From GPU/FPGA to ASICs

It’s exciting to see CKB mainnet’s hashrate reaches 2PH/s today, as the result of the mining transition from GPU/FPGA to ASICs. Until now 4 ASICs have been announced and we’re still at the beginning of the hashrate bump. Thanks to this CPU -> GPU -> FGPA -> ASIC process and a mature ASIC manufactoring and distribution chain the mined CKBytes have been distributed to many different types of cohorts through the last year (a long period of time!). The simple design of Eaglesong lowered the ASICs manufacturing barrier as we expect, although the power of market is still far beyond our wildest imagination that 4 ASIC models emerged before Lina turns 4 months old. The network is decentralized and growing. If you want to know more about this transition here’s a good summary.

As the hashrate spiking it’s fun to examine how the network’s security changed. Because we have real GPU mining data we can calculate a rough equivalent GPU numbers of the hashrate today and compare it with Ethereum. Here’s a GPU mining performance table:

GPU CKB Hashrate ETH Hashrate Note
2060x6 4380M [1] / 4492M [2] (avg. 4436) 165M [2]
P104x6 3780M [1][2] 206.25M [2] / 240M [3] (avg. 223M) 240M = 320M * 6/8 (P104x8 eth hashrate is 320M in [3])
P106x6 2460M [1] / 2510M [2] (avg. 2485M) 144M [2]

With the table above, it’s easy to calculate how many GPUs are needed for today’s hashrate:

CKB hashrate ~2 PH/s = 2,000,000,000 MH/s
ETH hashrate ~171 TH/s = 171,000,000 MH/s

GPU # for CKB # for ETH CKB/ETH ratio
2060x6 450857 1036364 43.5%
P104x6 529101 766816 69.0%
P106x6 804829 1187500 67.8%

The numbers show that if all ETH miners are mining with multi-purpose device such as GPUs (worst case), Ethereum would have roughly 2x devices of CKB. In this case it would take about one half of ETH mining rigs to deploy a 51% attack against CKB, which is quite unrealistic as the same force could 51% attack Ethereum directly and it would be more lucrative. Based on this I think it’s safe to conclude that Hashrate renting attack is no longer profitable.

The next question is, how much does it cost for a bribing attack against CKB? This is a much harder question to answer due to the anonymous nature of PoW. The cost to bribe manufacturers, pools and miners are all different as they have different fixed cost and variable cost. Mining pools appears relatively easier to bribe because unlike manufactures and miners they do not need to purchase ASICs, but it’s also well known that pools sometimes participate in mining themselves which will change the dynamics. We’re at the dawn of CKB’s ASIC era, we should be able to answer these questions better when the Lina is grown up.

Refs.

[1] The GPU stats was get from here but it’s removed recently, please check here instead for reference.
[2] https://www.chainnews.com/zh-hant/articles/955636747786.htm
[3] https://www.sparkpool.com/token/ETH#gpu

9 Likes

【ASICs的问世,导致主网算力增强,看来Lina正在长大,免疫力也越来越好了,以下是本文的中文翻译】

很高兴能够看到 CKB 主网的算力在今天已经达到了 2PH/s ,这是挖矿的设备从 GPU/FPGA 到 ASICs 的结果。目前已经有四种 ASICs 的厂商已经公布了他们的ASICs矿机,而我们还在算力飞跃提升的最早期。由于PU -> GPU -> FGPA -> ASIC 这个进程和一个成熟的 ASICs 的进程和分配,挖出来的 CKBytes 已经在过去的一年(一个很长的时期)分配给了不同的阵营。如我们预期,Eaglesong 简单的设计降低了 ASICs 制造的门槛,虽然市场的力量还是远远超出我们原先最大胆的想象,竟然在 Lina 问世四个月以前就已经出现了四种 ASICs 的模型。这个网络是分散并且正在成长,如果你想了解更多关于这个转变的信息,这里有一个很好的总结
image

随着算力的增加,检测网络的安全性如何变化是一件很有趣的事情。因为我们有真正的GPU挖矿数据,我们可以计算出今天的算力大致等于多少个 GPU,并将其与以太坊进行比较。这是 GPU 挖矿的性能比较表

GPU CKB Hashrate ETH Hashrate Note
2060x6 4380M [1] / 4492M [2] (avg. 4436) 165M [2]
P104x6 3780M [1][2] 206.25M [2] / 240M [3] (avg. 223M) 240M = 320M * 6/8 (P104x8 eth hashrate is 320M in [3])
P106x6 2460M [1] / 2510M [2] (avg. 2485M) 144M [2]

如上表所示,我们可以轻易的计算出今天的算力我们会需要多少 GPUs

CKB 的算力 ~2 PH/s = 2,000,000,000 MH/s
ETH 的算力 ~171 TH/s = 171,000,000 MH/s

GPU # for CKB # for ETH CKB/ETH ratio
2060x6 450857 1036364 43.5%
P104x6 529101 766816 69.0%
P106x6 804829 1187500 67.8%

数字显示,如果以太坊的矿工如果用像 GPUs (最坏的情况)这种多目的型的设备,以太坊会需要 CKB 两倍以上的设备。在这种情况下,大约需要一半的 ETH 挖矿设备才能部署 针对 CKB 的 51% 攻击,这是非常不现实的,因为同样的算力也可以直接对以太坊发动 51% 攻击,这将是更有利可图的。基于这样的原因我可以很安心的得出这个结论: 算力租借攻击 不再是有利可图。

下个问题是,对 CKB 发动贿赂攻击需要多少成本?这是一个更难回答的问题,因为 POW 的本质就具有匿名性。贿赂矿机制造商,矿池和矿工的情况是各不相同,因为他们既有不同的固定成本,也有不同的可变成本。矿池似乎相对容易贿赂,因为不像矿机制造商和矿工,他们不需要购买ASICs,但众所周知,矿池有时自己也会参与挖矿,这将改变动态。我们正处于 CKB 的 ASIC 时代的黎明,当 Lina 长大时我们将更能回答这些问题。

参考资料

[1] GPU状态的资料在这里 获取 ,但最近好像删掉了,请参看这里的数据当作参考资料.
[2] https://www.chainnews.com/zh-hant/articles/955636747786.htm
[3] https://www.sparkpool.com/token/ETH#gpu

4 Likes