Search Posts

Ive yet to have someone explain to me why they think FPGAs are so bad but GPUs are so good.

Thats a red herring. Who is making that argument ?

The developers who have forked explicitly over their algorithms being listed in this thread.

The economies of scale argument also applies to GPU based mining. For example, with modest capital I could easily build a host system supporting up to 128 GPUs (since no one uses the PCIe for anything other than making them boot and comm that is lower than serial speed), supporting individually resetting and reinitializing them and all the benefits of smaller hosts, but at 5% of system cost instead of as much as 30%, and also provide power savings. The difference is Bitmain doesnt represent enough demand to independently control the FPGA market, and the companies in that market are not going to sacrifice decades of high margin business for a short term cash play.

There is no down side for TSMC or commodity memory manufacturers to take big money for cheap part orders from Bitmain or any large player.. There is a downside for Xilinx/Intel flooding the market with cheap FPGAs. Similarly you dont see NVIDIA and AMD letting Bitmain build custom mining GPUs with their chips at cheap prices. Companies with massive R&D into their chip products want to very carefully maintain control over markets to keep the balance between volume and margin exactly where it is most profitable.

I'm pretty sure intel's plan is to flood the market. Why else would they be developing hybrid cpu/fpgas (like APUs) or why would they bother to create the CCIX interconnect? They're definitely going to flood the market. The question is when. Xilinx will almost surely wait and only react hoping that Intel will also try to preserve high margins. The way I see it, Xilinx has one last chance to gain market share before Intel opens the flood gates. Get your options placed on XLNX while you still can

-- Looking forward to making money on their downfall.

Plus, as you and I both know, Quartus is capable of placement / routing that is orders of magnitude better than anything Vivado could do automatically without floorplanning.

I'm looking forward to my call with Intel next week! Hopefully I'll be able to get something rolling quickly and at a lower price point so we (crypto community as a whole) can avoid Xilinx all together.

Edit:

Oh ya, I find it hilarious that devs are changing their algos based on what we say in this thread. Which reminds me, I just developed this new code for cryptonight which makes use of a little hack that's able to bypass some steps to obtain a result quicker. I'm now mining monero at 100Kh/s per fpga.

Whoa. 100kh/s for monero is awesome. If its cnv7 then that's a game changer!