00:04:19 If RandomX fails then we can look at what went wrong, which assumptions turned out incorrect, etc. Based on that information we can make a rational decision to improve RandomX or switch to e.g. SHA3. 00:04:35 But RandomX didn't fail so this is a pointless discussion now. 00:17:46 has failure been defined? to me thats another wiggly point 00:18:47 e.g., if an "asic" pops up that is 1.2x as efficient, did randomx fail? is it considered an asic if it can also perform general compute? 00:20:06 Selsta: no disrespect but having a discussion around having a plan in place for the potential failure of Rx is ideal because the alternative is we are forced to make these important decisions in the middle of a crisis 00:20:09 if something pops up thats n% more efficient, is it considered failure considering that the rest of the CPU market will most likely release a newer generation that will compete with the new n% asic? 00:20:25 Good point gingeropolous 00:20:26 in due time? 00:21:49 i don't know if there are good answers, i don't have an answer. but this is what happens on the frontiers 00:39:46 garth: I'm just not a fan of predefined plans that don't take into account what exactly failed. 07:11:43 I agree with selsta, no point in making decisions ahead of time when we don't even have any actual proof. There are many things that could explain this nonce pattern 07:16:47 "has failure been defined? to..." <- That's an excellent point. Efficiency is definitely the key metric here, though it would require measuring the actual power consumption in addition to the hashrare (i.e. finding someone who owns such a - hypothetical - device and is willing to provide accurate measurements) 07:19:22 And from there we would look at the number of devices involved, and how much of the total nethash they are replacing by 'kicking out' less efficient (and thus less profitable) devices 12:42:37 there are at least 3 github issues filled with pages and pages of this discussion already 12:42:40 no need to retread 12:43:00 failure is an ASIC at least 2x more efficient than CPUs 12:44:02 and of course, we don't know that efficiency without measuring. we might guess if fulltime miners are seeing profitability decrease non-proportional to nethash increase 12:45:43 if profitability decreases solely because net hashrate has increased, that only means there are more miners. 12:46:09 could be botnets, could be rogue datacenter admins 12:47:14 there are some interesting technologies worth keeping abreast of. like wafer-scale compute, and processor-in-memory 12:47:51 but WSC is ultra expensive, since it literally dedicates an entire silicon wafer to 1 compute device 12:48:35 PIM is promising as a way to reduce RAM latency, but so far only uses tiny microcontroller-class processors. not meaningful until full 64bit CPUs are used 12:49:42 there's no other novel architectures on the horizon that offer any meaningful advances 12:51:15 anybody who thinks they can gain even a 20% advantage over CPUs by shaving off unused features is dreaming 13:00:41 at this point more than 50% of a CPU die is cache https://images.idgesg.net/images/article/2020/10/screenshot-2020-10-08-07.14.47-100861362-orig.jpg 13:01:45 you'd need to be able to turn off around 50% of a core's components to approach a 20% power savings 13:02:42 branch predictors, decoders, cache coherency control - they don't add up to that much 13:03:24 if you jettison cache coherency control then you need to dedicate cache per core instead of sharing like here 13:03:48 which means even more of chip power is dominated by cache power, and core power is even less significant 13:36:36 failure is an ASIC at least 2x more efficient than CPUs >>> ok, so we have failure defined. 13:36:59 albeit the impracticality of obtaining enough data to test the definition 13:38:46 do we have ASIC defined in this scenario? is a custom chip that is packaged up in a device as a monero miner an ASIC if the same chip is also on the market for general compute? 