Bitmain’s ASIC Miner: Dead on Arrival

Dead on Arrival

Even before Bitmain confirmed the availability of their Ethereum ASIC Miners in July, community developers had already started working on modifications to their Ethash algorithms to render the product dead on arrival.

The Monero community already reacted with the advent of the Cryptonite miner from Bitmain:

These ASIC miners will be utterly useless before the first customer takes the device out its packaging, if it even ships at all. This will go down in the annals of crypto-mining history as one of the most expensive blunders of all time. How could Wu and Zhan not see this coming?

GPU Miners Live On

The “Bitmain Stillbirth” will prove, once and for all, that hardware flexibility to adapt to algorithmic changes is just as important as raw hash rates. GPU based miners with the ability to load code that is executed by thousands of SIMD processors live on. The threat of ASIC based mining is finally going to be over. The Monero developers are even planning on mutating their algorithm every six months to prevent the chance of ASIC miners gaining any foothold whatsoever.

Stock market traders may be interested in buying back into NVIDIA and AMD. After Analysts reduced their outlook for GPU producers, their stocks dropped slightly. Once the realization of ASIC miner hopelessness permeates, the stock price of various GPU manufacturers should go back up, but with a vengeance. This does not bode well for the environment since GPU’s are notorious power guzzlers. As the number of enthusiast miners grow, at geometric rates, we’re looking at a serious ecological problem.

The “Middle Way” – FPGA Miners

ASIC miners are fast and power efficient, yet inflexible. GPU miners are slower than ASIC miners and flexible, but they consume much more power: several orders more per hash verification.

If Buddha were around and into crypto, he would probably advise, “Take the Middle Way, use FPGA miners”. An FPGA is essentially an ASIC that can be reprogrammed. Performance wise, there’s a slight penalty for being programmable. Approximately 20-50% performance is sacrificed for this flexibility. In terms of power consumption, ASICs and FPGAs are pretty much the same.

FPGA Feasibility: Hash Rates

Most FPGA platforms possess sufficient computing power. Up until recently, the problem with FPGA miners resided in memory bandwidth. Ethereum’s Ethash algorithm is memory hard, meaning the upper bound of hash calculations require a certain amount and bandwidth of RAM. Then new kinds of memory started to emerge making the endeavor completely feasible.

Hybrid Memory Cube chips emerged three years ago to boast a bandwidth of 160 GB/s. The memory technology let’s an FPGA handle up to 20 MH/s on a single interface with less pins reducing the routing complexity. It’s impressive yet certain problems arise because the chips reside outside of the FPGA. Chip to chip communication requires several gigabit transceivers which increases the cost of the FPGA. Regardless the technology is prevalent and experiments show promising results.

Just about a year ago Xilinx introduced in-chip High Bandwidth Memory on their Virtex FPGA series with a default capacity of 8 GB. HBM has a bandwidth of 460 GB/s and a single FPGA can theoretically support hash rates of up to 58 MH/s. The most beautiful aspect to all this is that it requires no high speed (gigabit) transceivers or even controllers with access mechanisms built into the FPGA.

FPGA Feasibility: Cost

Everyone that knows anything about FPGAs knows they’re not cheap. This is due to a limited market, and the need to recover investment and set up costs (research, development, and manufacturing) to produce them. The materials and operating costs required to produce a single chip is relatively small. Hence making 100 chips costs pretty much the same as producing 1,000 or 10,000 chip batches.

Prices shift when markets grow and marginal product revenue plays a big part in these pricing decisions. If the N-th unit of product produced is of negligible cost, then a producer can lower prices to accommodate a wider consumer base and dramatically increase revenues. As an example, perhaps 10,000 consumers buy the product if it costs $1,200 and pays for itself in 10 months. If the price drops to $320 and the product pays for itself in 2 months, then you could have 10,000,000 or more consumers buying the product. These numbers are completely feasible and the difference is $12M vs. $3.2B in revenue. Our relationships and partnerships show the right price points will come with the emergence of FPGA mining thanks to the Subutai Blockchain Router.