW, so I did not inline the image): https://i.imgur.com/aIyhu2b.png

I feel like i just got trolled by a fury lover ... NSFW..

Yup that's wolfie alright.

[arielbit]

Funny you comeback with FPGAs and not GPUs

[miner29]

Hes been working on FPGAs for a while.  Good to see results of his work. 

[fmz89]

Some of you may remember me from my old account, Wolf0 (which I no longer have access to, as I got rid of the email I used long ago.) I'm still developing miners. Recently, I've been working on a miner for a coin called Ergo. Ergo uses a new proof of work known as Autolykos2 ( described here: https://docs.ergoplatform.com/ErgoPow.pdf ), which requires a large dataset in memory to mine. This dataset increases in size every 51,200 blocks, with a 2 minute block time. Currently mined with GPUs, and widely believed to be FPGA and ASIC resistant, this is the first FPGA miner (that I know of). While I have no clue what I'll be doing with it yet (releasing or otherwise), it performs slightly better than a 3070 per FPGA, at nearly half the wattage.

My miner is still in alpha stage - just a proof-of-concept, but I am currently mining it with a JC35 (Jungle Cat 35P) FPGA board which features two XCVU35P FPGAs, each with 8GB of HBM2 memory. Performance (currently, it will get better) is 178MH/s or so. Power is around 65W, at 0.75V core and 1.15V memory. Clocks are 415Mhz (core) and 1100Mhz (memory). These can go lower or higher to gain efficiency or hashrate.

As usual, I have a screenshot (and as usual, it is NSFW, so I did not inline the image): https://i.imgur.com/aIyhu2b.png

Oh. Really you're legendary wolf0, where is the account now?, Cannot find that account in here anymore are you get banned?, But weird is wolf0 i do remember when first time i"m mining he didn't post like this and really different personalities, good to hear if you are real wolf0

[swogerino]

Impressed by the low power at only 65 Watt at 178 Mhsh or so.I am pretty sure that most of the miners here will be more than willing to see that miner go to 200-300 watt and see 4x or 5x the hash rate it currently has.That can be the miner all people living in countries like Germany and alike are waiting for because they pay really high electricity cost.I also hope you have success in your work and hopefully we will see more FPGA with better 16 GB HBM2 memory in not the so distant future.

[miner29]

Impressed by the low power at only 65 Watt at 178 Mhsh or so.I am pretty sure that most of the miners here will be more than willing to see that miner go to 200-300 watt and see 4x or 5x the hash rate it currently has.That can be the miner all people living in countries like Germany and alike are waiting for because they pay really high electricity cost.I also hope you have success in your work and hopefully we will see more FPGA with better 16 GB HBM2 memory in not the so distant future.

The JC board will not take that much power.  You can get a little more from water cooling but 200-300 watts will most likely end in tears or fire!

[Metroid]

I guess this is the end for ergo gpu mining as well hehe, less coin for gpus and more coins for asics and fpga's ehhe

[miner29]

I guess this is the end for ergo gpu mining as well hehe, less coin for gpus and more coins for asics and fpga's ehhe

Except those boards are very expensive and not available in large numbers.
The company that produced them is bankrupt. 

[Metroid]

I guess this is the end for ergo gpu mining as well hehe, less coin for gpus and more coins for asics and fpga's ehhe

Except those boards are very expensive and not available in large numbers.
The company that produced them is bankrupt. 

My point is if fpgas can double, triple its hashrate then an asic will be there in few months. Forget autolikos2 on gpus, well they can always fork like other coins have done.

[mo35]

will your bits be available for FK33 also , if/when its out ? good to see that there is some development for sqrl stuff, since seems like they are gone permanently in all shapes and forms.

[miner29]

I guess this is the end for ergo gpu mining as well hehe, less coin for gpus and more coins for asics and fpga's ehhe

Except those boards are very expensive and not available in large numbers.
The company that produced them is bankrupt. 

My point is if fpgas can double, triple its hashrate then an asic will be there in few months. Forget autolikos2 on gpus, well they can always fork like other coins have done.