13:39:29 while failure may have been hashed out in the pages of github issues, i dunno if asic definition has had the same attention 14:48:58 Actually we could define a threshold of "definitely asics", based on a single assumption about the average electricity cost for the whole network 14:51:28 The maximum hashrate that could be supported by the mining network is: Security Budget [$/s] * Highest known miner efficiency [H/kWh] / Electricity cost [$/kWh] 14:52:34 (Where Security Budget = Price [$/XMR] * Block reward [XMR/block] / Average block time [s/block]) 14:54:32 For example, right now - assuming an maximum efficiency of 130 H/s/W (top tier Ryzen) and an electricity cost of 0.10 $/kWh - the maximum nethash that could be supported is ~5.71 GH/s (193.9% of the current nethash) 14:55:40 That means that if everyone had Ryzens mining at 130 H/s/W and kept mining at breakeven (0 net profit) but without mining at a loss, the network would reach ~5.71 GH/s 14:55:52 * everyone had *only* Ryzens mining 14:57:22 So if the network were to grow beyond that, that would mean: either the average electricity cost of the whole network is lower than 0.10 $/kWh; or somebody is mining at a greater efficiency than that 14:58:35 And from there, we could work backwards and estimate their real efficiency relative to the % of the nethash they could own 15:03:49 Right now, an efficiency of 105-110 H/s/W earns ~0.16 $/kWh, so they are profitable (though not by much) as long as they pay less than that in electricity 15:15:15 Hmmm, I wonder if that threshold was ever crossed back in the Cryptonight days... The profit margins were pretty high though, so perhaps we never got quite that far because there weren't that many asics yet 15:16:56 The one thing that breaks this definition is botnets though. They effectively pay 0 $/kWh, so their profitability and the maximum network size are technically "infinity" 15:18:02 However, if we know the total % of nethash owned by botnets, we can account for it 15:37:03 "failure is an ASIC at least 2x more efficient than CPUs" this is too harsh. Even some CPUs can be more than 2x efficient compared other CPUs (Ryzen 3900X vs some Intel) 15:39:18 which means newer CPU generations can close a 2x gap. I think Zen4 will surprise everyone. 15:40:32 True. Also, 0.05 $/kWh is not that far off 0.10 $/kWh - and not far off from what some mining farms/datacenters with cheap energy can get 16:14:18 Maybe “failure is an ASIC at least 2X my re efficient than the most efficient CPU”? 16:14:26 *more 16:14:49 Which is a fun concept because it means failure is a sliding scale that adjusts with CPU innovation 16:34:29 garth by that definition, Ryzen is an ASIC for Intel CPUs :D 16:39:15 wen zen4 16:39:53 September 16:48:43 i wonder if that'll be as borked as the ddr5 17:33:09 I’m actually quite excited about zen4 as a RandomX cruncher. Certainly more than any CPU in a while 18:23:13 I think we may also add a constraint: not for sale to the general public 18:23:31 if everyone can buy one, decentralization is still preserved 18:25:25 Hmmm being able to buy one is not enough though. If there's a single manufacturer in control of all the production, then it's still centralized 18:25:45 I suppose 18:26:19 but so far, all previous ASICs have failed to meet this criteria anyway 18:27:07 is Ethash still ongoing with ASICs and GPUs mining in parallel? 18:27:31 they haven't changed pow and they haven't deployed pos yet 18:49:55 From what I've seen, yes. ASICs are still waaay more profitable, but from what I've gathered so far, the number of devices manufactured is quite limited 18:50:23 Profitability @0.10$/kWh: 18:50:39 - Vega56: 450% 18:51:07 - Radeon VII: ~700% 18:51:29 - Bitmain Antminer E9: 2000% 18:51:39 (right now) 18:53:53 2000% means that you get $2 from 1 kWh before electricity? 18:54:01 $0.1*20? 19:02:51 Bitcoin Antminer S19 Pro gets 290% profitability 19:03:24 Ryzen 3900X, tuned to 100 h/J gets 145% profitability ($0.145 worth of XMR from each kWh) 19:08:04 Hmmmm, whattomine advertises a "Bitmain Antminer E9 3.00 Gh/s, 2600W" asic, but it's not on Bitmain's site anymore 19:11:20 Their official youtube channel still has the announcement from last year though: https://www.youtube.com/watch?v=nNFa-HKTV4Y 19:11:26 So I'm guessing it's discontinued 19:12:58 Antminer E9 is for Ethereum 19:14:36 This is the problem with ASICs - they're basically impossible to buy. They're either discontinued/out of stock or you have to pass through full KYC and wait for months to get one. 19:14:40 Yeah, that's the one with 2000% profitability I was talking about 19:15:21 After all these years, ASICs haven't become a commodity 19:15:35 Yeah 19:16:19 Ok, here's a real one: Innosilicon A11, 1.5GH/s 2200W https://www.innosilicon.com/html/a11-miner/index.html 19:16:44 19k USD on discount, lmai 19:16:46 lmao 19:17:11 1178% profitability right now, but almost a full year to recover the cost 19:17:50 They're definitely pricing them to squeeze the miners in case the PoS switch actually happens (as if) 19:18:17 they're always pricing them depending on profitability 19:18:25 so buyers will get 1+ year ROI 22:26:12 After all these years, ASICs haven't become a commodity > wooo that’s a depressing point