Topic: DIY FPGA Mining rig for any algorithm with fast ROI - page 47. (Read 99494 times)

As described at a first pages, The easiest and planning investment is to buy some GPUs for this money.

No disagreement on that, given that the power costs eat about 27% of mining revenue.
ROI time, however, is the only metric even remotely worth considering. Improving power efficiency gets you from 73% net profit (based on 1080Ti mining ETH) to about 93% net profit (based on this supposed 5x efficiency improvement you mention).
The question them becomes, how long does that 20% profitability improvement take to cover the 500% increase in hardware cost. And that figure is starting to look like it is measured in years rather than months.

This is awesome.  The fewer people that get involved, the better.  I hope the FPGA community stays small, so profits don't get diluted.

To clear things up about the power & modifications I posted an update to the Downloads page:
http://zetheron.com/index.php/downloads/

Software will be provided that runs on all types of cards (modified and unmodified).  Keccak represents the worst case for heat and power consumption.  Looking at the chart of performance/profit, you'll see that a majority of the profits for FPGA mining lie in the Cryptonight family of algorithms.  These algorithms, being memory intensive, don't consume that much power or generate that much heat.  GPU_Hoarder has his CN7 (19.7-22KH/s = 10-12x Vega 64) algorithm running on an unmodified VCU1525.  There is always the option of waiting for the Bittware VU9P crypto card as well which can run power-hungry algorithms at full speed right out of the box.

If you are still curious about performance on the heat-intensive algorithms as it relates to the various modifications, the approximate numbers are:
- Out of the box: full performance on memory hard algorithms; 50-60% of speed on power intensive algorithms
- Remove the outer casing and blast a powerful fan (takes 2 minutes to do): 70% of speed on power intensive algorithms
- (+) drop core voltage (no mechanical mod needed) w/powerful fan: 80% of speed on power intensive algorithms
- Advanced/crazy mods: 100% of speed on power intensive algorithms

This only applies to the existing VCU1525.  The huge group order being organized by GPU_Hoarder & Senseless already has some of the mods done before you even receive your card, and as I mentioned, the Bittware card (though it costs more) has all the mods already done for you.

Still, despite this information, hopefully people don't buy; the fewer people that get involved, the better.

The proper use case for this FPGA VCU1525 board is for those who can keep their designs limited to 150 watts or less for 24/7 operation. As it relates to mining out of the box, one can expect a 5-6x throughput advantage over a GTX 1080 Ti on average, of course for 5-6x the price.

As it applies to me, I am always thinking about space and power, therefore it is a Big Win in favor of this FPGA board.

I do not care for the misleading hash rates on the original posting, but I commend @whitefire990 for job well done.

So, in brief, 53 pages of hope, sweat and tribal dancing is ended like ROI=never.

From no mention of modding in the first post to this is crazy.

Ya, reduction in core clock is not the best way to handle power usage. Reduction in luts would be a greater decrease in power consumption than making an equivalent drop in frequency.

153A (130W) at 354mhz and that would provide 8.5Gh/s
means 130/0.9/0.9=160W on the AC line
AC->DC12V 90% efficiency
and 12V->0.85V 90% efficiency

8.5G /160W=53.3M/W
This efficiency is not an advantage.
there was  multichip kintex7 board got 6.4G/120W in 2014

@GPUHoarder, can you share more info on 22KH/s CN7?

Ya, the same can be said for a CPU or GPU. But if you need liquid nitrogen to keep it cool, it's probably not sustainable either.

One thing you should know is that FPGAs will anneal at high power.
let's see you burn the board

17GH @ 306A would imply ~8-9GH @ 160A.
That doubles the ROI time.

It depends on the design, whitefire's designs are far more optimized than mine for those specific algo's. It should scale linearly with frequency. So if it's using 306A at 708Mhz, it should use 153A (130W) at 354mhz and that would provide 8.5Gh/s. It is a good idea to provide A/C for these cards. The colder the card the faster it can run.

@whitefire990 Could you reduce core clock to a more reasonable number and provide some feedback? I'm not sure everyone is going to want to run their card at those junction temps. Keep in mind, even with cooling modifications, At those thermal envelopes, someone running their card in an environment 5F warmer than you, even with the same cooling, may have problems.

Right, so what is the expected performance like at 225W power limit? Presumably about 1/3, so about 6GH/s?

He's drawing almost double the rated power. That's why it shuts down. It would be like using 600 watts overclocking a gpu.

It's 1x 8-pin + PCIe (225W total). The big problem is that the core is only rated for 160A (136W -- But he's shown that it definitely can handle more).

The bigger issue is cooling which is not so difficult.

That right there is a complete show stopper, IMO, and a very strong indication that this particular product is unfit for purpose. This kind of an issue on a product with a price tag in the $4K range is unforgivable.

GPU manufacturers can keep a 2-slot card going flat out at 300W with their active cooling solutions. If VCU1525 cannot do the same with 150W power limit (1x 6-pin + PCIe at max 75W each), that indicates a very poor design. I am very glad I waited for initial feedback before I ordered mine, and in light of this I won't be ordering these at all.

Has anybody got a line on pricing and availability of the mentioned Bittware VU9P card that will be fit for purpose out of the box?

Hi

ok, million dollar question, what is the best coin to mine on FPGA based on current set of variables > 8 x FPGA cards + all Hardware mods + best available bitstream.

regards

Just because you CAN mine it, does not mean you SHOULD mine it.

FPGA's need specialized software specific written for a certain algo, and if an algo is ASIC dominated, and the FPGA would not beat the current ASIC speed/power ratio, it is not sensible to create a specialized software when noone will use it.

On some algo's, the battle has been lost already. Take bitcoin for example, a FPGA would not beat out current Asics. You can mine bitcoin on your GPU, but you don't right?

Hi Lazyriver

i thought that was part of what FPGA represented, flexibility to mine whatever you wanted even the coins on asics ? unless do you know that FPGA are completely infeasible on FPGA ?

Decred has Asics. No reason to release FPGA support.