China's Bitcoin mining giant Bitmain is working on an Ethereum ASIC miner, the Antminer F3, according to reports in Chinese technology media.
The report states that the Antminer F3 will be mounted with three motherboards with six ASICs per motherboard, as well as 32 x 1GB DDR3 memory chips.
Apparently the new ETH ASIC chips will begin production later this month, with the ASICs hitting 200-200MH/s. The cost of each Antminer F3 will be between $2500 and $3000. The F3 miner is expected to hit the market in Q2 or Q3 2018, said reports.
72 GB DRAM outperforms most of the GPU-based mining rigs in the market now. Mining specialists say Bitmain's demand for DRAM memory is around 3,000 pieces of 12 inch wafers per month.
Bitcoin mining migrated from CPUs to GPU's to FPGAs and now to ASICs. However, Ethereum is putatively ASIC-resistant; its ethash proof of work (PoW) mining algorithm is "memory hard" or memory bound.
Without going into an inappropriate level of detail, this means that calculating the PoW requires choosing subsets of a fixed resource dependent on the nonce and block header. This resource (a few gigabyte size data) is a pseudorandom dataset called a DAG (Direct Acyclic Graph), which is regenerated every 30,000 blocks, or about every five days.
Vijay Pradeep, a Silicon Valley robotics engineer and angel investor, who is interested in memory hardness, said: "Note that this Bitmain ASIC has a bunch of DDR controllers, connecting it to a huge bank of DRAM. So, architecturally, this looks vastly different than the Bitcoin miners on the market today."
Predeep has posted about ways to customize Ethereum mining hardware, specifically how to design a high memory bandwidth FPGA/ASIC board.
He added: "The Bitmain approach is not a magic bullet for mining. With bitcoin mining, moving from GPU to ASIC gave an order-of-magnitude increase in efficiency (both in energy and hw cost). The Bitmain system will help ease a worldwide shortage on GPUs and increase Ethereum mining efficiency, but surely not by an order of magnitude.
"Bitmain is building custom hardware for a cryptocurrency that claims to be transitioning to proof-of-stake. There is lots of contention around the schedule and approach for transitioning to proof-of-stake, and with Bitmain making huge investments into Ethereum proof-of-work custom mining equipment, we now have another player who will actively lobby against Proof-of-Stake for Ethereum."
Bitmain recently announced an ASIC miner for decentralised storage blockchain token Siacoin. This caused some dismay for Sia which was planning its own ASIC "Obelisk" units. Sia Founder David Vorick stated that the Bitmain unit is far less energy efficient, costs more money, and is an objectively worse miner than the SC1. So people who ordered Obelisk units will still be receiving hardware of substantial value".
Vorick was one of the few people who said ASIC mining on Ethereum was likely to happen and that the R&D was probably underway.
Back in January, Vorick said in an email to IBT: "We had our chip devs look into both ethash and Equihash, and we feel that both would be fairly easy to make ASICs for. For equihash you just make a sorting network in hardware, and for ethash you make a giant lookup pipeline."
This story has a number of interesting permutations. It suggests that unsentimental profit-seeking miners consider Ethereum's shift to proof of stake to be suitably distant to merit creating an ASIC miner.
It's also interesting in terms of what it means for GPU mining in general, not to mention the likes of NVIDIA and AMD. It should be noted that GPU chip makers are skittish about talking publicly about how much they are making from cryptocurrency mining; they will typically say they don't know how many of their GPUs are sold for mining (whenever more than one GPU is sold at a time would be an indictor).
Another person who is interested in mining hardware and doubted Ethereum's ASIC-resistance is Bram Cohen, the inventor of BitTorrent and founder of Chia Network.
He said in an email: "ASIC resistance isn't generally such a hot idea. By attempting to make it cost prohibitive to make ASICs, you'll likely (perhaps inevitably) fail and wind up with it merely being very expensive, which will result in very little if any competition among ASIC manufacturers.
"A better approach is to try to be ASIC commodity, where a relatively modest budget can produce about as good of an ASIC as can possibly be made. Probably the best algorithm for that is SHA3 iterated a hundred times.
"SHA3 is particularly well suited to straightforward hardware implementation, and iterating a hundred times gets rid of most of the tricks which can be played to essentially make lousy chips which often fail but consume enough less power that the failure rate is worth it."
Bitmain says on the requests section of its website that response time will be impacted by Chinese holidays.