I think you need to go look at the backlog in semi conductor production before you say a few months to asic.   More like a year+ at this point.  Fpgas fork easily unlike asics.  But again it is not the threat you claim it is….there simply isnt  the magnitude of volume of fpgas and the cost is sufficient to keep it that way for now. 

[Metroid]

I think you need to go look at the backlog in semi conductor production before you say a few months to asic.   More like a year+ at this point.  Fpgas fork easily unlike asics.  But again it is not the threat you claim it is….there simply isnt  the magnitude of volume of fpgas and the cost is sufficient to keep it that way for now.  

They will only tell they are producing it publicly if they plan to sell or market it as to sell on a set date, many produced asics in the past and kept for themselves for years till they decided to sell because they had something else better to let in their warehouse. Bitmain is known to do that, anyway, those practices are common known, not sure why you asked me to check semiconductor backlog, they ask for discretion here, when lots of money is involved and they dont want people to know then nobody will know about it. I said few months probably there are prototypes already.

[dedizonesv2]

anyway, once ETH is underground, the +800 TH s of hashrate on ETH will have to go to another currency.

ERgo or ETC ?

In addition, more profitable Ergo but requires the GPU frequencies to be adjusted
ETC, there has just changed algo in the miner, not even need to review the settings thus saving time.

The Asics and FGPA will remain ...


(Firo, RVN, Cortex are energy gulfs)

[Wolf9466]

Impressed by the low power at only 65 Watt at 178 Mhsh or so.I am pretty sure that most of the miners here will be more than willing to see that miner go to 200-300 watt and see 4x or 5x the hash rate it currently has.That can be the miner all people living in countries like Germany and alike are waiting for because they pay really high electricity cost.I also hope you have success in your work and hopefully we will see more FPGA with better 16 GB HBM2 memory in not the so distant future.

The JC board will not take that much power.  You can get a little more from water cooling but 200-300 watts will most likely end in tears or fire!

That's not true, the JCC2L boards can take 450W (not per FPGA, total, so 225W per FPGA).

will your bits be available for FK33 also , if/when its out ? good to see that there is some development for sqrl stuff, since seems like they are gone permanently in all shapes and forms.

An FK33 is too small for this implementation.

I think you need to go look at the backlog in semi conductor production before you say a few months to asic.   More like a year+ at this point.  Fpgas fork easily unlike asics.  But again it is not the threat you claim it is….there simply isnt  the magnitude of volume of fpgas and the cost is sufficient to keep it that way for now. 

They will only tell they are producing it publicly if they plan to sell or market it as to sell on a set date, many produced asics in the past and kept for themselves for years till they decided to sell because they had something else better to let in their warehouse. Bitmain is known to do that, anyway, those practices are common known, not sure why you asked me to check semiconductor backlog, they ask for discretion here, when lots of money is involved and they dont want people to know then nobody will know about it. I said few months probably there are prototypes already.

This isn't always true. I mean, I'm announcing this one because I know of at least one other implementation, and people ought to know that it's possible.
Also, this implementation uses HBM2 memory, which would be *very* expensive to put on an ASIC. Not only do you need the HBM2 itself and an interposer, but you additionally need memory controller IP and such. It complicates a design massively.

[miner29]

Impressed by the low power at only 65 Watt at 178 Mhsh or so.I am pretty sure that most of the miners here will be more than willing to see that miner go to 200-300 watt and see 4x or 5x the hash rate it currently has.That can be the miner all people living in countries like Germany and alike are waiting for because they pay really high electricity cost.I also hope you have success in your work and hopefully we will see more FPGA with better 16 GB HBM2 memory in not the so distant future.

The JC board will not take that much power.  You can get a little more from water cooling but 200-300 watts will most likely end in tears or fire!

That's not true, the JCC2L boards can take 450W (not per FPGA, total, so 225W per FPGA).

I was speaking per fpga so yeah under the best conditions 25w more than i stated.
You know plenty of high end fpgas have been burned up by inability to keep cool in mining. 
And you know there are no where near as many of these FPGAs out there as even 30XX cards...magnitudes less.   

[Wolf9466]

I have exceeded 3MH/s per watt - so I rewrote the OP, and renamed the thread. Details and new screenshot are in the OP.