Topic: DIY FPGA Mining rig for any algorithm with fast ROI - page 73. (Read 99472 times)

There are some situations where having 2 cards could be beneficial and provide superior hashrates. An example, cubehash on lyra2rev2. The hard part about lyra2rev2 has nothing to do with memory. And, if anyone out there wants to make a FPGA resistant algo, better use a lot of cubehash. It would restrict fpga hashrates, at least temporarily. If lut density doubles or triples again, it would be no problem.

Hmmm, two accelerator cards will be daisy chained with pipelining and their performance will magically double. I will believe it when I see it like all other claims made by the OP.

Yeah, for X16r coins thats 16^2=256, for X16s coins thats 16!/14!=240. Certainly doable.

Partial reconfiguration - you don’t need every combination, just every building block.

As it relates to Ravencoin mining with FPGAs, OP will need to store over 300 million bitstreams to account for every possible combination. Better get back to the drawing board because this design will never work.

Who is 'they'?  This is interesting.  Any information that you can link to?

I think they already mining Ravencoin with fpga.

Are you still mining on AWS and saying that it is still profitable?  Did you make your own bitstream and stay within the voltage limits mentioned on this thread?

what a monster

This looks very promising.  heres for the future 

Thanks, great information, seem to confirm that this is for software developers only at this point.  I can find my way around AWS and linux but cannot code, so waiting until someone releases something to the community or is selling something.

No.

So 27pages of blahblablah. Is there any real proof of the declared hashrates on several algos?

Do you have any experience with these FPGAs? It would be nice to know how they compare to the VCU1525 for mining.

https://www.altera.com/products/boards_and_kits/dev-kits/altera/kit-s10-fpga.html

No, but it could decode / encode 16 streams of 4K video with <500ms latency simultaneously without stuttering or losing a single frame.

But can it run Crysis in 4K @ 60fps? 

I think you're confusing markets. Traditionally, fpgas (other than tiny ones used in various devices) were only used for ASIC design. That's not the case any more. They're becoming legitimately their own product that can be used by a large number of markets. I for one am interested to see what they can do to speed up SQL servers. There are a number of big data / relational database FPGA solutions coming to market. The next couple of years are going to be very exciting for the fpga and server markets -- all that fpga power is going to be coming to our door. Just wait until the marketing guys figure out what they can do with FPGAs and how quickly they'll be able to do it... Alright, maybe this isn't a good idea at all  

Intel's only current coprocessor offering is the Xeon Phi. Intel is planning a hybrid fpga cpu. This would be like having a cpu with a gpu except it would be a cpu with a fpga. Their goal is to put a co processor on all the server chips. This would allow offloading of various applications that would have performance / efficiency increases at a multiple of a gpu -- and -- allow offloading of tasks you can't even do on a gpu. For higher end customers that need more fpga space, they'll have co-processor pci-e cards.

Even with the margin erosion, by increasing the market size, they will have far more profit than they could by price fixing. Intel sells 100s of M of chips every year. They have the ability to completely destroy Xilinx who has been stagnant, take almost complete FPGA market share, while, simultaneously taking market share from AMD and Nvidia. Brilliant! Intel could lose market cap that would be equivalent to 100% of Xilinx market cap and it would just be a "bad day". They're not in the same playing field.

Have you seen what they (Intel) decreased the price of proto chips to? I was pretty shocked. That's going to take a big bite out of Xilinx's 'entry fee' / 'cover charge'. In the next couple of years we'll be buying Stratix 10 PCI-E boards at walmart for $600 a pop.

I doubt it.  Altera/Intel have no desire to erode their margins, and there are a lot more high dollar uses for FPGAs than mining....  Most of the companies I've worked for will spend $50k-$100k per FPGA, for hundreds of engineers, to pull in an ASIC design schedule. 

ASIC's are only really going to survive on coins that aren't building resistance into their algo's.

Then we have a the FPGA's and GPU's battling it out on all the rest 

ASICs will continue to dominate mining and GPU miners are very resilient. But for FPGAs, even without their magical powers, will still remain a viable alternative for some people.